Jamris 2012 vol 6 no 3

pISSN 1897-8649 (PRINT) / eISSN 2080-2145 (ONLINE)

VOLUME 6 N° 3 2012 www.jamris.org

Editor-in-Chief

Co-Editors:

Janusz Kacprzyk

Kaoru Hirota

Witold Pedrycz

Roman Szewczyk

(Systems Research Institute, Polish Academy of Sciences , Poland)

(Interdisciplinary Graduate School of Science and Engineering,

Tokyo Institute of Technology, Japan)

(ECERF, University of Alberta, Canada)

(PIAP, Warsaw University of Technology )

; PIAP

(Tijuana Institute of Technology, Mexico)

(Research & Advanced Engineering, Ford Motor Company, USA)

, Poland

(Polish Academy of Sciences; PIAP, Poland)

(BUMAR, Poland)

Oscar Castillo

Dimitar Filev

Janusz KacprzykMariusz AndrzejczakPlamen AngelovZenn BienAdam BorkowskiWolfgang BorutzkyChin Chen ChangJorge Manuel Miranda DiasBogdan GabryśJan JabłkowskiStanisław KaczanowskiTadeusz KaczorekMarian P. KaźmierkowskiJózef KorbiczKrzysztof KozłowskiEckart KramerPiotr KulczyckiAndrew KusiakMark LastAnthony MaciejewskiKrzysztof MalinowskiAndrzej Masłowski

Editorial Board:Chairman:

(Lancaster University, UK)

(Korea Advanced Institute of Science and Technology, Korea)

(Polish Academy of Sciences, Poland)

(Fachhochschule Bonn-Rhein-Sieg, Germany)

(Feng Chia University, Taiwan)

(University of Coimbra, Portugal)

(Bournemouth University, UK)

(PIAP, Poland)

(PIAP, Poland)

(Warsaw University of Technology, Poland)


(University of Zielona Góra, Poland)

(Poznań University of Technology, Poland)

(Fachhochschule Eberswalde, Germany)

(Cracow University of Technology, Poland)

(University of Iowa, USA)

(Ben–Gurion University of the Negev, Israel)

(Colorado State University, USA)



Executive Editor:

Associate Editors:

Statistical Editor:

Editorial Office:

Al. Jerozolimskie 202, 02-486 Warsaw, POLANDTel. +48-22-8740109,

Copyright and reprint permissionsExecutive Editor

Anna Ładan

Jacek SalachMaciej Trojnacki

Małgorzata Kaliczyńska

Industrial Research Institute for Automationand Measurements PIAP

Patricia MelinTadeusz MissalaFazel NaghdyZbigniew NahorskiAntoni NiederlińskiWitold PedryczDuc Truong PhamLech PolkowskiAlain PruskiLeszek RutkowskiKlaus SchillingRyszard Tadeusiewicz

Stanisław TarasiewiczPiotr TatjewskiWładysław TorbiczLeszek TrybusRené WamkeueJanusz ZalewskiMarek ZarembaTeresa Zielińska

[email protected]

[email protected]


(Warsaw University of Technology and PIAP, Poland)

(PIAP, Poland)

(Tijuana Institute of Technology, Mexico)

(PIAP, Poland)

(University of Wollongong, Australia)

(Polish Academy of Science, Poland)

(Silesian University of Technology, Poland)

(University of Alberta, Canada)

(Cardiff University, UK)

(Polish-Japanese Institute of Information Technology, Poland)

(University of Metz, France)

(Częstochowa University of Technology, Poland)

(Julius-Maximilians-University Würzburg, Germany)

(AGH University of Science and Technology

in Cracow, Poland)

(University of Laval, Canada)


(Polish Academy of Sciences, Poland)

(Rzeszów University of Technology, Poland)

(University of Québec, Canada)

(Florida Gulf Coast University, USA)

(University of Québec, Canada)


JOURNAL of AUTOMATION, MOBILE ROBOTICS& INTELLIGENT SYSTEMS

All rights reserved © 1

Publisher:Industrial Research Institute for Automation and Measurements PIAP

If in doubt about the proper edition of contributions, please contact the Executive Editor. , excluding advertisements and descriptions of products.The Editor does not take the responsibility for contents of advertisements, inserts etc. The Editor reserves the right to make relevant revisions, abbreviations

and adjustments to the articles.

Articles are reviewed

Articles2

Journal of automation, mobile robotics & intelligent systems

Volume 6, n° 3, 2012

3 37

17

47

23

51

27

56

6130

33

13 40

contents

Auditory Occupancy Grids with a Mobile RobotBrian P. DeJong

Event Detection in ECG, Carotid Pulse, Phonocardiogram, and Detection of Consecutive Systolic Time IntervalsAnna Strasz, Wiktor Niewiadomski, Małgorzata Skupi ski, Anna G siorowska, Dorota Laskowska, Rafał Leonarcik, Gerard Cybulski

Remote Monitoring System for Artificial HeartBartłomiej Fajdek, Marcin Stachura, Michał Syfert, Paweł Wnuk, Michał Barty

Accuracy of the Element Geometry Mapping Using Non–Invasive Computer Tomography MethodMirosław Grzelka, Lidia Marciniak, Bartosz Gapi ski, Grzegorz Budzik, Andrzej Trafarski,Joanna Augustyn-Pieni ek, Mariusz Gaca

The Models for the Comparison of Roundness ProfilesŠárka Tichá, Stanisław Adamczak

Advanced Rehabilitation Device Based on Artificial Muscle Actuators with Neural Network ImplementationKamil Židek, Ondrej Líška, Vladislav Maxim

Comparative Studies of Various Methods of Mounting the Implant Mandrel Within the BoneMarcin Zaczyk, Danuta Jasi ska-Choroma ska

Experimental Apparatus for SMA Actuator Testing Miroslav Dovica, Tatiana Kelemenová, Michal Kelemen

Simulation of Vehicle Working Conditions with Hydrostatic Pump and Motor Control Agorithm Peter Zavadinka, Peter Kriššak

The Branch & Bound Algorithm Improvement in Divisible Load Scheduling with Result Collection on Heterogeneous Systems by New Heuristic FunctionFarzad Norouzi Fard, Sasan Mohammadi, Peyman Parvizi

Surface Characterization by Accurate Measurement and Image Processing Systems on Machined Surfaces of Precision Cutting ToolsNuman M. Durakbasa, Ismail Bogrekci, Pınar Demircioglu, Gökcen Bas, Aslı Gunay

Metrology for Pressure, Temperature, Humidity and Airspeed in the AtmosphereAnna Szmyrka-Grzebyk, Andrea Merlone, Krzysztof Flakiewicz, El bieta Grudniewicz, Krzysztof Migała

Applying Reverse Engineering to Manufacture the Molds for the Interior Decorations IndustryFlorin Popi ter, Daniela Popescu, Dan Hurgoiu, Radu R ca an

Journal of Automation, Mobile Robotics & Intelligent Systems VOLUME 6, N° 3 2012

Articles 3

Auditory Occupancy Grids with a Mobile Robot

Brian P. DeJong

Submitted 7th July 2011; accepted 30th March 2012

Abstract: This paper presents the use of auditory occupancy

grids (AOGs) for mapping of a mobile robot’s acoustic environment. An AOG is a probabilistic map of sound source locations built from multiple measurements using techniques from both probabilistic robotics and sound localization. The mapping is simulated, tested for ro-bustness, and then successfully implemented on a three-microphone mobile robot with four sound sources. Using the robot’s inherent advantage of mobility, the AOG cor-rectly locates the sound sources from only nine measure-ments. The resulting map is then used to intelligently po-sition the robot within the environment and to maintain auditory contact with a moving target.

Keywords: auditory occupancy grids, sound localiza-tion, occupancy grids, mobile robot

1. IntroductionRobots are becoming ever more common in our homes,

offices, factories, military, and emergency-response units. In every application, the more information a robot has about its environment, the more versatile it is. Robots are equipped with cameras, ultrasonic sensors, laser range finders, accelerometers, and microphones. The informa-tion obtained is then used for mapping, locating, tracking, and informed decision-making.

While robotic vision has seen great advances, robotic audition is still in its infancy. Yet hearing – specifically, sound location – is not unimportant for robots. Assistive robots need to be able to respond to verbal commands such as “Come here” by finding where “here” is. Mechanic robots need to listen to factory machinery or car engines to detect unwanted or troublesome sounds. Sentry and security robots need to recognize suspicious noises, locate their origin, and investigate. Search and rescue robots need to locate the sound of survivors in smoky buildings and piles of debris.

Compounding the issue is the myriad of noises surrounding the robot. Human bystanders, cars, televi-sions, radios, plumbing, air vents, and machinery create unwanted noises. The robot itself generates noises from its internal fans, motors, and drive systems. Furthermore, these noises echo off of nearby surfaces, creating phantom sources and locations.

This paper explores how to use one of the robot’s unique advantages: its mobility. As a robot moves, it can keep track of the auditory information to generate a sound-based map of the environment. The effect of

location-specific echoes and self-generated noise is diminished when information from multiple locations is compiled, while the stationary sound sources materialize out of the noise. What results is a probabilistic contour map of the auditory scene, such as that in Figure 1. Given these auditory occupancy grids (AOGs) [1], a robot is better equipped to listen, such as by moving to a quiet location, maintaining a line-of-hearing to a target, or comparing two time-different AOGs of the same area.

The following sections cover the background, simu-lation, robustness, and implementation of these AOGs. First, we begin by briefly reviewing techniques used in this paper, and discuss related research in robotics. Then, we present the creation of these grids from multiple measurements, both in simulation and implementation with a mobile robot. Given these maps, we explore several of their uses.

2. BackgroundAOGs combine techniques from two well-established

research fields. The mapping component comes from probabilistic robotics while the auditory processing is based on sound localization. We briefly review funda-mentals of each, and then discuss the limited research that intersects them.

The term auditory occupancy grid is necessary to distinguish these maps from traditional noise maps. For

Figure 1. Auditory occupancy grid for the robot’s work-space, generated from nine measurements. Lighter color corresponds to higher probability. The four sound sourc-es are designated by asterisks, and are correctly located


Articles4

many years, acoustical and safety engineers have used noise maps for modeling potentially hazardous environ-ments or determining noise impact levels for proposed infrastructure [2]. Unlike AOGs, noise maps are straight-forward contour maps of sound pressure levels, and do not directly reflect source location.

2.1. Probabilistic roboticsProbabilistic robotics is the application of probability

theory to robotic sensing and mobility [3]. For example, suppose a robot is using a laser range finder to locate an object in its workspace. Traditional, deterministic robotics assumes the object to be wherever the sensor measures the strongest return. Probabilistic robotics, on the other hand, represents the sensory information as a probability distribution that inherently includes the uncertainty of the measurement. Standard probability mathematics can then be used to combine multiple measurements from multiple sensors to define a more confident belief about the environment.

The first step in this process is the converting of raw sensor data into a probability distribution. This algorithm is called the inverse sensor model [4] because it is uses the data measured by the sensor to describe a model for that sensor. A simple example of a sensor model is one that scales the sensor data to a range of probabilities between arbitrary bounds:

[ ] [ ]min max min max: :data data p p→ . (1)

The bounds represent the highest confidence assigned to any single sensor measurement, since the strongest sensor return is mapped to pmax and the weakest to pmin. More complicated models exist, such as filtering algorithms. For example, Grabowski et al. [5] devel-oped a dynamic inverse sensor model to improve sonar mapping amid specular reflection. Similarly, Thrun [6] used maximum-likelihood filtering of sensor data to improve the mapping and reduce possible conflicts between multiple measurements. Note that the prob-ability distribution is not a probability density function – the probabilities do not necessarily sum to unity.

Once the sensor data is in terms of probabilities, it can be combined with other measurements to improve the robot’s belief. Combining multiple probabilities is gener-ally done by a Bayes filter [7]. Given a prior belief and a conditional sensor measurement, the updated posterior belief is

( ) ( )1 1prior cond

posteriorprior cond prior cond

p pp

p p p p⋅

=⋅ + − ⋅ −

.

In numerical algorithms, this equation becomes unstable with values near zero and one, so the log odds (LO) method [8] is often used:

ln

1i

ii

posterior prior sensor initial

pLOp

LO LO LO LO

=−

= + −

where i is any variable and LOinitial is the log odds of the default probability (usually 0.5, or 50%). Then, the poste-rior probability can be recovered by

111 posteriorposterior LOp

e= −

+ .

This is also called the binary Bayes filter.Probabilistic beliefs are then used for various purposes,

such as mapping the robot’s environment, location of the robot within that environment, tracking of objects, or planning of paths or control.

In this paper, however, we focus on the mapping and assume accurate knowledge of the robot’s pose. Mapping of the auditory scene is done by a grid of cells where each cell has a probability of being occupied by an object (first introduced by Elfes [9]). This grid is called an occupancy grid [3], certainty grid [10], or evidence grid [11], and is usually used for mapping of the physical environment.

Occupancy grids are most useful for stationary, i.e., static, environments, although there has been some recent research in applying them to dynamic situations. For example, Wolf et al. [12] used separate static and dynamic occupancy grids to map a robot’s environment, while using a landmark-occupancy grid to locate the robot within it.

2.2. Sound localizationWhile occupancy grids come from probabilistic

robotics, the auditory processing techniques used in AOGs come from the well-established field of sound localization.

Sound can be localized based on two or more record-ings at known locations. Suppose a source is emitting a sound, s(t), which is recorded by two microphones of known locations. The microphones record different sounds, ma(t) and mb(t), each containing an attenuated

1 The frequency domain equation follows directly from the time do-main equation given properties of Fourier transforms and cross cor-relations.

Figure 2. Sample Time Difference on Arrival (TDOA) for two microphones spaced 1 meter apart, assuming the sound travels 343 m/s


Articles 5

(due to distance, medium, directionality, etc.) and time-delayed version of the source with unknown noise. Given the time difference between the two recordings, called the Time Difference on Arrival (TDOA), the source must be located somewhere on a surface in space that corre-sponds to that TDOA. In two dimensions, this surface is a parabola – see Figure 2; in n-dimensional space, one microphone pair restricts the source to a one-less-dimensional ((n-1)-dimension) space.

The TDOA between two microphones is found by performing a cross correlation of the signals, either in the time domain,

( ) ( ) ( )t a bcorr m t m t dτ

τ τ τ= −∫ ,

or in the frequency domain,

( ) ( ) ( ) jt a bcorr M M e dωτ

ωτ ω ω ω−= ⋅∫ ,

where τ is a time shift.1 Here, ( )aM ω and ( )bM ω are the Fourier transforms of microphone recordings ma(t) and mb(t), ( )bM ω is the complex conjugate of ( )bM ω , and

1j = − . The best estimate for the TDOA is thus the τ that maximizes the cross correlation.

The frequency domain technique is standard because of its accuracy. The accuracy in the time domain is restricted by the sampling frequency: mi(t-τ) must be known, so τ must be in timestep increments. Figure 3 shows the effect of this increment for a sample micro-phone placement and a sampling time of 5000 Hz. On the other hand, any τ can be tested in the frequency domain, so the frequency domain analysis allows for more accu-rate calculation of the TDOA. However, the frequency analysis requires a Fourier transform of the signals, which can be computationally intensive in near-real-time appli-cations.

In most applications, three or more microphones are used. Given n microphones, there exist (n-1) indepen-dent pairings and respective TDOAs. Thus, with three microphones in two dimensions, the source can be trian-gulated to the intersection of the two parabolas. In three

dimensions, four microphones are needed. Additional microphones (e.g., eight [13,14], thirty-two [15], or even 128 [16]) are often used to provide more comparisons and improve robustness.

Another way of approaching the localization is to assign a likelihood value to each position in space based on the time delay corresponding to that position. The Steered Response Power (SRP) [17] for microphone pair (a,b) at each robot location x is defined as

( ) ( ) ( ) xjab a bM M e dωτ

ωω ω ω−= ⋅∫F x ,

where τx is the expected TDOA for that location. These correlations add to give the SRP given all microphone pairs:

1 1( ) ( ) ( ) x

n nj

a ba b

M M e dωτ

ωω ω ω−

= =

= ⋅∑∑∫F x . (3)

This likelihood function thus maps the probabilities around the microphone array.

The SRP can be tuned to certain frequencies by including a prefiltering weighting in the cross correla-tion, W(ω):

1 1

( ) ( ) ( ) ( ) x

n nj

a ba b

W M M e dωτ

ωω ω ω ω−

= =

= ⋅∑∑∫F x . (4)

This is called the Generalized Cross Correlation (GCC) [18]. The most common weighting is the Phase Transform (PHAT) [14,19],

1( )( ) ( )PHAT

a b

WM M

ωω ω

= ,

because it reduces the effect of echoes – the echoed frequencies appear stronger in the microphone recordings but are normalized by the weighting term. The PHAT weighting also reduces the effect of unequal preamplifiers on the microphone signals. With a PHAT weighting, the GCC reduces to an integral of phase-based terms.

In the literature, there has been much work done on sound localization, although usually with set of stationary microphones distributed around a room. For example, Stillman and Essa [20] used a four-microphone array for localization in a smart room. Mungamuru and Aarabi [21] used models of source and microphone directivity to improve upon the GCC PHAT algorithm for stationary microphone arrays distributed about a simulated room. Also, Aarabi [19] combined ten weighted likelihood maps from two-microphone arrays to locate three sound sources.

2.3. Auditory sensing in robotsThere has been some research on sound localization

using human-like robots. For example, Nakadai et al. [22] built a biologically inspired two-microphone humanoid robot that localizes sound sources using inter-aural phase and intensity differences. Unfortunately, they found that the front-back ambiguity couldn’t be solved without active audition [23], such as by rotating the microphone array [24]. Huang et al. [25] used a three-microphone array and biomimetic processing to determine direction

Figure 3. Comparison between (a) time domain and (b) frequency domain cross correlations for 5000 Hz signals. The color bands correspond to points in space that have the same corr(τ) values. The time domain is restricted by the sampling frequency while the frequency domain can be more detailed given finer increments of τ


Articles6

to two sound sources. Implementing it on a mobile robot, they then combined visual and auditory information to localize a human speaking [26].

For tracking sound sources from a mobile robot, Valin et al. [13] used an eight-microphone array, beam-forming, and particle filtering. (In a previous work [27], they implemented a simplified method for determining the angle to a source based on a far-field assumption.) Likewise, Sasaki et al. [15] used a thirty-two-microphone array on a mobile robot to map two moving sound sources.

Using auditory evidence grids as an intersection between probabilistic robotics and sound localization was first proposed by Martinson and his colleagues. They implemented auditory evidence grids on a mobile robot via a four-microphone array. They were successful in locating two sources, but needed post-processing to locate three [28]. They were also able to determine source volume and directionality by having the robot approach each source to investigate it [29]. In addition, they have successfully applied their mapping to human tracking and speech recognition, for improved human-robot interaction [30].

3. Creating AOGs on a mobile robotGiving a mobile robot the ability to localize sound is

not necessarily easy. In many applications, the robot’s environment does not have a previously distributed static microphone array, and the mobile robot must carry the array with it. All sound measurements must be taken at the robot’s location with limited distance between micro-phones, greatly reducing the accuracy of localization. Figure 4 shows a comparison of sound localization for fixed three-microphone arrays of two different radii. The figure also illustrates another common issue with closely spaced microphone arrays: the array can locate the angle to the source accurately, but not the distance.

environment to verify the AOG algorithm. Second, we map randomly-located sound sources to get a sense of its robustness. Third, we implement it on a mobile robot for one environment, as a proof of concept. While these results are not meant as an in-depth experiment, they do show that AOGs are successful in many environments. For both simulation and implementation, we use a three-microphone array for simplicity; however, the algorithm holds for any number of microphones.

3.1. SimulationLet us illustrate the procedure via the simulation we

created in MATLAB. Suppose we have a mobile robot in a two-dimensional 5x5 meter workspace with four omni-directional, dimensionless sound sources (see Figure 5). We grid the workspace into 0.1x0.1 meter cells; for each cell we want to determine the probability that it holds a sound source. Thus, each cell initially has a probability of 0.5. This grid is the AOG.

Figure 4. Sound localization for two arrays of different radii. The closer array has poorer localization, espe-cially along the radial direction

Figure 5. The three-microphone (labeled a-c) mobile ro-bot’s workspace, showing sound source locations (num-bered 1-4) and path (plus signs represent measurement locations).

For the simulation, the four sound sources were modeled as omnidirectional signals consisting of a sum of 250 sinewaves (p) with uniformly distributed frequency (ωp) and phase (ϕp), and with pressure attenuation based solely on distance between source k and microphone i (dk,i). We numbered the sources as 1-4 (e.g. Figure 5). For each microphone recording, the sources were indi-vidually time-delayed (τ) based on distance.

The robot carried a three omnidirectional microphone array, equally distributed on a 0.2-meter flat circle centered at the robot’s center (Figure 5). We labeled the microphones as a-c. Because the sound signals were sums of sinewaves rather than recordings, they could be time-shifted analytically to create omnidirectional microphone signals (ma(t), mb(t), and mc(t)). For each microphone, the four source signals were then combined along with uniformly distributed random noise (n(t)). The general magnitudes of the sources (ck) were 1.0, 0.5, 0.5, and 1.0, with a noise magnitude of 1.0 for each. Mathematically, source k was

( )250

1( ) sink k p p p

ps t c c tω ϕ

=

= +∑ ,

A mobile robot, however, has the advantage of mobility over static microphone arrays. As the robot traverses its environment, it can take multiple sound measurements that can be fused together to make a clearer, more accu-rate occupancy grid of the auditory scene. This AOG is similar to (but not directly analogous to) an elevation map of an area.

The following sections detail a validation of the AOG concept. First, we simulate the robot, microphones, and


Articles 7

where ck is the general magnitude of source k (1.0, 0.5, 0.5, or 1.0), and cp is the magnitude of sinewave p (randomly distributed). Thus, microphone recording for microphone i was

( )( ),

4 250

,1 1

( ) noise( ) attenuated_and_delayed_ ( )

( ) sin ,k

k i

isources

ci p p i k pd

k p

m t t s t

n t c tω τ ϕ= =

= +

= + − +

∑

∑ ∑

These source and microphone models are simplistic, but serve the purpose of our simulation. The simulation also ignores any robot-generated noise (such as from onboard motors or fans).

At each robot pose throughout the workspace, we

obtain a 3-meter-radius likelihood map for sound sources, F(x) (Equation 4), for the surrounding area based on 1024-sample simulated microphone recordings. This is approximately equal to a sampling rate of 5000 samples per second (the maximum sampling rate for our physical implementation discussed later). Thus, frequencies above a frequency of 2500 Hz (which includes much of the audible range) can cause aliasing. The likelihood map is centered on the robot and covers a subset of the overall workspace. We chose a radius of 3 meters based empiri-cally on the useful range of our microphone array. To calculate F(x), we used the GCC with PHAT weighting (Equation 4) using Fast Fourier Transforms. Because the position-dependent time delays, τx, are independent of the microphone recordings, we calculated them beforehand.

Figure 6 shows a sample likelihood map for a situa-tion with only one sound source. Each microphone pair defines a parabola of high likelihoods. When combined, the result is one area of high likelihood. The three weaker arcs in Figure 6d are artifacts of the individual pairings. Note that the simulation involves more complicated like-lihood maps, since the multiple sources create multiple parabolas in each microphone pairing.

Next, the likelihood values are converted to prob-abilities to form a single-measurement probabilistic map, which we call the sensor grid. Our inverse sensor model scales the likelihood values to a range between 0.1 and 0.9 probability (as in Equation 1), based on the minimum and maximum values seen at that the corresponding robot pose. This in an arbitrary model and has implications on the quality of the AOG, as we will discuss later. Math-ematically, our scaling for any likelihood, F(x), is

( ) ( )( )minmax min

0.9 0.10.1sensorp F FF F

−= + ⋅ −

−x x ,

where Fmax and Fmin are the maximum and minimum likelihoods.

The simulated robot traversed the space as shown previously in Figure 5, generating sensor maps and fusing them together using the binary Bayes filter (Equation 2).

The resulting AOG is shown in Figure 7. In the figure, color corresponds to probability, with white representing highest probability and the highest contour lines repre-senting 90% confidence (contours are at 23, 45, 68, and 90%). Visually, the algorithm successfully located the four sources, even though the microphones are tightly grouped on the mobile robot. Furthermore, while each likelihood map is relatively coarse (Figure 6d), the resulting AOG is quite accurate.

To localize the sound sources from the AOG, we used a threshold of 70% probability to isolate peaks. This threshold can significantly affect the number and loca-tions of the sources – in general, increasing the threshold decreases the number of sources found, and can increase or decrease the location accuracy depending on the shape of the peaks. The 70% used here was chosen empirically from the robustness test discussed later as a robust and consistent threshold across various workspaces. Here, using averages for the peaks weighted based on each

Figure 6. Likelihood maps for one source, for (a) mi-crophones a and c, (b) microphones a and b, (c) micro-phones b and c, and (d) combined

Figure 7. Auditory occupancy grid for simulated robot and sources (represented by asterisks)


Articles8

nearby point’s probability,

i

i

ipeak

pp

= ∑∑

x

x

xx .

we calculate each peak’s weighted center. The four sources were localized within 0.19, 4.3, 4.1, and 4.3 cm, respectively.

3.2. Robustness of the algorithmAs a test of the robustness of the AOG algorithm to

source location and distribution, we simulated 100 four-sound-source environments and located the sources in them. For each environment, the four sound sources were randomly placed among the nine possible locations shown in Figure 8. The AOG simulation mapped each acoustic environment, and then located sources based on the resulting peaks. From initial tests, we empirically chose a threshold of 70% and a minimum peak weight of 1.5 (units of probability).

The results show that AOGs are relatively robust to

source positioning in the environment. Of the 100 tested trials, the algorithm correctly found four sources 65% of the time, and incorrectly found three or five sources 19% and 16% of the time, respectively. However, in the three-source results there were no false positives (all sources found were valid), and in the five-source results there were no false negatives (all valid sources were found). Thus, the AOG located 381 out of the 400 sources (95%).

For the correctly located sources, the AOGs averaged a maximum localization error per environment of 8.7 cm, with the worst localization error of 22 cm. Figure 9 shows the percent of sources found with each error, and within each error (probability and cumulative density functions). The majority of sources were found with very small errors: over 50% were found within 3 cm, while 90% of the sources were found within 9 cm. This is surprisingly accurate for only nine sound measurements.

3.3. Implementation Next, we implemented AOGs on an in-house wireless

mobile robot (see Figure 10) as a proof of concept. Our goal was to validate the simulation and algorithm, not to conduct a full experiment.

The robotic system consists of two computers running MATLAB’s xPC Target software: a host desktop where the algorithms are programmed, and the robot’s onboard target PC-104 stack. The robot’s computer includes an Advantech PCM-3375 CPU (533 MHz) carrying a 512 MB CompactFlash card as a hard drive, a Sensoray 526 DAQ board, and miscellaneous input/output/protec-tion circuitry. Algorithms are programmed on the host computer in Simulink, compiled into C code via Real-Time Workshop, ported to the robot via xPC, and run there as a kernel. Data can be ported back to the host computer for additional analysis and plotting.

The robot uses three Audio-Technica MT830c omnidi-rectional condenser lavalier microphones and drive motor encoders. The microphones are attached to 20-cm arms extending radially, as modeled in the simulation, with onboard preamplifiers.

The four sound sources were selected to give a range Figure 8. Robot’s path (same as before) and possible source locations for the robustness test

Figure 9. Number of sources found in the robustness test with each error value (probability density), and within each error value (cumulative density)

Figure 10. Photograph of the mobile robot, with PC-104 stack, three-microphone array, and wireless Ethernet bridge


Articles 9

of sounds typical of a mobile robot’s environment. They were (recall the numbering as shown in Figure 5):1. A knocking at approximately 5 Hz, to create distinct

high-frequency beats in the microphone recordings.2. A recording of classical music, with a mix of long

tones and regular beats.3. A recording of a human speaking, with irregular

patterns and frequencies.4. An electric shaver, with relatively constant frequen-

cies.The sound sources were not all at the same volume –

source 1 (tapping) and source 4 (shaver) were both louder than the other two (as in the simulation). The sources were positioned facing up (perpendicular to the robot’s workspace) to minimize directionality of the sound, since the current algorithm assumes omnidirectionality.

The robot traveled around the 5x5 meter workspace as shown previously in Figure 5. At each location, it briefly stopped moving and recorded 0.2 seconds of sound at 5000 Hz, generated a sensor grid for that location from a 1024-sample correlation window, and updated its AOG.

Figure 11 shows a comparison between sensor grids from the simulated and real robots, when the robot is in the center of its workspace. In the plots, the color corre-sponds to probability of a source, with white representing 90% probability, to black representing 10% probability. The sensor maps correctly locate the sounds sources since the white beams point to them. The sensor maps in Figure are similar, giving credence to our simulation. Recall that the simulation used simple models for the sound sources and microphones.

The overall AOG is shown previously in Figure 1, where lighter colors correspond to higher probability. The map correctly locates the sound sources, with the highest contour lines representing 97% confidence (contours are at 24, 49, 73 and 97%). Even though the four sources had various volumes, they were all strongly located (see Figure 12): distance errors were 10.3, 7.3, 8.4, and 0.4 cm, respectively. This AOG is similar to the one from the simulation.

In addition, two small peaks can be seen in the bottom right and upper right of the grid (in both the simulation and the implementation) – these are false peaks can be eliminated with further measurements, or simply based on their size. From analyzing the map as it was generated, it appears that these peaks are artifacts of the sensor grids radial-direction inaccuracy.

The AOGs are still effective at locating the sound sources in this application when using time-domain correlations. Figure 13 shows the robot’s AOG using time-domain correlations (same contours). While not as accurate as the previous GCC version, it still locates the four sources from nine measurements. The sensor maps generated during the time-domain analysis are coarser (recall Figure 3), and unweighted like the GCC, yet the AOG algorithm is still successful. This result is promising for applications where the Fourier transforms needed by the GCC are too computationally intensive.

The AOGs show that the robot successfully located the four various sources with only nine measurements. The robot now has an understanding of the auditory land-

Figure 11. Sensor maps for the central location in the workspace, for (a) the simulation and (b) the implemen-tation. Color corresponds to probability, from white for 90% to black for 10%. The white beams correctly point to the sources (asterisks)

Figure 12. Sources’ locations as calculated from the AOG. All sources were correctly located within 20 cm


Articles10

scape and can use that knowledge for various tasks. In the following section, we present and implement several uses for the AOG.

4. Using AOGsWe now discuss several uses for the AOG, using the

real-world AOG found via the implementation.

4.1. Moving to a sourceless locationIn some situations, such as when the robot is listening

for commands or suspicious sounds, it may be beneficial for the robot to move to a position of low source-proba-bility. This is not the same as moving to the quietest loca-tion in the workspace, but it is moving to the area that is least likely to contain an interfering source. We can find this location from the AOG, using a similar method to the source locating. Using a threshold of 5% probability and weighted average of the largest space, we get the map

and location shown in Figure 14. When commanded, the robot moved to this location.

4.2. Line of hearingAnother situation that benefits from the acoustical

knowledge is when attempting to track a moving source that the robot is listening to. Suppose the robot is trying to listen to a source of interest, such as a human speaking. Clearly, the best method for hearing the human is to move next to it. But suppose the human is moving in the given workspace across the bottom border from left to right, while the robot is constrained to the top border – the robot is listening to the human from across the room.

In this application, the robot can use standard beam-forming techniques [13] to focus the listening towards the human, but it is also advantageous for the robot to posi-tion itself along its border such that there are no sources between it and the human. That is, sources between the robot and human will hurt the robot’s listening capability.

The robot can use its mobility and the AOG to main-tain a low-source-probability “line-of-listening” with the source of interest.

We implemented this with a discrete path-step algo-rithm. A human moved across the bottom border from left to right at a constant speed. The robot started on the left of the top border and moved intelligently along that border, at twice the speed of the human. At each time step, the robot had several possible locations to move to – to the left or to the right – based on its speed (and bounded by the edges of the map). For each possible location, there exists a line-of-listening, l(x) with n points, from robot to human made up of map locations and their corresponding probabilities. The robot assigned a weight, w(x), to each possible location based on the probability

Figure 13. Auditory occupancy grid for the robot’s work-space using time-domain correlations. The sound sourc-es are still located

Figure 15. Line-of-listening tracking of human speaker. The human moved along one border, while the robot moved along the opposite so as to minimize the prob-ability of a sound source between them. The second plot compares the intelligent tracking to default tracking

Figure 14. Sourceless location as calculated from the AOG


Articles 11

that a source exists somewhere on that line, by combining each point’s probability via standard statistics:

( )(1 2) (1) (2) (1) (2)

( ) ( ) (1 2 3... ) .p p p p p

w p l p n∪ = + −

= = ∪ ∪ ∪x x

It then moved to the position with the lowest weight. This weighting could be of another form, such as maximum probability seen, average probability, etc. Figure 15 shows the path taken by the human and robot. It also shows the probability of a source along the line-of-listening for both a default exact-position tracking and the intelligent weighted tracking. The weighted tracking reduced the probability of sound interference, and thus improved the ability of the robot to listen to the human.

5. Conclusions and Future WorkWe have successfully mapped four sound sources in

two dimensions using a three-microphone mobile robot. The simulation, robustness test, and implementation show that high quality, accurate AOGs are achievable from only nine measurements, for various source locations. We have also successfully demonstrated several uses for these auditory maps.

We are continuing to research AOGs. For example, the AOGs presented here were created when our mobile robot paused its motion to maintain position and eliminate self-generated noise while the sound segments were recorded. This procedure may not be desirable in all applications. Can the maps be generated on the fly without any pauses?

In addition, the robot’s workspace did not include any sound-echoing surfaces or obstacles that typically degrade a robot’s listening capability. The AOGs should be robust enough to strong reverberations – are they?

AOGs could prove useful in some three-dimensional applications, such as factories or car engines. Once again, the AOGs should be easily extendable – are they?

One limitation of all occupancy grids is that they are designed to record static information. As mentioned earlier, work by Wolf [12] and others (e.g., [31]) suggest that occupancy grids can be used for dynamic environ-ment, and there has been limited robotic sound localiza-tion of moving sources (e.g., [13,15]). How can AOGs be applied to dynamic environments?

The inverse sensor model used here was a simple scaling from likelihood values (from the cross correla-tion) to a probability range. This model has limitations when the likelihood map is near unity. For example, if the robot is far from all sources, the likelihood map will be near zero – each cell receives a low likelihood from the cross correlation. With our sensor model, however, the slightly-more-likely cells are scaled to high probabilities, even though the likelihood map implies that they probably don’t contain sources. One way to mitigate this effect is to use the maximum and minimum likelihoods seen at any robot pose as the boundaries for the scaling. However, this is post hoc information, and we found it gives too much weight to only one or two likelihood maps. How can the inverse sensor model be improved?

Finally, in this paper we have demonstrated only a few sample applications for AOGs, although many more

exist. What lessons can be learned from the application of AOGs to new situations?

AUTHORBrian P. DeJong – Central Michigan University, Mount Pleasant, MI 48859, USA; e-mail: [email protected]

References[1] DeJong B.P., “Auditory occupancy grids: Sound

localization on a mobile robot”. In: IASTED Inter-national Conference on Robotics and Applications, 2010.

[2] Martinson E., Arkin R., “Noise maps for acousti-cally sensitive navigation”. In: Society of Photo-Optical Instrumentation Engineers, 2004.

[3] Thrun S., “Probabilistic algorithms in robotics”, AI Magazine, vol. 21, no. 4, 2000.

[4] Thrun S., Burgard W., Fox D., Probabilistic Robotics, MIT Press, Cambridge, MA, 2005.

[5] Grabowski R., Khosla P., Choset H., “Autonomous exploration via regions of interest”. In: IEEE/RSJ International Conference on Intelligent Robots and Systems, 2003.

[6] Thrun S., “Learning occupancy grid maps with forward sensor models”, Autonomous Robots, vol. 15, no. 2, 2003.

[7] Leon-Garcia A., Probability and Random Processes for Electrical Engineering, 2nd ed. Addison-Wesley, 2004.

[8] West M., Harrison J., Bayesian Forecasting and Dynamic Models, 2nd ed. Springer-Verlag, New York, 1997.

[9] A. Elfes, “Sonar-based real-world mapping and navigation”, IEEE Transactions of Robotics and Automation, 1987.

[10] Moravec H.P., “Sensor fusion in certainty grids for mobile robots”, AI Magazine, vol. 9, no. 2, 1988.

[11] Schultz A.C., Adams W., “Continuous localiza-tion using evidence grids”. In: IEEE International Conference on Robotics and Automation, 1998.

[12] Wolf D.F., Sukhatme G.S., “Mobile robot simulta-neous localization and mapping in dynamic envi-ronments”, Autonomous Robots, 2005.

[13] Valin J.-M., Michaud F., Rouat J., “Robust local-ization and tracking of simultaneous moving sound sources using beamforming and particle filtering”, Robotics and Autonomous Systems, vol. 55, no. 3, 2007.

[14] Rabinkin D.V., Renomeron R.J., Dahl A., et al., “A DSP implementation of source location using microphone arrays”, The Journal of the Acoustical Society of America, vol. 99, no. 4, 1996.

[15] Sasaki Y., Kagami S., Mizoguchi H., “Multiple sound source mapping for a mobile robot by self-motion triangulation”. In: IEEE/RSJ International Conference on Intelligent Robots and Systems, 2006.

[16] Kagami S., Tamai Y., Mizoguchi H., and Kanade T., “Microphone array for 2D sound localization


Articles12

and capture”. In: IEEE International Conference on Robotics and Automation, 2004.

[17] DiBiase J.H., Silverman H.F., Brandstein M.S., “Robust localization in reverberant rooms”, Micro-phone Arrays: Signal Processing Techniques and Applications, Springer, 2001.

[18] Knapp C.H., Carter G., “The generalized correlation method for estimation of time delay”, IEEE Trans-actions on Acoustics, Speech and Signal Processing, 1976.

[19] Aarabi P., “The fusion of distributed microphone arrays for sound localization”. Journal of Applied Signal Processing, vol. 4, 2003.

[20] Stillman S. , Essa I., “Towards reliable multimodal sensing in aware environments”. In: ACM Interna-tional Conference, 2001.

[21] Mungamuru B. , Aarabi P., “Enhanced sound local-ization”, IEEE Transactions on Systems, Man, and Cybernetics, vol. 34, no. 3, 2004.

[22] Nakadai K., Matsuura D., Okuno H., Kitano H., “Applying scattering theory to robot audition system: Robust sound source localization and extraction”. In: IEEE/RSJ International Conference on Intelligent Robots and Systems, 2003.

[23] Nakadai K., Lourens T., Okuno H., Kitano H., “Active audition for humanoid”. In: National Conference on Artificial Intelligence, 2000.

[24] Nakadai K., Hidai K., Okuno H., Kitano H., “Epipolar geometry based sound localization and extraction for humanoid audition”. In: IEEE/RSJ International Conference on Intelligent Robots and Systems, 2001.

[25] Huang J. , Ohnishi N., Sugie N., “A biomimetic system for localization and separation of multiple sound sources”, IEEE Transactions on Instrumenta-tion and Measurement, vol. 44, no. 3, 1995.

[26] Huang J., Supaongprapa T., Terakura I., Wang F., Ohnishi N., Sugie N., “A model based sound local-ization system and its application to robot naviga-tion”, Robotics and Autonomous Systems, no. 27, 1999.

[27] Valin J.-M., Michaud F., Rouat J., Letourneau D., “Robust sound source localization using a micro-phone array on a mobile robot”. In: IEEE/RSJ International Conference on Intelligent Robots and Systems, 2003.

[28] Martinson E., Schultz A., “Auditory evidence grids”. In: IEEE/RSJ International Conference on Intelligent Robots and Systems, 2006.

[29] Martinson E., Schultz A., “Robotic discovery of the auditory scene”. In: IEEE International Conference on Robotics and Automation, 2007.

[30] Martinson E., “Improving human-robot interaction through adaption to the auditory scene”. In: ACM/IEEE International Conference on Human-Robot Interaction, 2007.

[31] Graves K., Adams W., Schultz A., “Continuous localization in changing environments”. In: IEEE International Symposium on Computational Intel-ligence in Robotics and Automation, 1997.


Articles 13

Event Detection in ECG, Carotid Pulse, Phonocardiogram, and Detection of Consecutive Systolic Time Intervals

Anna Strasz, Wiktor Niewiadomski, Małgorzata Skupi ska, Anna G siorowska, Dorota Laskowska, Rafał Leonarcik, Gerard Cybulski

Submitted 26th June 2011; accepted 2nd September 2011

Abstract: We developed a program which allows measurement

of consecutive time intervals between chosen events in ECG, carotid pulse, and phonocardiogram. Currently it is possible to determine following systolic time intervals (STI): PEP – pre-ejection period, Q-S2 – time between trough of Q wave and aortic valve closure, Q-D - time between trough of Q wave and dicrotic notch, S2-D - time between aortic valve closure and dicrotic notch, as well as QQ interval.Measurements were performed on 30 young, healthy sub-jects. Subjects were supine, they performed two-minute isometric handgrip (HG) twice. First HG was followed by four-minute rest, second HG by two-minute occlusion of the working arm.Preliminary analyses revealed: 1/ the QQ interval changes were reflected weakly or not at all in changes of STI, 2/ shortening of QQ during handgrip was paralleled by slight decrease of Q-D and Q-S2, 3/ during occlusion, when QQ intervals returned to baseline also Q-D and Q-S2 returned to baseline, despite sustained elevation of arterial pressure, 4/ there were distinct oscillation in the time course of Q-S2 intervals, time course of Q-D in-tervals was relatively smooth, thus S2 and D may reflect different events contrary to common notation.

Keywords: systolic time intervals, ECG, polyphysiogra-phy, automatic signal analysis

1. IntroductionIt is believed that analyses of systolic time intervals

(STI) may provide information on heart muscle con-tractility and sympathetic activity [1]-[8]. We undertook a study to verify reliability of STI as indices of SNS ac-tivity. Determination of STI requires combined analysis of few biological signals. We used following signals: ECG, carotid pulse, and phonocardiogram and performed ‘manual’ analysis of few chosen intervals which are be-lieved reflect changes in sympathetic activity induced by experiment. We decided to replace cumbersome pro-cedure with automated detection of key events, what in turn allowed determination of consecutive systolic time intervals in the whole period of observation.

2. Method2.1. Events detection

Selected events were detected in the following order:1) Qi – trough of Q wave (ECG), local minimum preced-

ing peak of Ri wave (Fig. 1),2) Si – trough of S wave (ECG), local minimum follow-

ing peak of Ri wave (Fig. 2.),

Fig. 3. E – beginning of blood ejection from left ventricle; carotid pulse

Fig. 2. S – trough of S wave; ECG

Fig. 1. Q – trough of Q wave; ECG

Fig. 4. M – local maximum in carotid pulse curve imme-diately following E


Articles14

3) Ei – beginning of blood ejection from left ventricle (carotid pulse), local minimum of carotid pulse curve immediately following Si (Fig. 3),

4) Mi – local maximum of carotid pulse curve immedi-ately following Ei (Fig. 4),

5) Di – dicrotic notch (carotid pulse), local minimum oc-

Fig. 5. D – dicrotic notch; carotid pulse

Fig. 6. S2 – beginning of second ton caused by aortic valve closure; phonocardiogram

Fig. 7. Traces, events, time intervals. Traces: from the top: ECG, carotid pulse, phonocardiogram. Events: Q (trough of Q wave) – ‘*’; E (beginning of blood ejec-tion from left ventricular) – ‘o’; D (dicrotic notch) – ‘x’; S2 (beginning of second ton caused by aortic valve clo-sure) – ‘∆’. Time intervals: PEP (pre-ejection time pe-riod); Q-S2 (interval between Q and S2); Q-D (interval between Q and D); S2-D (interval between event S2 and D); QQ (interval between consecutive Q).

curring during predetermined period (250 ms) after Mi (Fig. 5),

6) S2i – beginning of second heart tone caused by aortic valve closure (phonocardiogram), maximum peak of the first positive deflection of the S2 complex preced-ing dicrotic notch (Fig. 6).

2.2. Determination of time intervals between chosen events

Time of event Xi appearance is denoted as t(Xi). Fol-lowing time intervals have been determined (Fig. 7.):1. pre-ejection period: PEPi = t(Ei) – t(Qi),2. time between trough of Q wave and aortic valve clo-

sure: Q-S2i = t(S2i) – t(Qi),3. time between trough of Q wave and dicrotic notch:

Q-Di = t(Di) – t(Qi),4. time between aortic valve closure and dicrotic notch:

S2-Di = t(Di) – t(S2i),5. QQi interval: QQi = t(Qi+1) – t(Qi).

2.3. Experiment and case presentationThirty young, healthy subjects participated in the

study. Subjects were supine, they performed two-minute isometric handgrip (HG) at 30% of the maximal volun-tary contraction. First HG was followed by four-minute rest. After second HG, two-minute occlusion of the working arm was applied, followed by two-minute rest. The study was undertaken to verify reliability of some in-dices of SNS activity; among them PEP and PEP/LVET.

Fig. 8. Presentation of time courses of chosen time inter-vals for case A. S2–D curve remains almost flat through the experiment


Articles 15

2.3.1 Case presentationCase A (Fig. 8)

One may see that changes of Q-S2 curve are followed faithfully by changes of Q-D curve. The constant time lag between Di (dicrotic notch) and S2i (aortic valve clo-sure) may be interpreted as transition time from aortic valve to carotid artery; thus Di and S2i are the same event observed in different places. Accordingly the constancy of time lag from Di to S2i (ca. 30 ms) is evident in time course of S2–D curve; this curve remains almost flat through the experiment.

The presentation of continuous time courses of cho-sen time intervals reveals that 1/ the distinct oscillations of QQ seems to not influence strongly these intervals, 2/ shortening of QQ during handgrip is paralleled by slight decrease of Q-D and Q-S2, 3/ during occlusion, when QQ intervals return to baseline values also Q-D and Q-S2 return to the baseline values, i.e. Q-D and Q-S2 do not depend on arterial pressure which remains elevated.Case B (Fig. 9)

It is evident that Q-D oscillate only slightly, the oscil-lations of Q-S2 are much greater. This is also evident in S2-D interval (from 10 ms to 70 ms). Assuming that the detection of S2 and D is correct it is rather impossible that the S2 and D reflect the same event. The possible ex-planation may be given in the paper of M. F. O’Rourke, T. Yaginuma, and A. P. Avolio [9].

They stated that usually the beginning of the diastolic wave, termed dicrotic notch, occurs immediately after aortic valve closure. These two events are sometimes

believed to be synonymous. However they may stem from different processes.

Closure of aortic valve is predominantly a cardiac phe-nomenon whereas dicrotic wave is predominantly a vas-cular one.

Is it possible that this explanation applies also to our results; thus S2 reflects the cardiac event – valve closure, whereas D reflects beginning of the diastolic wave. If such hypothesis holds true next question arises: why Q-D interval is much more stable than Q-S2.

Case B demonstrates also the constancy of PEP which changes only slightly during handgrip despite consider-able shortening of QQ.Case C (Fig. 10)

Maybe case C provides proof that the Q-S2 oscillations are not an artifact but true physiological phenomenon. It may be seen that characteristic of oscillation changes during handgrip and occlusion.

Fig. 9. Presentation of time courses of chosen time in-tervals for case B. Q-D curve oscillates only slightly, the oscillations of Q-S2 curve are much greater

Fig. 10. Presentation of time courses of chosen time in-tervals for case C. This case may prove that the Q-S2 oscillations are not an artifact but true physiological phenomenon.

3. ConclusionThe automated beat-to-beat event detection and time

intervals calculation allows to present continuous time course of systolic time intervals. Such presentation dem-onstrates oscillatory pattern of intervals changes, interde-pendence or lack of thereof between time intervals.

The time courses analysis of QQ, Q-D (time between trough of Q and dicrotic notch) and Q-S2 (time between trough of Q and aortic valve closure) time intervals dem-


Articles16

onstrated interesting features of relationships between these time intervals. 1/ The QQ duration influences weakly Q-D and Q-S2

intervals length, especially if the QQ variability is of respiratory origin.

2/ On the other hand the clear-cut shortening of Q-D and Q-S2 intervals seems to coincide with situation when increase of sympathetic activity is to be expected.

3/ Though, in some subjects S2 (aortic valve closure) precedes D (dictrotic notch) by constant time period, in others a delay between S2 and D was variable ranging-from 20 up to 70 ms. It shows that contrary to commonly held belief S2 and D do not reflect the same event.

AuthorsAnna strasz1, Wiktor Niewiadomski1,2, Małgorzata Skupińska3, Anna Gąsiorowska1,4, Dorota Laskows-ka1, Rafał Leonarcik1, *Gerard Cybulski 1,3

1 Department of Applied Physiology, Mossakowski Medical Research Centre Polish Academy of Sciences

2 Department of Experimental and Clinical Physiology, Medical University of Warsaw

3 Institute of Metrology and Biomedical Engineering De-partment of Mechatronics, Warsaw University of Tech-nology

4 Laboratory of Preclinical Studies in Neurodegenerative Diseases, Nencki Institute of Experimental Biology

*Corresponding author

references[1] S. S. Ahmed, G. E. Levinson, C.J. Schwartz, P.O.

Etringer, “Systolic time intervals as measures of the contractile state of the left ventricular myocardium in man”, Circulation, vol. 46, no. 3, 1972, pp. 559-571.

[2] J. T. Cacioppo, G. G. Berntson, P. F. Binkley, K. S. Quigley, B. N. Uchino, A. Fieldstone, “Autonomic cardiac control. II. Noninvasive indices and basal response as revealed by autonomic blockades”, Psy-chophysiology, vol. 31, no. 6, 1994, pp. 586-598.

[3] H. Schächinger, M. Weinbacher, A. Kiss, R. Ritz, W. Langewitz, “Cardiovascular indices of periph-eral and central sympathetic activation”, Psychosom Med., vol. 63, no. 5, 2001. pp. 788-96.

[4] K. A. Brownley, A. L. Hinderliter, S. G. West, S. S. Girdler, A. Sherwood, K. C. Light, “Sympathoadren-ergic mechanisms in reduced hemodynamic stress responses after exercise”, Med Sci Sports Exerc., vol. 35, no. 6, 2003, pp. 978-986.

[5] A. D. Goedhart, G. Willemsen, J. H. Houtveen, D. I. Boomsma, E.J. De Geus, “Comparing low frequen-cy heart rate variability and pre-ejection period: two sides of a different coin”, Psychophysiology, vol. 45, no. 6, 2008, pp.1086-1090.

[6] J. H. Meijer, S. Boesveldt, E. Elbertse, H.W. Berendse, “Method to measure autonomic control of cardiac function using time interval parameters from

impedance cardiography”, Physiol Meas., vol. 29, no. 6, 2008, pp. S383-S391.

[7] C. M. Licht, S. A. Vreeburg, A. K. van Reedt Dort-land, E. J. Giltay, W. J. Hoogendijk, R. H. DeRijk, N. Vogelzangs, F. G. Zitman, E. J. de Geus, B. W. Pen-ninx, “Increased sympathetic and decreased para-sympathetic activity rather than changes in hypo-thalamic-pituitary-adrenal axis activity is associated with metabolic abnormalities”, J. Clin. Endocrinol. Metab., vol. 95, no. 5, 2010, pp. 2458-2466.

[8] J. B. Hinnant, L. Elmore-Staton, M. El-Sheikh, “De-velopmental trajectories of respiratory sinus arrhyth-mia and pre-ejection period in middle childhood”, Dev Psychobiol., vol. 53, no. 1, 2011, pp. 59-68.

[9] M. F. O’Rourke, T. Yaginuma, A. P. Avolio “Physi-ological and pathophysiological implication of ven-tricular/vascular coupling”, Annals of Biomedical Engineering, vol. 12, 1984, pp. 119-134.


Articles 17

Remote Monitoring System For Artificial Heart

Bartłomiej Fajdek, Marcin Stachura, Michał Syfert, Paweł Wnuk, Michał Bartys

Submitted 27th June 2011; accepted 27th September 2011

Abstract: This paper presents a general description of the plat-

form for remote monitoring and supervision of heart assist devices (POLPDU). To facilitate a more in-depth under-standing of this system and its development, firstly, a brief discussion on severe heart failure problem are given. Then, the short description of the core modules of the system is included. The main emphasis is placed on the communica-tion between the heart assist devices and the central plat-form server. Finally, the directions of development of the system are discussed.

Keywords: remote monitoring system, telemetry, artificial heart, remote devices maintenance

1. IntroductionNowadays a growing number of patients are living

with heart failure. Cardiovascular diseases (CVD) are the leading cause of death and the main causes of illness and disability in developed countries. According to Central Sta-tistical Office (GUS) [1] the CVD are still the main causes of death (in 2010 – 46% of all deaths) despite the fact, that in Poland since the beginning of the 90s downward trend in mortality caused by cardiovascular diseases is observed. In many cases, e.g. for patients with severe or end stage heart failure, cardiac transplantation is the therapy of the last chance. However, there is still a large gap between the number of potential candidates for heart transplantation and the number of available hearts [2]. One of the pos-sible treatment method, in case of the severe heart failure, is application of mechanical heart supporting. It is used to partially or completely replace the function of failing heart. A growing number of patients, which require long-term support, involves the use of mechanical devices that can be used outside the hospital. Current technology can provide information and control systems that improve the patient’s safety and quality of life by the support of the users of heart support system, i.e. patients, doctors, technical stuff and device supplier.

In order to be able to monitor and supervise the heart support system in on-line mode, independently of the patient location, a remote monitoring system called CMS2 was developed. The system can monitor the device-related information such as control pressures, pressures in the air tanks and physiological-related information such as EKG signal. All the information can be transmitted in on-line mode from a patient’s unit to the central server. Stored data in central database can be analyzed by the qualified service staff. The monitoring center also provides wide range of functions connected with devices maintenance.

2. POLCAS heart support systemSince 1991, the Foundation for Cardiac Surgery

Development (FRK) in Zabrze has been working on developing the artificial heart. Nowadays, the Polish system for heart support POLCAS[3] consists of the arti-ficial ventricle POLVAD-MEV and the three controllers POLPDU-401, POLPDU-402 and POLPDU-501 (Fig.1).

Fig. 1. The family of POLPDU control units

Presented devices are designed to handle only one patient. The control units of the 401 and 402 series may be used only in hospital due to its big size, method of control and type of power supply. The control unit of 501 series is the latest product of FRK. Due to its much smaller size and weight it is significantly more mobile solution. For this reason, it can be also used during super-vised treatment conducted outside the hospital.

The construction of the drivers that belongs to the POLPDU family is very similar. The main difference between control units from 401 and 402 series is different source of the pneumatic supply. The POLPDU-402 con-trol unit requires a connection to an external air supply from the hospital pneumatic system for proper work. While the POLPDU-402 is completely independent from the external pneumatic installation. It is equipped with the onboard compressor which supplies artificial heart cham-bers. The main difference between 400 and 500 series is the type of the electro-pneumatic transducer that is responsible for generating modulated pressure wave sup-plying artificial heart chambers. All of the devices which belongs to the POLPDU family are equipment with two independent, redundant electro-pneumatic circuits, which can be used to control one or two pneumatic chambers in two modes: with and without the synchronization with the patient’s ECG signal. In case of a failure of one of the control paths, the system automatically switches to the redundant circuit.


Articles18

The control units are equipment with measuring system that collects all basic parameters of the unit (e.g., the con-trol pressure in the left/right circuit, the pressure in the air tanks – vacuum and overpressure). The drivers are also capable to measure and collect additional parameters of the artificial ventricle. It is possible to observe all the stored parameters on the device console screen (in case of 401 and 402 control units) and on a computer screen that is connected to the device (in case of 501 control unit). Regardless of the possibility to monitor the parameters locally, the units are equipped with the telemetry systems responsible for sending the parameter values to external devices and systems.

3. Objectives and scope of the monitoring systemDuring the design process of the Polish central artifi-

cial heart monitoring system (called CMS2) the following assumptions were made:• The primary tasks of the system are current moni-

toring and collecting archive parameters and mea-surements from particular devices and supporting the management of the life cycle of the devices that are part of the heart supporting system. Each unit is defined as specific, independent hardware component (e. g. artificial heart chamber, POLPDU driver unit, ECG device). Devices operate in sets (assemblies) that are attached to the patient.

• The system is designed to handle several considered type of devices. However, there must be a possibility to extend it to handle the facilities being developed in the future.

• The system design is modular so it can be expanded and launched in stages.

• The additional modules that realize advanced analysis of collected data in order to carry out device diagnos-tics and to support the patient’s medical diagnosis are planned to be designed and implemented.

• Selected functions of the system will be available remotely via the Internet, without the need to install dedicated software on the user’s machine. Different tools used to develop web applications will be used to fulfill this task.

It was assumed that, in the first stage of work, the following system functions will be implemented: moni-toring devices parameters and variables in real-time (requires elaborating embedded communication soft-ware for devices and communication modules for cen-tral server), creating archival databases and management of the devices configurations supporting basic issues of device maintenance.

4. Structure of the systemThe CMS2 system consists of several interacting

components that are operating on different devices and servers. The general structure of the presented system is shown in Fig. 2. One can distinguish the following ele-ments of the system:• Communication module (unit). Is is installed on the

POLPDU device. It is responsible for communication with the outside world via dedicated exchange proto-cols. The protocol was specially designed for the pur-pose of elaborated system. In the case of POLPDU-

402 it is an independent hardware module, while for POLPDU-501 it is an independent software module.

• Connection module. It is an independent module that runs on the monitoring center server. Its main task is to manage communication flow (in soft-real-time mode) between the monitored devices and the monitoring center in terms of data acquisition from devices as well as transferring control messages to them.

• Virtual console. It is an independent software that can be run on standard PC. It is used to directly com-municate (without the central database), with selected POLPDU driver unit. The central database is used only to authorize the virtual console and check if the user has privileges to monitor particular POLPDU unit.

• Data import and export module. It is an indepen-dent module that runs on the monitoring center server. It is responsible for reading and writing process data and devices configuration from / to the database of the central monitoring system.

• Monitoring and diagnostic module. The module is responsible for automated analysis of the measure-ment data for the purpose of advanced monitoring and diagnostics.

• Process and devices configuration database. In the CMS2 system two MySQL database are running on the monitoring center server. The first one is used to store the devices configuration. The second is a data-base used to store process variables and alarms col-lected in on-line mode from monitored devices.

• Graphical user interface module. The set of mod-ules that run on the www server and web browser side responsible for realizing the graphical user inter-face to all the data stored in monitoring center.

Fig. 2. The simplified structure of CMS2 system


Articles 19

5. CommunicationThe star topology was adopted in a communica-

tion network used to exchange the data between telem-etry modules, that runs on the POLPDU control units, and monitoring center. A network node (the connection module), that is installed on the main server, is the net-work coordinator. The POLPDU control units (402 and 501 series) are the sub-nodes which are equipment with the communication interfaces, respectively IK-002 and PK-CMS2. The methods of data exchange between the CMS2 central server and POLPDU control units are shown in Fig. 3.

frames using specially implemented functions. It is also responsible for conversion of transmission. The transmis-sion speed between the POLPDU-402 and the telemetry module is fixed to 115200 b/s. The speed of the data exchange between the telemetry module and wireless module is also constant and equal 38400 b/s. The pro-posed mode of communication is a real-time mode in the sense that each of the frame received from the coordinator is transmitted immediately to the POLPDU-402 driver, and each of the frame received from POLPDU-402 is passed immediately to the coordinator when the com-munication channel to the coordinator is released by the previous transfer.

The IK-002 communication protocol implements three different types of the data formats:• the format designed exclusively for the transfer of

slow changing measurement data - the data are sent periodically every 960 ms,

• the format designed exclusively for the transfer of fast changing measurement data - the data are sent periodically every 60 ms,

• the mixed format used to transfer slow and fast changing measurement data. In this case, the fast changing measurement data are sent periodically every 60 ms and the slow changing measurement data every 960 ms.

In order to provide proper time synchronization for all POLPDU-402 control units in the communica-tion system, which is compatible with IK-002 protocol, the principle of distributing time signals only from one source was adopted.

The central for remote monitoring and service is responsible for distributing the time signal to all control units. The system also assumes that each of the telemetry module has its own local real-time clock (LRTC). Every local real-time clock is counting time with 1 ms resolu-tion.

Fig. 3. The methods of data exchange between CMS2 server and POLPDU control units

Each of the network sub-node communicates only with the coordinator of the network. The peer-to-peer network transactions are not allowed, i.e. the transaction between any other nodes in the network.

The right to initiate the transactions in the network have all nodes subordinated to the network. The simulta-neous two-way communication mode is implemented to communicate between the coordinator and the selected network node. The network nodes exchange information using communication frames.

5.1. POLPDU 402 communicationThe control unit POLPDU-402 was equipped only

with the telemetry module that allows to setup the com-munication channels using the GSM wireless communica-tion network. This method of communication was chosen mainly because of the hardware limitations imposed by the designers of the POLPDU-402 unit. Since the com-munication was intended to informational purposes only the reliability level of such solutions was satisfactory. The communication protocol used to exchange the infor-mation between control units and the communication module installed on a central server is a specialized pro-tocol called IK-002 (Fig. 4).

The POLPDU internal protocol [4] is used to exchange the information between the POLPDU-402 control unit and the telemetry module. The telemetry module acts as simultaneous two-way transmitter of the POLPDU protocol frames. It interprets communication

Fig. 4. The simplified block diagram of the communica-tion system between the POLPDU-402 control unit te-lemetry interface and the monitoring center


Articles20

5.2. POLPDU 501 communicationThe communication module is mounted on the LPC2350

microcontroller [5], which is responsible for realization on the graphical interface of the heart support control unit POLPDU-501. The controller runs on the Linux oper-ating system and the measurement data are recorded in real-time. The communication process between the control device and communication module is realized by indepen-dent process and is carried out using pipelines (FIFO queues stored in the volatile memory of the microcontroller). The three pipelines are created: the first and second pipeline for storing measurement data (respectively slow and fast changing data with corresponding time-stamps), the third pipeline is dedicated for the orders that are sent remotely from virtual console.

The measurement data, received by the communication module, are stored in the cyclic buffers. From that buffer the data are sent to the central monitoring system server. The Internet is the main communication channel of the data transfer. The TCP/IP protocol is used, which provides cor-rect addressing of data with checksum and retransmission of lost packets support. In case of absence (or termination) of the wired connection stored data are sent via GSM network using the GPRS protocol. In such a way the redundancy of communication channels is realized.

The communication between POLPDU control unit and monitoring center server is preceded by the authen-tication of the device in the monitoring center. To make the authentication the unique ID number (optionally: together with its geographical position obtained from the GPS module) is send by the communication module. In a case of acceptance the main server sends the frame which confirms that the connection was established, oth-erwise a frame with the special error code that refuse the connection is sent. After receiving the confirmation frame from the server the communication module sends a request to receive a list of variables, which should be transmitted to the main server. This procedure is repeated every time when there is a loss of communication with the server. The exchange of the data between the com-munication module and the server via Internet channel is done using the dedicated protocol called PKCMS2, which was developed with the following assumptions:• encryption of the connection is optional and is real-

ized as encrypting the entire communication channel,• time-stamp is transmitted by the control device unit,• connection (TCP/IP tunnel) is maintained all the

time when the communication is active,• the measurement data packets are divided into two

groups: fast-changing data that are sent every second (in each package there are 100 - 200 measurement data for each variable), slow-changing data sent every second,

• the POLPDU control unit is responsible for the ini-tialization of the connection,

• the protocol during transmission of rare (non-cyclic) data is based on pure text. The cyclic transmitted data are transmitted in binary form;

• the protocol allows to define and distinguish fast-changing and slow-changing signals.

The header of each transmitted message starts with the information indicating the version of the protocol.

After the protocol number a unique device identifier (ID) is sent. The header ends with an information about mes-sage length.

5.3. Communication moduleThe connection module is a process that runs on the

central monitoring server. Its primary task is to handle com-munication with POLPDU devices. It is responsible for authentication of different devices and archiving received data in the central database via the import/export database module. The module is also responsible for authentica-tion of the virtual consoles and transmission to them the information derived from POLPDU devices. The autho-rization of the virtual console is performed in two stages: in the first step the user of the virtual console retrieves from a database 32-byte, disposable authentica-tion key for the connection. Then the key is sent to the connection module, where it is verified. Acceptance of the virtual console may be achieved only after positive key verification.

5.2. Virtual consoleIn order to allow remote monitoring and supervision

of the POLCAS system the dedicated software called Vir-tual Console (VC) was designed and implemented. The graphical interface of the application is shown in Fig. 4.

The application is used by the technical staff during active work of the control units (connected to the patient) as well as passive work (without the patient – service mode). The software can also be used for training the medical personnel. For this reason, the graphical inter-face of the application has been adopted in a similar way to the real interface of the control panel of the POLPDU unit. The application is compatible with POLPDU-402 and POLPDU-501 control units.

It is possible to connect with any POLPDU control unit, which was registered in the CMS2 system with use of VC. The software allows continuous monitoring of the basic parameters of the control unit (e.g., the con-trol pressure in the left chamber, the control pressure in the right chamber, the parameters of the control signals, the basic technical parameters: the pressure in the tanks: hypertension, vacuum, the major power status, battery charge status, etc.), parameters of the artificial heart (e.g., the blood-part pressure, the instantaneous volume of the artificial chamber, etc.) and several additional medical parameters (e.g., the patient’s ECG signal ). The key parameters used for assessing the proper operation of the device and artificial ventricular, so-called fast-changing

Fig. 5. The graphical interface of the virtual console ap-plication


Articles 21

signals, are visualized as waveforms. As a simple diag-nostics the alarm system is implemented. When any of the monitored parameter exceeds its safe range a special alarm message is displayed.

The application has been equipped with the supervi-sion module, which allows to change the basic parameters of the POLPDU control unit remotely (e.g., the discharge/suction pressure for the left/right path, the operating fre-quency of the left/right path, etc.). For the security rea-sons, turning on the remote supervision mode is possible only after the approval of the working mode of the control unit made by the local operator. Also positive verifica-tion of the user permissions on the central server CMS2 is required.

The software can operate in the two modes:• Direct mode. The application communicates directly

with the POLPDU control unit using the GPRS net-work. It is an emergency mode. It is used mostly in the case of the failure of the communication with central server.

• Intermediate mode. The application communicates with the connection module on CMS2 central server (the server communicates with the POLPDU control units). This is the default operation mode.

When the application is running in direct mode the connection redundancy is ensured (there are two inde-pendent communication channels – supported by two different GSM providers). When the application lose the connection via the active channel it is automatically switched to the redundant one. The switching process to redundant patch is signaled by the application with the appropriate message. In the intermediate mode, the redundancy of the connection is provided on the CMS2 central server (the communication via Ethernet and two different GPRS paths).

6. Monitoring centerThe primary CMS2 system server performs several

basic task such as: the device lifecycle management (storage and tracking configuration and use of devices and components), gathering, processing and sharing the archival measurement data, advanced diagnostic of the devices, permissions and access control management. The system has the ability to remotely access the Internet network without installing any additional software. In this case, it was decided to produce the prototype in accor-dance with the Web-desktop technique, which is based mainly on JavaScript an Ajax technology. The server part

was implemented in PHP language. The MySQL database was used to store the system configuration and archival data. To accomplish the additional computing tasks, like diagnostics, the special PHP and C++ modules were implemented.

In the field of devices configuration management the center provides authorized access to data and allows to perform the necessary operations connected with the lifecycle management of the devices that are the part of artificial heart assemblies (sets), such as:• management of individual devices (defining

according to the patterns, identifying the set of mea-surement variables and parameters, setting the opera-tional parameters),

• assembly management (creating, deleting, etc.),• recording service activities that are performed at the

level of devices and assemblies,• analyzing the history of use of particular devices in

particular assemblies,• user management (including the division into the dif-

ferent roles and groups) and the advanced permission management system in which permissions are allo-cated at the level of devices, users, groups, roles and individual activities,

• monitoring of the virtual console log entries.In addition, the Web interface of the CMS2 system

allows both the on-line data monitoring from the par-

Fig. 6. The graphical interface of the supervision module of the virtual console

The Start / Stop of the pneumatic path

The optional settings

The select parameter

mode

The selection of additional functionsThe

navigation buttons

Fig. 7. An example of the graphical web interface: the artificial heart sets configuration management

Fig. 8. The preview of the current state of the artificial heart control unit in web browser


Articles22

ticular device or assemblies, as well as viewing and ana-lyzing the archival data. The data can be presented in a graphic form (automatically or manually updates charts), and can be exported as files for use in external systems. The CMS2 system is equipped with the features to sim-plify the process of searching and navigating of the his-torical data.

7. ConclusionThe paper presents an outline of the remote moni-

toring system of the artificial heart control units. The presented system consists of two parts: (a) mounted on each of the monitored devices (the communication modules) and (b) the central monitoring system. During the design and implementation of the various functional modules, the emphasis was mainly placed on the adapta-tion of the selected technologies to the specific system. It was assumed that this system will be handled by both the engineers and those without in-depth technical knowledge (the cardiologists and the medical support staff).

The main part of the system operates in a soft real time mode. It is connected with the data acquisition from distributed artificial heart support units (assemblies). The data acquisition deals with the following issues: collecting the measurements with a fixed sampling rate, assigning time stamps to the measurement data, tasks related to the internal transfer of streaming data from artificial heart control unit to the communication interfaces. These tasks are performed by the telemetric packages installed on the particular devices, the central communication module and data import/export services running on central server. This part of the system is fully implemented and now is under tests.

The second part of the system is connected with the management of the life-cycle of particular devices and the configurations (assemblies) in which they are working. It provides the user interfaces to configuration data. It also enables on-line device monitoring based on the data stored in database (it does not use direct communication with the devices). The current version of this part of the system implements basic features and functions. The future development of the system will be focused on those modules. In the next steps of devel-opment the following features will be added: manage-ment of patients and doctors (data storage, configuration and access), tracking the process of device selling and lending, detailed and configurable device description and definition, tracking the supply chain connected with device production and device storage.

Acknowledgements This work was supported by the National Centre for

Research and Development (NCBiR) under the Project “Development of metrology, information and telecom-munications technologies for the prosthetic heart” as part of multiannual “Polish Artificial Heart” Program (task 2.1 – Developing a system of automatic control and su-pervision of work for extracorporeal cardiac prosthesis).

AUTHORSBartłomiej Fajdek*, Marcin Stachura,Michał Syfert,Paweł Wnuk,Michał Bartyś – Politechnika Warszawska, Instytut Au-tomatyki i Robotyki, ul. Św. A. Boboli 8, 02-525 War-saw, Poland,[email protected],[email protected], [email protected], [email protected],[email protected]


References:

[1] http://www.stat.gov.pl – Podstawowe informacje o rozwoju demograficznym Polski w latach 2000 – 2010, Główny Urząd Statystyczny

[2] T. Zielinski, A. Browarek, M. Zembala, J. Sadowski, M. Zakliczynski, P. Przybylowski, K. Roguski, A.B. Kosakowska, J. Korewicki and POLKARD HF investigators, Risk Stratification of Patients With Severe Heart Failure Awaiting Heart Transplanta-tion--Prospective National Registry POLKARD HF, Transplantation

[3] Kustosz, R.; Gawlikowski; M., Darłak,M.; A. Kapis: Pneumatyczny system wspomagania serca POLCAS do zastosowań długoterminowych, Śląskie Warsz-taty Biotechnologii i Bioinżynierii Medycznej – BIO-TECH-MED. Silesia 2005, Cardiac Surgery Development Foundation, Zabrze 2005.

[4] Czak, M.: Szczegółowa specyfikacja protokołu POLPDU w warstwie fizycznej i w warstwie łącza danych, Opracowanie Fundacji Rozwoju Kardiochi-rurgii im. Zbigniewa Religi, Zabrze 2010, s. 17.

[5] http://www.embeddedartists.com


Articles 23

Accuracy of The Element Geometry Mapping Using Non-Invasive Computer Tomography Method

Mirosław Grzelka, Lidia Marciniak, Bartosz Gapinski, Grzegorz Budzik, Andrzej Trafarski, Joanna Augustyn-Pieniazek, Mariusz Gaca

Submitted 15th October 2011; accepted 12th November 2011

Abstract: Non-invasive method of the measurement and identifi-

cation of shape and geometrical characteristics of meas-ured detail based on the computer tomograph is widely applied in medicine with high efficiency. The paper is aimed at presenting the metrological analysis of the ac-curacy of shape reproduction. The examined shape was described due to the tomography and compared to the results of coordinate measurement. The comparative ac-curacy analysis of the tomography was performed for the typical details, as well as for the masters made with high accuracy. Using the results of the analysis, the respec-tive models of measured details were created and used as a nominal in the further investigations. The results of the measurement with tomography were compared with the results achieved from the measurement by means of typical coordinate measuring machines and by means of 3D measurement optical scanner. Analysis of the gained values of the form deviations and important geometrical characteristics enabled to perform complex analysis of the shape reproduction of the examined detail with the method of computer tomography. The main purpose of the performed researches was the determination of the reproduction accuracy of surface measured with non-invasive computer tomography meth-od. The investigations and analysis have been performed with the following equipment: computer tomograph, 3D printer and coordinate optical scanner supported with specialized software for advanced CAD data analysis.

Keywords: computer tomography, optical coordinate measuring technique, reverse engineering, accuracy of workpiece mapping

1. IntroductionThe computer tomography became one of the basic

diagnostic and measuring tool in modern medicine. Continuous development of this technique, which is oriented to both minimize radiation and obtain maximum data, determine innovation in solving technical problems. In the past, computer tomography systems was applied only to the diagnosis. Nowadays, there are more and more new applications of this non-invasive measuring technique not only for medicine but also for archeology, mechanical engineering, metrology, criminology or even art.

The goal for researchers was to evaluate accuracy of geometric feature mapping using non-invasive computer tomography. Research and analysis were conducted using computer tomograph, special software dedicated to analysis and the transformation of measurement data,

optical 3D measuring scanner, advanced software for CAD data analysis.

Processing of the measurement data and its transforma-tion into 3D model depends on the type of the examined model surface. In the presented researches, the authors have concentrated on the 3D freeform surfaces describing any type of detail. For the elements with curved surfaces and lines (2D and 3D) and with vertical and horizontal surfaces related to the coordinate system of tomography, the tolerance could be defined as a distance between the point of the CAD model and the point of the real model in the line orthodox to the CAD model surface. Figure 1 and 2 show the correlations between the layer thickness and other geometrical parameters of the model in the inter-section along y-axis.

The correlations between the parameters presented in the Figure 3 could be written as follows:

– for the outer flat declined surface

sinhd γ= ⋅ (1)

– for the outer curved surface

2 2 2 sinh hd r r γ= − + + + ⋅ ⋅ (2)

Where: P(x,y,z) – point in the measured surface, d – distance between the point in the curve and the measuring point, r – curve radius, h – distance between subse-quently captured pictures, N – normal vector in the point P of the surface, n{nx, ny, nz} – elementary vector.

Geometrical analysis in the direction is dependent on measuring parameters of the tomograph. There is an opportunity of its regulation in order to meet expected accuracy with respect for technical limits of the measuring device. According to that, both the accuracy of used measuring system and algorithms of data trans-formation into digital 3D form are factors which deter-mine the accuracy of element mapping with computer tomography.

Fig. 1. Influence of the sampling step (tomograph photos) on the model accuracy: a) thickness 1.0 mm, b) thickness 0.5 mm, c) thickness 0.2 mm


Articles24

2. Research results The first part of research was conducted on the

computer tomography Hitachi ECLOS 16 which was shared by KIE Sp. z o. o. in Poznan. The distance between the x-ray source and each separate picture was 0.5 mm for every single element that was put to the tests.

The second part of research, which dealt with the eval-uation of the surface mapping accuracy using computer tomography and the optical 3D measuring scanner, was conducted in accordance with the process of reverse engi-neering (RE). Elements selected to research are charac-terized by geometrical features and surfaces that are the most popular in both engineering industry and medicine.

On each step of the geometry mapping process in use were available methods and algorithms with simulta-neous verifying the precision of each transformation in order to guarantee the accuracy of element mapping on the highest possible level. Particular attention was paid to the individual characteristic element and geometric features formed by the free surfaces. In the pictures of

Fig. 2. Layer modeling methods: a) negative tolerance, b) positive tolerance, c) mixed tolerance

Fig. 3. Geometrical relations in the model’s layers: a) intersection of the model with flat side, b) intersection of the model with curved side

Fig. 4. Spherical element. Deviation of the surface iden-tification ± 0.3 mm. Diameter deviation 0.0135 mm (least square fitting element)

Fig. 5. 3D surface with various curvatures and small de-tails. Deviation of the surface identification ±0.5 mm

Fig. 6. View of a pig spinal verterbra and image of a pig spine measurement on tomography

particular elements (Figures 4 and 5), there are presented following parameters: a) picture of element, b) results gained from tomography, c) initial 3D model, d) modi-fied 3D model, e) 3D element gained from coordinate scanner, f) comparison (inspection) – deviation of the model.

3. ConclusionsMetrological analysis of the whole reproduction

process of the geometrical characteristics of an object measured with computer tomography, specialized CAD software and 3D coordinate scanner leads to the following conclusions:

– deviation of the described process for the elements consisting of 3D freeform flat and convex surfaces (respectively concave ones), compared to the whole dimensions of the object does not exceed ±0.2 mm,

– accuracy of the spherical surface reproduction with substantial curvature related to the whole dimensions does not exceed ±0.3 mm (Fig. 4),

– analysis of the surface built up with freeform surfaces of various and variating curvature based on the proposed methodology does not exceed ±0.5 mm (Fig. 5),

– analysis of geometric deviations of mapping the circle described by free surfaces with varying and vari-able curvature (Fig. 6), based on the methodology devel-oped does not exceed ±0.7 mm (Fig. 7).


Articles 25

The reproduction process of the freeform surface is very complicated from technical point of view and requires advanced equipment and software. Computer tomograph supported with Rapid Prototyping system and coordinate measuring technique enables to perform the metrological analysis of the geometrical element with certain accuracy. The measurement method is non-inva-sive and in some applications is irreplaceable.

The performed investigations and metrological anal-ysis lead to the ability to describe and determine the reproduction accuracy of any kind of object.

AUTHORS

Mirosław Grzelka* Poznan University of Technology, Institute of Mechanical Technology, Division of the Metrology and Measurement Systems, Poznan, Poland, [email protected]

Lidia Marciniak – Poznan University of Technology, Institute of Mechanical Technology, Division of the Metrology and Measurement Systems, Poznan, Poland

Bartosz Gapiński – Poznan University of Technology, Institute of Mechanical Technology, Division of the Metrology and Measurement Systems, Poznan, Poland,

Grzegorz Budzik – Rzeszow University of Technology, Rzeszow, Poland.

Andrzej Trafarski – Poznan University of Technology, Institute of Mechanical Technology, Division of the Metrology and Measurement Systems, Poznan, Poland.

Joanna Augustyn-Pieniążek – AGH University of Science and Technology, Cracow, Poland.

Mariusz Gaca – Student of Poznan University of Tech-nology, Poznan, Poland.


References[1] Dowsett, Kenny, Johnston, The Physics of Diag-

nostic Imaging. Chapman & Hall Medical, 1998, ISBN 0-412-40170-3.

[2] Tai C., Huang M., “The processing of data points basing on design intent in reverse engineering”, International Journal of Machine Tools & Manu-facture, no. 40, 2000.

[3] Cygnar M., Miechowicz S., Budzik G., “The Influ-ence of CT Scanning Parameters on The Inaccu-racy of The Computer Tomography Imaging of The Turbine Blade For Reverse Engineering”. In: Inter-national Conference on Machining and Measure-ment of Sculptured Surface, MMSS 2006, Institute of Advanced Manufacturing Technology, Kraków 2006, pp. 281-288.

Fig. 7. The results: a) image of a model directly from the tomography, b) model with smoothing in the software for data analysis, c) comparison of the CT measurement results with model obtained from direct measurements on the optical scanner

a)

b)

c)


Articles26

[3] Cuilliere J.C., “An adaptive method for the auto-matic triangulation of 3D parametric surfaces”, Computer Aided Design, Elsevier, vol. 30, no. 2, pp. 139-149.

[4] Luhmann T., Robson S., Kyle S., I. Harley I., Close Range Photogrammetry: Principles, Tech-niques and Applications, Published by Whittles Publishing, Scotland, UK 2006, ISBN 0-470-10633-6.

[5] Luhmann Th., Nahbereichsphotogrammetrie, Wichmann, Heidelberg, 2000. (in German)

[6] Ratajczyk E., „Tomografia komputerowa w pomi-arach geometrycznych 3D”, PAK, no. 2, 2011, pp. 220-223. (in Polish)

[7] Ratajczyk E., „Tomografia komputerowa CT w zastosowaniach przemysłowych. Cz. I. Idea pomiarów, główne zespoły i ich funkcje”, MECHANIK Miesięcznik Naukowo-Techniczny, no. 2,2011, pp. 112-118. (in Polish)

[8] Ratajczyk E., „Tomografia komputerowa CT w zastosowaniach przemysłowych. Cz. II”, MECHANIK Miesięcznik Naukowo-Techniczny, no. 03, 2011, pp. 226-231. (in Polish)

[9] Ratajczyk E., „Tomografia komputerowa CT w zastosowaniach przemysłowych. Cz. III. Tomo-grafy i ich parametry, przykłady zastosowań”, MECHANIK Miesięcznik Naukowo-Techniczny, no. 04, 2011, pp. 326-331. (in Polish).

[10] Ratajczyk E., „Tomografia komputerowa CT w zastosowaniach przemysłowych. Cz. IV. Opro-gramowania, parametry dokładności i metody ich wyznaczania”, MECHANIK Miesięcznik Naukowo-Techniczny, no. 05-06, 2011, pp. 474-483. (in Polish)

[11] Wieczorowski M., Miechowicz S., Markowska O., Budzik G., Grzelka M., “Methodology and analysis of the reproduction and fitting the implant or filling destined for deformed scull In the processes of RE and RP with coordinate metrology”. In: IC2B’2009, International Conference on Bioengineering & Biomaterials IC2B’2009, 18-20.03.2009, Meknes, Marocco.

[12] Grzelka M., Trafarski A., Wieczorowski M., Staniek R., “Metrological analysis of the reproduced detail accuracy with the non-invasive computer tomog-raphy method combined with coordinate measure-ments”. In: IC2B’2009, International Conference on Bioengineering & Biomaterials, 18-20.03.2009, Meknes, Marocco.

[13] Grzelka M., Budzik G., Markowska O., Gapiński B., „Analiza dokładności wykonania implantu czaszki z wykorzystaniem technologii trójwymiarowego druku (3DP) i współrzędnościowego skanera 3D”. In: IX Sympozjum Modelowanie i pomiary w medy-cynie. Krynica 10-14.05.2009. Ed. by Janusz Gajda, Wydawnictwo Katedry Metrologii AGH, 2009, pp. 197-202, ISBN 978-83-61528-08-1. (in Polish)

[14] Grzelka M., Trafarski A., „Metrologiczna analiza dokładności odtworzenia kształtu bada-nego elementu nieinwazyjną metodą tomo-grafii komputerowej z wykorzystaniem współrzędnościowych pomiarów”. In: IX

Sympozjum Modelowanie i pomiary w medyc-ynie, Krynica 10-14.05.2009. Ed. by Janusz Gajda. Wydawnictwo Katedry Metrologii AGH 2009, s, 183 – 190, ISBN 978-83-61528-08-1. (in Polish)

[15] Budzik G., “Possibilities of utilizing 3DP tech-nology for foundry mould making”, Archives of Foundry Engineering, vol. 7, issue 2, 2007, pp. 65-68.

[16] Liu W., Rapid Prototyping and engineering appli-cations – a toolbox for prototype development, Taylor & Francis Group, 2008.


Articles 27

The Models for The Comparison of Roundness Profiles

Šárka Tichá, Stanisław Adamczak


Abstract: Roundness is one of the mostly frequently observed

geometrical deviations. A profile of a machine part con-sisting of simple harmonic courses or their sums de-pends on a number of factors. Modeling of geometrical deviations requires applying the Fourier to extend each periodical function into an infinite series of harmonic components. Using of harmonic analysis we are able to determine the values of amplitudes of single components and their phase shifts. These parameters can be used for example for comparison of roundness profiles by means of correlation coefficients. The obtained correlation co-efficients are concrete parameters, on the basis of which it is possible to determine the degree of conformity and unity between the compared profiles. The comparison of evaluated roundness profiles can be represented as math-ematical statistical models. In these models is detected sequence of single steps, which are necessary for deter-mination of required correlation coefficients [1, 2] .

Keywords: roundness, correlation coefficients, math-ematical models, comparison

1. IntroductionThe functional surfaces of rotary machine component

need to be as accurate as possible from the point of view of geometrical quality. To assure good co-operation of

rotary parts (e.g. couple journal-bearing), following are necessary:– high dimensional accuracy, – high form and position accuracy,– low surface waviness,– low roughness parameters.

One of the most relevant factors of rotary surfaces macro geometry is roundness. The value of roundness deviation DR (EFK) and profile for have an effect on the functioning of a machine part and indirectly on the cooperation of ro-tary surfaces. In case of a rotation, the profile form directly influences vibrations.

A profile of machine part at dependence on various in-fluences can consist of the simple harmonic components (oval, tri-lobbing, square…), or their sums. This depends on the surface machining conditions of during the manufac-turing process. During modeling of geometric deviations Fourier’s series are used. Thus, each periodical function can be extended into an infinite series of harmonic compo-nents. Irregularities of roundness profile correspond to the irregularities periodically occurring on basic wavelength, which is equal to perimeter of profile. After a harmonic analysis it is possible to assign amplitude and phase shift to all irregularities which define the size and the charac-ter of irregularities on the surface of machine part. These parameters can be used for additional processing, e.g. for comparison of roundness profiles courses [2].

Fig. 1. The model of comparison roundness profiles measured by the relative method for various parameters a and b


Articles28

2. The comparison of roundness profilesMostly methods for the comparison of roundness pro-

file involve evaluation on the basic of roundness devia-tion DR (EFK). However, this evaluation is not objective, nor does it concern the geometrical form of machine part as the whole. Two components of roundness profile can be evaluated by the same value of DR, but in fact the courses will be different [2].

Objective evaluation of geometrical form of rotary machine parts requires applying appropriate measuring methods. The instruments used for that register round-ness profiles courses in the form of measuring signal. The signal it then processed and the resulting parameters can be applied for instance to:– comparison of roundness profiles obtained with

different methods of measurement,– comparison of roundness profiles obtained with one

method of measurement,– the evaluation of form changes of parts caused by

aging,– assessment of machine-tools from viewpoint of their

reliability etc.Roundness profiles can be compared objectively us-

ing a statistical method, of calculation of correlation. The correlation coefficients can be calculated either by means of the Pearson method (linear correlation) or by the less known and less frequently used Spearman method (range correlation). The obtained correlation coefficients are then used for the determination of the degree of conform-ity and unity between the compared profiles.

3. The models for the comparison of round-ness

It is possible to represent the concept of comparison of the evaluated roundness profiles as mathematical mod-els. These models are composed of a sequence of steps

necessary to determine the correlation coefficients for roundness profiles. The correctness of developed models was verified experimentally, see ref. [2].

Figure 1 shows mathematical model for the compari-son of roundness profiles obtained by measurement with a relative method at different values of method parameters a and b. It is necessary to transform measured profiles Fn(j) into real ones Rn(j). This transformation requires application of complicated mathematical formulas. The result of the harmonic analysis of the transformed pro-files are values of amplitudes and phase shifts of each harmonic components, which are the used for calculation of correlation coefficients. On the basis it is possible to determine the type of correlation and degree of unity.

Figure 2 presents a mathematical model used for comparing two roundness profiles measured with non-reference instruments (i.e. with a change in radius). The principle of non-reference measurement is simple and so is the mathematical model, which does not require any complicated transformation. The measured profile is the same as the real one. From harmonic analysis of meas-ured profiles we obtain values of quantities demanded for the calculation correlation coefficients.

The third mathematical model is applied to the com-parison of roundness profiles measured with relative and non-reference methods (Fig. 3). This model is more com-plicated than model two became of the complex trans-formation theory of relative profile measuring. The steps following the transformation are identical with those in other models.

In all the three models it is necessary to take account of the influence of several disruptive factors on the ac-curacy and objectivity of comparison. There are mainly method errors, errors of the measuring system and well as errors made during the mathematical processing of in-formation.

Fig. 2. The model of comparison roundness profiles measured by the non-reference methods


Articles 29

4. ConclusionAt present much emphasis is put on evaluation of form

accuracy of rotary machine parts. In practice the methods and procedures have to be adjusted for each case indi-vidually. The suitability and objectivity are assessed by comparing the obtained harmonic components of round-ness profiles. Also, it is essential to calculate the coef-ficients of correlation between the values of amplitudes and of phase shifts of each harmonic component. It is possible to represent the algorithm of this comparison by the means of mathematical models and procedures ena-bling calculation of correlation coefficients. These mod-els can be used, after small modifications, for the rep-resentation of the conception concerning the evaluation of form changes of rotary machine parts resulting from ageing or the assessment of technological heredity etc.

ACKNOWLEDGEMENTS

This paper was created under the Grant of the Minis-try of Education, Youth and Sports Czech Republic No. 2548/2011/F1

AUTHORSŠárka Tichá –VŠB-Technical University, Department of Working and Assembly, 17. listopadu 15, 708 33 Ostrava-Poruba, Czech Republic, e-mail: [email protected]ław Adamczak* – Politechnika Świętokrzyska, al. Tysiąclecia Państwa Polskiego 7, 25-314 Kielce, Poland,e-mail: [email protected]


References [1] Adamczak S., Odniesieniowe metody pomiaru zary-

sow okragłości maszyn.. Politechnika Świętokrzyska, Kielce 1998. 181 p. PL ISSN 0239-4979. (in Polish)

[2] Tichá Š. Využití korelačního výpočtu k porovnávání profilů kruhovitosti. Doktorská disertační práce.VŠB-TU Ostrava 1999. (in Czech)

Fig. 3. The model of comparison roundness profiles measured by the relative and non-reference method


Articles30

Advanced Rehabilitation Device Based on Artificial

Muscle Actuators with Neural Network Implementation

Kamil Židek, Ondrej Líška, Vladislav Maxim

Submitted 27th June 2011; accepted 26th September 2011

Abstract: In this paper the rehabilitation device for upper arm

based on artificial muscles is introduced. Presented au-tomated rehabilitation device has three degrees of free-dom: 2 DOF in arm and 1 DOF in elbow that provides almost all basic rehabilitation exercises. Artificial pneu-matics muscles will be tested in connection with spring and antagonistic connection. This system provides lifting and falling of arm construction based on patient force. There is possibility to generate help force during rehabili-tation or opposite load. Artificial muscles are controlled by pneumatic valves terminal from micro computer based on MCU. Higher level control system provides artificial intelligence implementation based on neural network for prediction and change of load according sensor values history (incremental sensor, pressure sensor). For proto-type testing there is described usability of industrial robot to test precision of load and trajectories during rehabili-tation. This automatic rehabilitation device will help to reduce therapeutics work with patient, automate and im-prove rehabilitation process.

Keywords: rehabilitation, automation, artificial muscle, control

1. Introduction Automated rehabilitation is nowadays in fast develop-

ment in physical therapy [3]. Automated rehabilitation is a special branch of rehabilitation medicine focused on devices that can be used by people to recover from physi-cal trauma. The first results in this area are described for example in these articles [5], [6]. Automatized machines are very suitable for implementation to rehabilitation area. They replace manual procedures by autonomous exercises. There are three main areas of physical therapy: cardiopulmonary, neurological, and musculoskeletal. Though automated rehabilitation has applications in all three areas of physical therapy, most of the work and development is focused on musculoskeletal uses. Mus-culoskeletal therapy assists in strengthening and restor-ing functionality in the muscle groups and the skeleton, and in improving coordination. In the current paradigm of physical therapy, many therapists often work with one patient, especially at the early stages of therapy. Auto-matic rehabilitation allows rehabilitation to occur with only one therapist, or none with adequate results. Auto-mated systems allow more consistent training program with automated tracking patient’s progress and shifting the stress level accordingly, or making recommendations to the human therapist. In the future automated rehabili-

tation promises effective results. As the technology de-velops and prices decrease, rehabilitation systems will be available in everyday life.

2. Construction Construction of prototype device is mainly based on

standardized aluminum profiles and rotary joints. All actuators are based on pneumatics artificial muscles. Artificial muscles are suitable for these devices because of their flexibility especially in end positions. Present-ed automated rehabilitation device has three degrees of freedom: 2 DOF in arm and 1 DOF in elbow that pro-vides almost all basic rehabilitation exercises as it was described by [1]. Artificial pneumatics muscles will be tested in connection with spring and antagonistic connec-tion according design [4]. This system provides lifting and falling of arm construction. Possibility to generate help force during rehabilitation or opposite load is there. Artificial muscles are controlled by pneumatic valve ter-minal from micro computer based on MCU. Higher level control system provides artificial intelligence based on neural network for prediction and change of load accord-ing sensor values history. Simplified kinematics scheme of rehabilitation device is displayed in Fig. 1.

1

2

3

Fig. 1. Kinematics scheme of rehabilitation device

The mechanism is fixed to double chair for rehabilita-tion in a comfortable sitting position. Rehabilitation sys-tem is designed for both arm (left, right), but not in same time. The patient must change chair for adequate arm.

3. Control systemThe main control part is based on 8bit MCU (ATMEG-

A128L microcontroller) which control pneumatics artifi-cial muscle and cooperate with sensors, detailed scheme is pictured in Fig. 2. The main output part for switching the electromagnetic valves is integrated transistor array, which is directly connected to the microcontroller output. Device is equipped by display and keypad for monitoring of rehabilitation process and practices selection. Micro-controller communicates with a PC by serial link (US-


Articles 31

ART). There is possibility to connect device to mobile PC by USART to USB transducer.

The basic control algorithm for the control of the reha-bilitation process consists of three parts: a) Regulatory part,b) Protective part,c) User part.

The regulatory part of the algorithm ensures that the rehabilitation device copying required trajectories. An important feature for controlling the rehabilitation is pressure sensing of patient arm. Based on this property we can achieve a suitable speed of shoulder rehabilitation practices in the prescribed mode.

The protective part of the algorithm is designed to en-sure safety of the patient during rehabilitation exercises, where for example: in case of detecting of acceleration level over the certain threshold device has to stop the movement of limb within a few milliseconds. The impor-tant elements for detection is included acceleration sen-sor, gyroscope and temperature sensor of human body (to monitor of the muscles during practice).

The user part of the algorithm ensures communication between the user and the microcontroller. There is pos-sible to choose several types of rehabilitation practices with various parameters. All data during practice are dis-played on the display unit. Main display is not able to dis-play all values at once, so individual information rotated cyclically in a time loop [7].

Diagram in the Fig. 3 illustrate rehabilitation device control system for first prototype with external measur-ing card and description of safety circuits. High level control system is based on Industrial PC alternatively OPLC or Tablet PC with additional information from testing sensors.

4. Neural Network implementationUtilization of artificial intelligence is widely applied in

present. There are many experiments with various algo-rithms, methods and their combination e. g.: neural net-works, theory of learning machines (machine learning), fuzzy logic, genetic algorithms, experts systems etc. As it was mentioned above the pneumatic artificial muscle is now unused mostly in reason of complicated control because of there is high non-linearity.

Standard types of regulator fail what is main rea-son of using neural networks. Sequence of operation is visible in the Fig. 4 on the left. It describes operation of rehabilitation device. In diagram is rehabilitation device represented by operation system. In the Fig. 4, right there is displayed neural network with assigned specific values of four inputs to one output important for correct function NS. There is used NS with back propagation teaching.

Proposed NS has been able to learn with acceptably mistakes of learning in less than 80 cycles. After leaning NS there is next step to create tested set of data. After creating tested set of data there was tested NS and inter-pretation of testing NS. Interpretation is running on basis of comparison real and expected results of classification.

Fig. 2. Principle scheme of control system

MCU

ATMEGA128

SENSOR OF ACCELERATE

MEMS

ACCELEROMETER

TEMPERATURE

SENSOR

SPEED SENSOR

INCREMENTAL

SENSOR OF

ROTATE

GYRO

VOLTAGE DIVIDER

LCD DISPLAY

AMPLIFIER + SIGNAL

ADJUSTMENT

PRESS SENSOR

FOR PATIENT

HAND

BUTTONS 8x

(START, STOP,

TOTAL STOP, etc.)

ARRAY OF

PNEUMATIC

VALVES 6x

USART/USB

T

PRESSURE SENSOR FOR

PNEUMATICS

CIRCUIT

Fig. 3. Diagram of Rehabilitation Control System

Fig. 4. Neural network and control scheme of implemen-tation

Fig. 5. Neural network and control scheme of implemen-tation


Articles32

It is defined some of coefficients of estimate precision of classification. All vectors were classified correct. So it sit about aptitude of application NS in rehabilitation de-vice. In Fig. 5 is displayed Error graph of created neutral network.

5. Industrial Robot testing SystemTesting platform is based on articulated robot with

5 DOF Mitsubishi RV-2AJ [2]. The robot is controlled from external C# application and serial port. Rehabilita-tion device is connected to end component of robot by flexible coupling. We can reach any position in 3D robot workspace to define testing trajectory easy in drawing area. Testing device can help check safety of rehabilita-tion device before testing with life patient. Simulation of testing device and rehabilitation system is displayed in Fig. 6.

AUTHORS

Ing. Kamil Židek* – Technical University of Kosice, Faculty of mechanical engineering, Department of Bio-medical engineering automation a measuring, Košice, 042 00, Slovakia, [email protected].

doc. Ing. Ondrej Líška – Technical University of Kosice, Faculty of mechanical engineering, Department of Biomedical engineering automation a measuring, Košice, 042 00, Slovakia, [email protected].

Vladislav Maxim – Technical University of Kosice, Fac-ulty of mechanical engineering, Department of Biomedi-cal engineering automation a measuring, Košice, 042 00, Slovakia, [email protected].


References[1] Cuccurullo Sara J., Physical medicine and rehabili-

tation board review, Demos Medical Publishing, 2010, p. 938.

[2] Hopen J. M., Hosovsky A., “The servo robustifica-tion of the industrial robot”. In: Annals of DAAAM for 2005, 19-22 October 2005, Opatija, DAAAM International, Vienna, Austria 2005. ISSN 1726-9679, pp. 161-162.

[3] Kommu I.S. et al., Rehabilitation Robotics, I-Tech Education and Publishing, Vienna, Austria 2007. ISBN 978-3-902613-01-1. P. 638.

[4] Piteľ J., Balara M., Boržíková, J., “Control of the Actuator with Pneumatic Artificial Muscles in An-tagonistic Connection”, VŠB – Technical University of Ostrava, vol. LIII, no. 2, 2007. ISSN 1210-0471, pp. 101106.

[5] Pons J.L., Rocon E., Ruiz A.F., Moreno J.C., “Up-per-Limb Robotic Rehabilitation Exoskeleton: Tremor Suppression”. In: Rehabilitation Robotics, Bioengineering Group, Instituto de Automática Industrial – CSIC, Spain, 2007. ISBN: 978-3-902613-04-2, pp. 453-470.

[6] Sarakoglou I., Kousidou S., Nikolaos G. et al., Exoskeleton-Based Exercisers for the Disabilities of the Upper Arm and Hand in Rehabilitation Ro-botics, UK, 2007, pp. 499-522.

[7] Sun J., Yu Y., Ge Y., Chen F., “Research on the Multi-Sensors Perceptual System of a Wearable Power Assist Leg Based on CANBUS”. In: Pro-ceedings of the 2007 International Conference on Information Acquisition, 2007.

Fig. 6. Simulation of testing device and rehabilitation system

6. ConclusionThe project is using artificial muscle as joint actuator

because of silent operation and flexibility during move-ment, start and end position. The developed automated rehabilitation device will save therapeutics capacity, pro-vide improving in prediction of increasing and decreas-ing load during rehabilitation exercises according patient progress. System is monitored by many sensors during operation together with low level safety circuit. Predic-tion of load is based on integrated neural network algo-rithm. There is designed prototype testing system based on industrial robot. Next works after successful testing process will be development of mobile version without chair as orthosis (exoskeleton) for direct rehabilitation in patient household.

AcknowledgementsThe research work is supported by the Project of the

Structural Funds of the EU, Operational Program Re-search and Development, Measure 2.2 Transfer of knowl-edge and technology from research and development into practice: Title of the project: Research and development of the intelligent non-conventional actuators based on ar-tificial muscles ITMS code: 26220220103.

VEGA-0185-10 Study of power semiconductor conver- terswith high power conversion efficiency.


Articles 33

Comparative Studies of Various Methods of Mounting The Implant Mandrel within The Bone

Marcin Zaczyk, Danuta Jasi ska-Choroma ska


Abstract: The paper presents results of studies pertaining to eval-

uation of a quality (compressive strength of mounting the implant within the bone) of mounting the implant mandrel within the osseous tissue for various relevant techniques.The presented results are related to selected techniques of mounting the implant mandrel within a bone. The compar-ison has been carried out for cement-less mandrels with a smooth mandrel of the endoprosthesis, porous cement-less endoprostheses as well as mandrels made of bioactive materials.

Keywords: mounting of an implant mandrel within the osseous tissue, endoprosthesis, implant-bone contact

1. IntroductionA development of the biomedical engineering makes

many numerous types of implant mandrels available. They employ various techniques of mounting within the osseous tissue. The paper contains a comparison of the most often applied techniques of mounting the implant mandrel within the osseous tissue. The main aim of the paper has been a comparison of one of the parameters describing the quality of the mounting of the implant within the osseous tissue for various techniques of joining the mandrel with the osseous tissue. A list of the same parameters describing the quality of the mounting for each particular technique of mounting allows one to univocally evaluate, which solution is the most effective, and its application for selected individual characteristics of a patient will make the implantology more successful.

2. Methodology of the studiesThe comparative studies of various ways of mounting

the implant mandrel within the osseous tissue was preceded by an overview of the relevant techniques of mounting the mandrel within the osseous tissue. For further studies, one has chosen, among other things, methods that are commonly applied in the implantology and a prototype method based on resorbable materials. The following techniques were selected: – mounting of an implant with a smooth surface of the

mandrel,– mounting of an implant with a porous surface of the

mandrel,– mountingofanimplantwithasuperficiallayermade

of a resorbable material [1,2,4].The choice was determined by the fact that an

analysis of the data obtained during various methods

of experimental studies would not univocally define which solution is superior over the rest. An imprecise description of the parameters that can be found in the related literature could result in an erroneous evaluation of a solution. In order to prevent against such situation, one elaborated his own methodology of experimental studies that was applied for the above selected techniques of mounting the implant mandrel within the osseous tissue [7]. A reliable comparison is possible only when we know how given factors influence a studied parameter, or when all the factors are constant for all the studied objects. The experimental comparative studies presented later in the text were performed according to a methodology, where the factors influencing the studied object were the same for all the tests, and the tests on the studied objects were performed with application of

Fig. 2. Diagram of realization of the recording of the dis-placements of the mounted mandrel with respect to the osseous tissue

Weight Generating Thrust onto the Implant Mandrel Direction of the Thrust Displacement Sensor in Z-Axis

Studied Specimen

Balance with Strain Gauges

Displacement Sensor in X-Axis

Displacement Sensor in Y-Axis

Fig.1. Block diagram of the test station for determining displacements of the mandrel with respect to the osseous tissue, which it was mounted in


Articles34

the same technique, while keeping identical conditions in repeatable studies.

Realization of the studies consisted in a compression test of a studied object that had been prepared before-hand, and recording a displacement of the mandrel mounted within the osseous tissue with respect to the tissue. The measurements were performed using a system being a part of a whole test station (Fig.1).

In order to verify the obtained results, the displace-ments were recorded repeatedly for the same technique of mounting. The tests were repeated for the techniques of mounting the mandrel within the osseous tissue, which had been selected beforehand. The applied load exerted by the weighted mandrel generated a thrust onto the osseous tissue.

The data obtained during the studies were automati-cally recorded in real-time (Fig. 2).

3. Object of the studiesThe object of the studies were three techniques of

mounting the mandrel within the osseous tissue. In order to realize laboratory studies there were prepared objects reproducing the particular techniques of mounting the mandrel within the osseous tissue.

There were made three specimens for each technique. As the object of studies reproducing the mounting

of a smooth mandrel within the osseous tissue, there was applied a sleeve (imitating a fragment of the man-drel) mounted with the closer shaft of a cow thigh bone. A fragment of the closer shaft of the thigh bone originated from a freshly taken cow osseous tissues. The sleeve was made of a high-quality implant steel (cobalt alloy) complying with ISO 5832-12 standard. Mounting of the sleeve within the osseous tissue was realized with the same technique as implantation of the endoprosthesis during a real surgery (Fig. 3).

environment. The cylindrical frame was made out of a cobalt alloy (ISO 5832-12 standard). The material filling the frame was a synthetic osseous substitute by the name of Bio-Oss (Fig. 5)[3, 4].

Fig. 3. Structure of the studied object presenting a ce-ment-less technique of mounting a smooth mandrel with-in the osseous tissue

Sleeve Imitating a Smooth Implant Mandrel Closer Shaft of a Cow Thigh Bone Compressed Osseous Tissue

Fig. 4 Structure of the studied object presenting a ce-ment-less technique of mounting a porous mandrel within the osseous tissue

Porous Sleeve Imitating a Porous Implant Mandrel Closer Shaft of a Cow Thigh Bone

Compressed Osseous Tissue

Fig. 5. Structure of the studied object presenting a tech-nique of mounting a mandrel with a layer of resorbable material

Cylindrical Frame Imitatingthe Implant Mandrel

Closer Shaft of a Cow Thigh Bone

Filling with an Osseous Substitut

The second studied object was a cement-less mounting of a porous mandrel within the osseous tissue (Fig. 4). The cylindrical sleeve had pores of ca. 1 mm created on its surface.

The third studied object was mounting of a cylindrical frame filled with a material resorbable in the osseous

There were made three pieces of each object of studies presented above. The objects, before and after the tests, were stored in temperature of 3-5°C and rela-tive humidity of 90%, in order to inhibit the process of decaying and oxidation of a fresh osseous tissue [4,5].

4. Results of the studiesThe studies carried out allowed one to create a charac-

teristic of the displacements of the mounted mandrel with respect to the bone, which the mandrel was mounted in. In the following tables (Tab. 1-3) there are listed aver-aging values of the displacements of the mounted man-drel with respect to the bone, which the mandrel was mounted in. The tables present results obtained for three studied solutions.


Articles 35

Tab.1. Averaging values of the displacements for the first solution of the mounting, calculated on the basis of three tests

No.Type of the object

of studies

Applied load Resultant displacement

[N] [mm]

1

Obj

ect w

ith a

por

ous

smoo

th s

urfa

ce o

f the

man

drel 0 0

2 30 0.00977

3 60 0.078163

4 90 0.034196

5 120 0.151441

6 200 0.24426

7 300 0.478749

8 400 0.571568

9 500 0.571568

10 600 0.718124

Tab. 2. Averaging values of the displacements for the second solution of the mounting, calculated on the basis of three tests


of studies


[N] [mm]

1

Obj

ect w

ith a

por

ous

surf

ace

of th

e m

andr

el

0 0

2 30 0.058622

3 60 0.058622

4 90 0.058622

5 120 0.097704

6 200 0.381045

7 300 0.644846

8 400 0.688813

9 500 0.688813

10 600 0.688813

Tab. 3. Averaging values of the displacements for the third solution of the mounting, calculated on the basis of three tests


of studies


[N] [mm]

1

Obj

ect w

ith th

e cy

lindr

ical

fram

e an

d os

seou

s su

b-st

itute

0 0

2 30 0.039082

3 60 0.039082

4 90 0.039082

5 120 0.039082

6 200 0.068393

7 300 0.087934

8 400 0.092819

9 500 0.801172

10 600 0.810943

5. Analysis of the obtained resultsThe obtained results allowed one to determine value

of the displacements in the direction of the applied load for various mountings of the mandrel within the osseous tissue with respect to the exerted thrust (Fig. 6). The fol-lowing chart shows a trend of variations of the displace-ments that take place between the bone and the mandrel that is mounted within it. There is also visible a limit of losing a stable mounting of the mandrel within the osseous tissue.

6. DiscussionMounting of the implant within the osseous tissue

operates correctly only when it receives certain supply of mechanical energy as a result of the acting forces, whose magnitude, time and frequency of acting are within appropriate intervals. The performed studies proved that each of the studied solutions presents an approximate character of collaboration between the implant and the osseous tissue. The most advantageous solution is the one that features stability limits at the possibly highest load. As far as the studied types of mounting are con-cerned, the most advantageous behavior was that of the object of studies having the mandrel with a layer of a resorbable material [6].

7. ConclusionsA challenge in the experimental studies where osseous

tissues are used is a variability of the structure of the osseous tissue over the time. In an organism, the osseous tissue, owing to processes of modeling, can progressively adapt to external conditions taking on a form and struc-ture optimal for particular conditions. Therefore, implan-tation disturbs the optimal structure of the osseous tissue, resulting in micro-injuries and changes in the distribution of the stress within the osseous tissue. It disturbs also the equilibrium of the mechanical energy between the tissue structure and a function fulfilled in the motion system [1,4,5,6]. Maintaining variable and advantageous stimuli having a certain dynamics of interactions is possible owing to application of an appropriate implant mandrel and its mounting within the osseous tissue. In the paper, one presented a characteristic courses of the interactions between the implant mandrel and the osseous tissue as a result of the applied load. One compared characteristics of behaviors of particular solutions of mounting the man-drel within the osseous tissue.

Fig. 6. Results of measurements of the displacements of the mounted mandrel with respect to the osseous tissue as dependent on the applied load

LIMIT POINT OF LOOSING STABILITY FOR THE THIRD SOLUTION

DISPLACEMENT [mm]

LOAD [N]


Articles36

AuthORsMarcin Zaczyk* – Division of Design of Precision Devices, Institute of Micromechanics and Photonics, Faculty of Mechatronics, Warsaw University of Technology, ul. Boboli 8, 02-525 Warsaw, Poland:tel. (+48) (22) 2348572, fax. (+48)(22) 2348601.E-mail: [email protected]

Danuta Jasińska-Choromańska – the same Divison. tel. (+48)(22) 2348604E-mail: [email protected]


References[1] Marczynski W., Leczenie zaburzeń zrostu i ubytków

tkanki kostnej, Wydawnictwo Bellona, Warsaw 1995 (in Polish)

[2] Morloock M., Schneider E., Bluhn A., et al., “Duration and frequency of every day activities in total hip patients”, Journal of Biomechanics, vol. 34, 2001, pp. 873-881.

[3] Pietruska M.D., “A comparative study on the use of Bio-Oss® and enamel matrix deriva-tive (Emdogain®) in the treatment of periodontal bone defects”, European Journal of Oral Science, vol. 109, no. 3, 2001, pp. 178-181.

[4] Zaczyk M., Jasińska-Choromańska D., Zjawiskokontaktuendoproteza-kość.In:Materiały II semi-narium inżynierii wytwarzania, Kalisz 2008, Poland, pp. 230-236. (in Polish)

[5] Zaczyk M., Jasińska-Choromańska D., Miśkie-wiczA.,„Stanowiskadobadań jakościosadzeniatrzpieniaimplantuwkości”.In:Materiały konfer-encyjne Majówki Młodych Biomechaników,Ustroń2011, Poland. (in Polish)

[6] ZaczykM.,Jasińska-ChoromańskaD.,KołodziejD.,„Ocenadokładnościwynikóweksperymentalnegobadania połączenia implant-kość”, PAK, vol. 57, no. 9, 2011. (in Polish).


Articles 37

Experimental Apparatus for Sma Actuators Testing

Miroslav Dovica, Tatiana Kelemenová, Michal Kelemen

Submitted 26th June 2011; accepted 3rd September 2011

Abstract: Paper deals with measuring apparatus for shape mem-

ory alloy (SMA) wire actuators. The apparatus has been developed as didactic tool for exercises in mechatronic study program. Apparatus enable contact-less measuring of static and dynamic characteristic of SMA actuators.

Keywords: SMA actuator, measuring, mechatronics

1. Introduction The most famous SMA material is Nitinol, which is

an alloy of nickel and titanium. It has been discovered in Naval Ordance Laboratory in sixty years of twentieth century. Phenomenon of SMA occurs in more then 20 alloy types. The SMA actuators are made as wire, spring or ribbons shape.

Nitinol wire actuator named as FLEXINOL 250LT with 0.25 mm in diameter has been tested in this paper. Recommended activation temperature is 70 °C. Recov-ery force is recommended at value of 9,3 N. Activation electric current is defined at the 1000 mA. Limit maxi-mum stress (pull force) is 172 MPa [3].

In this paper, the contraction and pull recovery test stand has been developed for testing of SMA wire actua-tor [1, 2, 3, 4, 5, 6, 7].

2. Apparatus arrangementApparatus (Fig. 1, 2) consists of a frame with an arm

and two pulleys for guiding of nylon wire connected with SMA wire actuator. One end of the Flexinol wire is attached to the frame and second free end is connected via nylon wire with bias weight. Nylon wire is guided with two pulleys to the place for hook with bias weight. There is a reference point (Fig. 3) from permanent mag-net placed on the nylon wire. Deformation of the Flexinol wire is represented with the reference point (permanent magnet) and position of the magnet is measured via Hall position sensor SS495A. Consequently, output sensor voltage represents the deformation of the Flexinol wire.

It is necessary to do calibration of the hall position sen-sor with length etalons (Fig. 4). The calibration process associates the sensor output voltage with Flexinol wire deformation.

Results from calibration process are shown in Figure 4. Output sensor voltage has nonlinear behavior which corresponds to Flexinol free end position. The measured points are as result of average from 10 measurements. Expected uncertainty of measurement for measurement of position is 0.2 mm. Uncertainties of length etalons have been neglected. Relation between the free end po-sition and output sensor voltage has been fitted via poly-nomial model mentioned in Figure 5.

Thermal activation of the SMA can be easily driven by electrical current from power supply via Joule heating. Cooling of the SMA can be realized via heating radiation into surroundings at the room temperature.Fig. 1. Measuring apparatus arrangement

Fig. 2. Realization of SMA actuator measuring apparatus

Fig. 3. Reference point (magnet) for contact-less measuring of SMA deformation


Articles38

Heating and cooling of the Flexinol causes change of the Flexinol free end position. Fig. 6 shows this dependence of free end Flexinol position on value of electric current, which shows highly nonlinearity and hysteresis behavior. The characteristic shown in Fig. 6 has been measured with bias weight 1 kg, which corresponds to maximum recommended pull stress.

Dynamic characteristic has been tested through the step response testing. Excitation electric current is shown in Fig. 7 and it has been controlled via microcontroller. Step response has been measured via measuring adapter MF624 with personal computer in Matlab/Simulink en-vironment.

The designed test bench also enables to cyclic test life-time of actuator.

It is possible to determine activation and deactivation time for Flexinol which are correspond with heating and cooling time. Figure 7 shows that cooling is slower then heating. Step response is tested for overloaded mode and it is possible to say that overloading causes a decreasing of the heating and cooling time. Students also can iden-tify mathematic model of actuator, which can be inserted into complex model of the mechatronic system.

3. ConclusionThe measuring apparatus is used as didactic tool for

practical exercises in mechatronic subjects. Students can practically try to use SMA wire actuator. They can test various condition of actuator using.

Shape memory alloy has a lot of advantages as clean, silent and spark free operation, high biocompatibility and excellent corrosion resistance. They are also free of parts such as reduction gears and do not produce dust particles. Actuators based on this principle are very often used in robotic and mechatronic application [8, 9, 10, 11, 12].

AcknowledgementsThe authors would like to thank to Slovak Grant

Agency – project VEGA 1/0022/10 „A contribution to the research of measuring strategy in coordinate measur-ing machines”. This contribution is also the result of the project implementation: Centre for research of control of technical, environmental and human risks for permanent development of production and products in mechanical engineering (ITMS:26220120060) supported by the Re-search & Development Operational Programme funded by the ERDF.

Fig. 4. Calibration process of contact-less measuring of SMA deformation

Fig. 5. Calibration process characteristic

Fig. 6. Static characteristic of SMA wire actuator

Fig. 7. Actuator step response testing


Articles 39

AUTHORSMiroslav Dovica* – Technical University of Košice, Da-culty of Mechanical Engineering, Letná 9, 042 00 Košice, Slovak Republic, e-mail: [email protected]. Tatiana Kelemenová – Technical University of Košice, Da-culty of Mechanical Engineering, Letná 9, 042 00 Košice, Slovak Republic, e-mail: [email protected]. Michal Kelemen – Technical University of Košice, Da-culty of Mechanical Engineering, Letná 9, 042 00 Košice, Slovak Republic, e-mail: [email protected].


References[1] Andrianesis K., Koveos Y., Nikolakopoulos G., and

Tzes A., Experi-mental Study of a Shape Memory Alloy Actuation System for a Novel Prosthetic Hand, ed. by: Corneliu Cismasiu, ISBN: 978-953-307-106-0, Publ. InTech, August 2010, pp. 81-105.

[2] E. M. Severinghaus “Low Current Shape Memory Alloy Devices” U.S. Patent US6969920B1, Nov. 29, 2005. Monodotronics.

[3] DYNALLOY Inc., Flexinol Technical Characteristics, [online]. Document Version 2011/01/02 T15:31:00Z 2011 [cit. 2011-01-02]. available online: <http://www.dynalloy.com>.

[4] Shik Hoi-yin, “Metal has memmory??”, [online]. Doc-ument Version 2011/01/02 T17:45:00Z 2011 [cit. 2011-01-02]. available online: http://www.phy.cuhk.edu.hk/phyworld/iq/memory_alloy/memory_alloy_e.html.

[5] P. Drahoš, O. Čičáková, “Measurement of SMA Drive Characteristics”, Measurement Science Review, vol. 3, section 3, 2003, pp. 151-154.

[6] ASTM F 2082 – 01 Standard Test Method for Deter-mination of Transformation Temperature of Nickel-Titanium Shape Memory Alloys by Bend and Free Re-covery.

[7] Janke L., Czaderski C., Motavalli M., Ruth J., “Ap-plications of shape memory alloys in civil engineering structures – Overview, limits and new ideas”, Materials and Structures, vol. 38, June 2005, pp. 578-592.

[8] Vitko A., Jurišica L., Kľúčik M., Duchoň F., “Context Based Intelligent Behaviour of Mechatronic Systems”, Acta Mechanica Slovaca, vol. 12, 2008, part 3-B, pp. 907-916. ISSN 1335-2393.

[9] Vitko A., Jurišica L., Kľúčik M., et al., “Embedding In-telligence Into a Mobile Robot”, AT&P Journal Plus, no. 1: Mobile robotic systems, 2008, pp. 42-44. ISSN 1336-5010.

[10] Olasz A., Szabó T., “Direct and inverse kinematical and dynamical analysis of the Fanuc LR Mate 200IC robot”. In: XXV. International Scientific Conference mi-croCAD, March-April 2011, p. 37-42.

[11] Lénárt J., Jakab E., “Machine vision used in robotics”. In: OGÉT 2010, XVIII. International Conference on Mechanical Engineering, 2010, p. 272-274. (In Hun-garian)

[12] Antal D., “Dynamical modelling of a path controlled vehicle”. In: XXIV. Micro CAD, International Scientific Conference, 2010, pp. 1-6.


Articles40

Simulation of Vehicle Working Conditions

with Hydrostatic Pump and Motor Control Algorithm

Peter Zavadinka, Peter Kriššák

Submitted 26th June 2011; accepted 16thSeptember 2011

Abstract: Emission rules have significant impact on market of

mobile working machines. They bring new challenges for transmission, working, controlling and vehicle manage-ment functions. Extensive knowledge from design, control system, optimization, automation, signal processing, sys-tem modeling etc. are welcome to get answer how best to produce more efficient, productive machines. In case of new concepts is appropriate to create complex dynamic model of machine. Presented paper describes a four wheel drive model and simulation of a mobile working machine and the concept of hydrostatic pump and motor control al-gorithm in selected conditions.

Keywords: hydrostatic pump and motor, control algo-rithm, mobile working machine, energy savings

Subsystems operate with signals and kinematic and power variables (Figure 1). The variables and signals pres-ent input and output data coming to and from individual subsystems. The direction of vehicle movement is defined by the sign of wheel’s speed. The positive value of speed belongs to vehicle forward drive. The negative value of speed belongs to vehicle reverse drive. Analogous to the motoring mode (for example during vehicle acceleration the energy is transferred from engine to wheel) the value of the wheel’s output torque is positive. During braking mode (for example during downhill drive the energy is transferred from wheel to engine) the value of the wheel’s torque is negative. Velocity and force (angular velocity and torque) are important for mobile working machine power train simulation. Flow and pressure are important for hydrostatic transmission analyze.

In the case of 2D model of mobile working machine [15], [16], [17] the power train consists of front and rear axle (wheel), central differential gear (transfer gear), gear-box, hydrostatic transmission and combustion engine (CE). Model is useful for design, testing and optimization of hydrostatic transmission control. The control options of speed gears shift and combustion engine speed are as-sumed during hydrostatic transmission control design.

The duty cycle of mobile working machine must be defined at the beginning of hydrostatic transmission con-

1. Dynamic model of mobileThere are not too many authors who write about dy-

namic models of mobile working machines. Useful works in this area are written by [2], [3], [7], [9], [14], [18].

Power train structure of mobile working machine can be divided to subsystems [11]. This allows in simulation tool (e. g. MATLAB/ Simulink) modular assembling in dependence on the type of power train [15], [16].

Figure 1. Structure of mobile working machine power train with kinematic and power variables consisted of Combustion engine 1, Pump 2, Motor 3 (pump and motor presented hydrostatic transmission), Clutch 4, Gearbox 5, Central differential gear 6, Front differential gear 7, Rear differential gear 8, Front left wheel 9, Front right wheel 10, Rear left wheel 11, Rear right wheel 12, Accelerator (throttle) pedal 13, HST control 14 and Shift control 15


Articles 41

trol design [8]. Every type of mobile working machine (MWM) has a characteristic duty cycle; so universal duty cycle definition is nearly impossible [6]. The duty cycle can be divided to vehicle transportation to workstation – Mode A and vehicle work on workstation – Mode B.

A telescopic loader can be selected for instance with following modes: vehicle transportation to workstation (uphill and downhill driving) and work on workstation (stuff loading and unloading) as it is shown in Figure 2 as Mode A and Mode B [17].

2. Driving modesControl of driving modes is a method how to steer the

power train modes of a mobile working machine. With hydrostatic transmission (HST) it is possible to change driving mode according to driver or safety demands. It means that the driver can change MWM driving mode by pressing correspondent button in dependence on current situation.

A regulation of pump and motor is typical for tele-scopic loaders, when the pump and motor displacement is dependent on the combustion engine speed [15]. The primary regulation (regulation of pump) is active by low-er engine speed. The secondary regulation (regulation of motor) is active by the higher engine speed. The primary and secondary regulation is not active together. While en-gine speed increasing the pump displacement is increas-ing too (primary regulation). When all the options of pri-mary regulation are carried out, the secondary regulation starts to be involved, that it is mean that the displacement of motor is reduced. The hydrostatic transmission control acts similarly by decreasing of the engine speed.

Sophisticated control algorithms developed by tractor producer Fendt (Vario transmission) or John Deere (Au-toPower transmission) are described in [1].

Normal drive mode principle is shown in Figure 3. The speed is controlled by accelerator pedal; the speed gear shift is controlled by joystick by this mode. The swash plate and bent axis angles are linear dependent on the engine speed (as shown in Figure 4). The beginning of secondary regulation is in 1600 min-1, where the maxi-mal engine torque is available. This setting of hydrostatic transmission control allows secure machine operation. There is no focus on operation costs. The working mode can be used in the duty cycle or machine transportation to workstation.

Figure 2. Selected duty cycle with Mode A and Mode B where 1 and 4 – forward drive, 3 and 6 – reverse drive, 2 – loading, 5 – unloading

Figure 3. Normal drive mode with Combustion engine 1, Pump 2, Motor 3, Gearbox 4, Accelerator pedal con-trol 5, Microprocessor 6 and Joystick control 7

Figure 5. Optimal drive mode with Combustion engine 1, Pump 2, Motor 3, Gearbox 4, Microprocessor 6 and Joystick control 7

Figure 4. Normal drive mode, Pump angle αHG and Motor angle αHM vs. Combustion engine speed


Articles42

Optimal drive mode principle is shown in Figure 5. This driving mode allows vehicle control only with joystick. The pump, motor, engine and gearbox are con-trolled in dependence on driver demand. The swash plate and bent axis angles are linear dependent on the engine speed (as shown in Figure 6). The beginning of second-ary regulation is in 1100 min-1. This value was adjusted by simulations. This setting of hydrostatic transmission control allows decrease of fuel consumption. It is opti-mal mode regarding operation costs. The working mode can be used in the duty cycle (the automatic gear shift is blocked) or machine transportation to workstation.

Figure 6. Optimal drive mode, Pump angle αHG and Mo-tor angle αHM vs. Combustion engine speed

3. Control of hydrostatic transmissionsThe controlled pumps and motors must contain hy-

draulic, electric or electro-hydraulic control system. This control system drives swash plate angle, bent axis angle, valves etc. in dependence on the input signals (current, pressure, voltage...). Basic control options of pump and motor are following:

Two-position control.Multi-position control.Proportional control (continuous control).As a pump and motor control were selected electro-

hydraulic proportional controls from Sauer-Danfoss pro-ducer. These controls allow continuous change of pump and motor displacement.

3.1. Pump controlBased on previous demands the Electrical Displace-

ment Control (EDC) was chosen for 24 V variant of ve-hicle [12].

The EDC consists of a pair of proportional solenoids on each side of a three position, four-way porting spool. The proportional solenoid applies a force input to the spool, which ports hydraulic pressure to either side of a double acting servo piston. Differential pressure across the servo piston rotates the swash plate, changing the pump’s displacement from full displacement in one di-rection to full displacement in the opposite direction.

EDC is current driven control (as shown in Figure 7) requiring a Pulse Width Modulated (PWM) signal. PWM allows more precise control of current to the solenoids. The PWM signal causes the solenoid pin to push against the porting spool, which pressurizes one side of the ser-vo piston, while draining the other. Pressure differential across the servo piston moves the swashplate. A swash-plate feedback link, opposing control links, and a linear

spring provide swashplate position force feedback to the solenoid. The control system reaches equilibrium when the position of the swashplate exactly balances the in-put command from the operator (solenoid). As hydraulic pressures in the operating loop changes with load, the control assembly and servo- swashplate system work constantly to maintain the commanded position of the swashplate.

When the control input signal is either lost or re-moved, or if there is a loss of charge pressure, the spring-loaded servo piston will automatically return the pump to the neutral position.

Figure 7. Pump displacement vs. control current

3.2. Motor controlBased on previous demands the Electro-hydraulic

Proportional Control was chosen for 24 V variant of ve-hicle [13].

Displacement can be changed electro-hydraulically (as shown in Figure 8) under load in response to an elec-trical signal from minimum displacement to maximum displacement and vice versa. The displacement changes are proportional to the electrical signal. The electrical signal must be a PWM signal.

Figure 8. Motor displacement vs. control current

Electro-hydraulic Proportional Control is equipped with Pressure Compensator Override (PCOR). The con-trol can be overridden by Pressure Compensator Over-ride using high loop pressure. When the high pressure level equals to Pressure Compensator Override start pressure setting, the PCOR valve is activated. The motor displacement increases. This causes decreasing of system pressure to acceptable value and ensures optimal power utilization throughout the entire displacement range of the motor. Pressure Compensator Override start pressure is adjustable from 110 to 370 bar.


Articles 43

Electro-hydraulic Proportional Control should be equipped with Hydraulic Brake Pressure Defeat system or Electric Brake Pressure Defeat system.

4. Control algorithm model of hydrostatic unitsThe developed hydrostatic transmission control is

based on the previously mentioned driving modes. In the normal driving mode vehicle speed is controlled by accelerator pedal. The demanded engine speed is output from control block (subsystem) to combustion engine. The engine speed feedback is used for displacement con-trol of pump and motor (as shown in Figure 3). A shift of speed gear upwards is realized with forward joystick movement and a shift of speed gear downwards is real-ized with reverse joystick movement.

The optimal drive mode operates like a normal mode but with two differences. First difference is in the way how the vehicle speed (Combustion engine speed) is set. The speed in the optimal mode is controlled by joystick. The swash plate angle and bent axis angle dependence on the engine speed is shown in Figure 5. Second difference is automatic speed gear shift in the optimal drive mode. The automatic speed gear shift is dependent on the en-gine speed analogous to pump and motor displacement. It can be switched off.

Both drive modes are prepared to reversal drive and braking with hydrostatic transmission. During reversal drive the displacement of pump is opposite. The devel-oped model of hydrostatic transmission control is de-scribed in [16].

4.1. Mode A duty cycle simulationA testing of vehicle’s simulation model according to

mode A is shown in Figure 2.The vehicle was tested for various driver demands

and various speeds. The road surface was assumed to be a dirt track. The mobile working machine was unloaded. During the testing of simulation model by normal drive mode a change of speed gear in 4th second was simulated. The optimal drive mode shifted speed gears automati-cally in dependence on combustion engine speed. The demanded engine speed was set up by accelerator pedal (in optimal drive mode by joystick) to demanded values according to Table 1. Figure 9 represents simulation by selected vehicle parameters during Mode A. In this simu-lation the normal drive mode of power train control is ac-tive (vehicle/engine speed is set up by accelerator pedal). Figure 10 represents previous simulation with optimal drive mode of power train control (vehicle/engine speed is set up by joystick).

The Table 1 represents comparison between normal drive mode and optimal drive mode during Mode A simulation. As it is shown here, the drive mode (setting of hydrostatic transmission control) can significantly in-fluence vehicle performance and operating costs. On the basis of intuitive premises and accessible information of telescopic loader’s from manufacturers, obtained results can be considered as realistic. The simulation results must be considered carefully at the beginning of simu-lation. According to Table 1 it can be deduced, that for higher engine speed (vehicle speed) the optimal drive is more suitable in duty cycle – Mode A.

Table 1. Normal and optimal control algorithm comparison

NORMAL DRIVE MODE

Demanded CE speed [min-1]

Fuel consump-tion [l]

Range time [s]

1500 0.1610 280.5

1800 0.1322 150.3

2000 0.1216 98.4

2200 0.1246 61.6

OPTIMAL DRIVE MODE


Fuel consump-tion [l]

Range time [s]

1500 0.1874 398.0

1800 0.1141 110.6

2000 0.1150 77.1

2200 0.1083 51.5

OPTIMAL vs. NORMAL DRIVE MODE


Fuel consumption [%]

Range time [%]

1500 +16.4 +41.9

1800 -13.7 -26.4

2000 -5.4 -21.6

2200 -13.1 -16.4

Figure 9. Mode A simulation, Normal drive mode – field path


Articles44

4.2. Mode B duty cycle simulationThe testing of MWM power train model according to

duty cycle - mode B is shown in Figure 2.The road surface was assumed to be a field road. In-

fluence of cornering drive was neglected. The work func-tion (loading soil with dipper) was simulated as force load (2000 N) acting on the vehicle last two meters of forward vehicle drive by loading the soil. The vehicle load from soil’s weight was simulated as a vehicle’s mass

Table 2. Normal and optimal control algorithm compari-son for tested range - mode

NORMAL DRIVE MODE

Loading resistance [N], Load [kg], Fuel consumption [l]

Range time [s]

2000 (13m-15m), 500 (15m-45m), 0.0356 70

OPTIMAL DRIVE MODE

Loading resistance [N], Load [kg], Fuel consumption [1]

Range time [s]

2000 (13m-15m), 500 (15m-45m), 0.0276 62

OPTIMAL vs NORMAL DRIVE MODE

Loading resistance [N], Load [kg], Fuel consumption [%]

Range time [%]

2000 (13m-15m), 500 (15m-45m), -22.5 -11.4

Figure 10. Mode A simulation, Optimal drive mode – field path

Figure 11. Mode B simulation, Normal drive mode – field path

Figure 12. Mode B simulation, Optimal drive mode – field path


Articles 45

increase (+500 kg) from the moment of loading till un-loading as illustrated in Figure 2. During the simulation the first speed gear was engaged. The demanded com-bustion engine speed was set up by accelerator pedal (in optimal drive mode by joystick). The maximal demanded value of the combustion engine speed was 2000 min-1. The characteristic of demanded combustion engine speed was set up so that the vehicle could drive required dis-tance without using mechanical brakes.

The results of simulation for Mode B and normal drive mode (combustion engine speed is set up by accelerator pedal) are shown in Figure 11. The Figure 12 represents results of the same test but for optimal drive mode (com-bustion engine speed is set up by joystick).

Analogous to previous tests obtained results can be considered as realistic (with respect to used parameters and equations). The test results must be again considered carefully at the beginning of simulation. An influence of the loading and weight increase was minimal. The differ-ences can be detected only by detailed search.

The Table 2 represents comparison between normal drive mode and optimal drive mode during Mode B sim-ulation. Drive mode (setting of hydrostatic transmission control) again significantly influenced simulation results. The loading with enabled optimal drive mode significant-ly decreased fuel consumption and loading cycle time.

5. ConclusionsIn this paper the simulation model of four wheel ve-

hicle power train with control algorithm was presented.Individual subsystems in model: combustion engine,

hydrostatic transmission, gearbox, central differential gear, wheels, vehicle chassis and control were modeled separately by blocks. Modular structure increased speed and efficiency of mobile working machine power train modeling. The individual blocks were created in MAT-LAB-Simulink program. A model of each subsystem (block) was equipped with mask, what increased user’s comfort of parameters setting.

Detailed modeling of subsystem dynamics is difficult and imprecise because of model complexity and difficul-ties by acquiring of some parameters. For example the tire modeling is one of the most difficult fields of ve-hicle’s modeling nowadays. Almost all modeled subsys-tems are non-linear.

The developed model shows strong non-linear behav-ior, which decelerates simulation mainly at the beginning.

The designed linear control algorithm shows relative-ly promising results. But it is still only base conception of the control algorithm. All performed simulations show acceptable results however in some cases (start, gear shift) it is necessary to consider physical and mathemati-cal complexity of model’s structure. During the simula-tion, it is important to know relationships and principles behind the user interface of these blocks; otherwise the simulation results can be misinterpreted by the engineer.

All performed simulations show acceptable results however in some cases (start, gear shift) it is necessary to consider physical and mathematical complexity of model’s structure. During the simulation, it is important to know relationships and principles behind the user in-terface of these blocks, otherwise the simulation results can be misinterpreted.

Developed models of subsystem contain some op-tions, whose using is not confirmed yet. The tendency of this application settings results from need of including the inertia during acceleration.

Despite of this we can assume that the reduction of fuel consumption can be done by suitable modification of pump and motor control characteristics. It can be de-duced, that the hydrostatic transmission setting in the optimal drive is more suitable. For selected application and testing cycles the using of primary regulation is more suitable for lower engine speed and smaller speed range.

In the future, it is necessary to continue with develop-ing of new blocks of drive and control and also with de-bugging of the initial conditions and parameter settings. Many of these were only guessed because of unavail-ability. This work provides good base for further model development of mobile working machine.

AUTHORS

Peter Zavadinka* – Ustav mechaniky teles, mechatron-iky a biomechaniky, Fakulta strojniho inzenyrstvi, VUT v Brne, Technicka 2896/2, 616 69 Brno, Czech Republic, [email protected]

Peter Krissak* – Technical centre, Sauer-Danfoss a.s., Kukucinova 2148-84, 017 01 Povazska Bystrica, Slovak Republic, [email protected]

* Corresponding author

References

[1] Bauer F., Sedlak P., & Smerda, T., Traktory. Praha: Prori Press, 2006.

[2] Carlsson E., Modeling Hydrostatic Transmission in Forest Vehicle. M.S. Linkopings Universitet, 2006.

[3] Carter D., Load modeling and emulation for an earthmoving vehicle power train. M.S. University of Illinois at Urbana-Champaign, 2003.

[4] Carter D. E., Alleyne A. G., “Earthmoving vehicle power train controller design and evaluation”. In: Proceedings of the American Control Conference, 30thJune 2004- 2nd July 2004, pp. 4455-4460.

[5] Grepl R., “Real-Time Control Prototyping in MAT-LAB/Simulink: Review of tools for research and education in mechatronics”. In: IEEE International Conference on Mechatronics (ICM 2011), 13th-15th April 2011, Istanbul, Turkey.

[6] Grepl R., Vejlupek J., Lambersky V., Jasansky M., Vadlejch F., Coupek P., “Development of 4WS/4WD Experimental Vehicle: platform for research and education in mechatronics”. In: IEEE International Conference on Mechatronics (ICM 2011), 13th-15th April 2011, Istanbul, Turkey.

[7] Kiencke U., Nielsen L.,: Automotive Control Sys-tems: For Engine, Driveline, and Vehicle. 2nd ed., Springer-Verlag: Germany, 2005.

[8] Kriššák P., Kučík P., “Computer aided measure-ment of hydrostatic transmission characteristics”,


Articles46

Hydraulika i Pneumatyka, 4/2005, p. 22-25, INDEX 37726, ISSN 1505-3954.

[9] Prasetiawan E., Modeling, simulation and control of an earthmoving vehicle power train simulator. M.S. University of Illinois at Urbana-Champaign, 2001.

[10] Rill G., Vehicle dynamics: Lecture notes. [Online] Regensburg: University of applied sciences, 2007, Available at: http:// homepages.fh-regensburg.de/~rig39165/skripte/ Vehicle_ Dynamics.pdf [Accessed 27 October 2008].

[11] Rill G., “Vehicle modeling by subsystems”, Jour-nal of the Brazilian Society of Mechanical Sciences and Engineering, [Online]. vol. 28, no. 4, 2006, p. 430-442. Available at: http://www.scielo.br/scielo.php?pid=S1678-58782006000400007&script=sci_arttext&tlng=en [Accessed 9 January 2010].

[12] Sauer-Danfoss: H1 Axial Piston Pumps 045/053 Single, 045/053 Tandem, 018 Single, 115/130 Sin-gle, 147/165 Single, 2009. [Online] Sauer-Danfoss. Available at: http://www.sauer-danfoss.com/stel-lent/groups/publications/ documents/product_lit-erature/ 11009999.pdf [Accessed 27 March 2009].

[13] Sauer-Danfoss: Series 51, Series 51-1, Bent Axis Variable Displacement Motors, Technical Infor-mation, 2003. [Online] Sauer-Danfoss. Avail-able at: http://www.sauer-danfoss.com/stellent/groups/ publications/documents/product_litera-ture/52010440.pdf [Accessed 15 February 2009].

[14] Tinker M., Wheel loader power train modeling for real-time vehicle. M.S. University of Iowa, 2006.

[15] Zavadinka P., Krissak P., Modeling and simulation of diesel engine for mobile working machine power train. In: Polish Society of Mechanical Engineers and Technicians - SIMP. Hydraulics and Pneumat-ics 2009: Domestic branch and turbulent global market. Wrocław, Poland 7-9 October 2009. SIMP: Wroclaw, Poland, 2009, ISBN 978-83-87982-34-8, pp. 339-348.

[16] Zavadinka P., Modeling and Simulation of Mobile Working Machine Power train. M.S. Brno Univer-sity of Technology, 2009.

[17] Zavadinka P., Kriššák P., Modeling and simulation of mobile working machine power train. In: Tech-nical Computing Prague 2009, 19.11.2009, Praha, Česká republika, 2009, p. 118, ISBN 978-80-7080-733-0.

[18] Zhang, R.: Multivariable robust control of non-linear systems with application to an electro- -hydraulic power train. Ph.D. University of Illinois at Urbana-Champaign, 2002.

[19] Zhang R., Alleyne, A. G. Carter D. E., “Robust gain scheduling control of an earthmoving vehicle power train”. In: Proceedings of the American Con-trol Conference, 2003, vol. 6, no. 4-6, June 2003, pp. 4969- 4974.


Articles 47

The Branch & Bound Algorithm Improvement in Divisible Load Scheduling with Result Collection on

Heterogeneous Systems by New Heuristic Function

Farzad Norouzi Fard, Sasan Mohammadi, Peyman Parvizi


Abstract: In this paper we propose a new heuristic function for

branch & bound algorithm. By this function we can in-crease the efficiency of branch & bound algorithm. Di-visible loads represent computations which can be arbi-trarily divided into parts and performed independently parallel. The scheduling problem consists in distribut-ing the load in a heterogeneous system taking into ac-count communication and computation times, so that the whole processing time is as short as possible. Since our scheduling problem is computationally hard, we propose a branch & bound algorithm. By simulating and compar-ing results it is observed which this result produces bet-ter answers than other methods, it means that branch & bound algorithm have less total average of relative error percentage in the variety Heuristic functions.

Keywords: divisible load scheduling, heterogeneous systems, branch & bound algorithm

1. IntroductionDivisible loads form a special class of parallelizable

applications, which if given a large enough volume, can be arbitrarily partitioned into any number of indepen-dently and identically processable load fractions. Divis-ible load theory (DLT) is the mathematical framework that has been established to study divisible load sched-uling (DLS) [1, 2]. The problem of working scheduling heterogeneous system has specific importance because of the necessity of optimize using calculating processors and also spending less time for performing of scheduling algorithms. In this paper we study divisible load sched-uling with result collection on heterogeneous which has star network. In a star connected network where the cen-ter of the star acts as the master and holds the entire load to be distributed, and the points of the star form the set of slave processors, the basic principle of DLT to deter-mine an optimal schedule is the AFS (All nodes Finish Simultaneously) policy [3]. In heterogeneous system, processors Efficiency, communication network topology and speed of network lines can be different. Scheduling works in heterogeneous system is computationally hard. One of the computation models is divisible load. Divis-ible load model originated in the late 1980s [4, 5]. Sur-veys of divisible load theory (DLT), including applica-tions, can be found in [1, 6]. DLT proved to be a valuable tool for modeling processing of big volumes of data [7, 8] includes image processing [9], signal processing, data mining and research in Database [10]; calculate linear algebra [11] and multimedia functions [12]. Distribut-

ing the load causes inevitable communication delays. To shorten them, the load may be sent to processors in small chunks rather than in one long message. This way the computations start earlier. Such multi-installment or multi-round divisible load processing was proposed first in [13]. Memory limitations for single-installment com-munications were studied in [14], where a fast heuristic has been proposed. In [15] it was shown that this problem is NP-hard if a fixed startup time is required for initiation of communications. In this theory we use master-slave model. The load located on master. Master computer divides divisible load between slaves, when slave com-puters received all load, start processing. Each slave computers after finishing of processing report the result to master. The problem consists in finding a communica-tion sequence, the schedule of communications from the originator to the workers, and sizes of transferred load pieces, so the total responding time becomes minimum. In previous researches amount of slave results hypoth-esized low so that we ignore time delay for sending this data to master; but nowadays, researchers hypothesizing time delay for returning back slave results to master com-puter. If M is number of computer, to consider different arrangements, time complexity is O〖(m!)〗^2. it has not already represented a certain algorithm with poly-nomi-nal time complexity that can produce answer less time in all cases but existent creative ways are LifoC, FifoC [16, 17], ITERLP [18], Sport [19, 20], GA [21], and Branch & Bound LifoC. Our aim is to suggest Branch and bound algorithm for solving divisible load scheduling with result collection on heterogeneous systems. The rest of this paper is organized as follows. In section 2 the prob-lem is formulated. Section 3 describes Branch and bound Copt algorithm for solving DLS problem. The results presented in section 4. The last section is dedicated to conclusions.

2. System model and problem definition The network model to be considered here consists of

(M + 1) processors interconnected through M links in a single-level tree fashion as shown in Fig. 1.

In this paper we assume star interconnection. A set of working processors is connected to a central server called master. A processor is a unit comprising a CPU, memory and a hardware network interface. The CPU and network interface can work in parallel so that simultaneous communication and computation is possi-ble. { } Is the set of computation parameters of the slave computers, and { } is the set of communication parameters of the network links. Is the reciprocal of the speed of processor pk, and Ck is the


Articles48

reciprocal of the bandwidth of link lk. In this model, L is the whole dividable load that exists in master computer. Since it does not damage problem, we suppose that L=1. The source p_0 splits L into parts and sends them to the respective processors for computation. Each such set of m parts known as a load distribution

. All processors follow a single-port and no-overlap communication model, implying that processors can communication with only one other pro-cessor at the time, and communication and computation cannot occur simultaneously. If the allocated load frac-tion is , then the returned result is equal to , where 0≤δ≤1. The constant δ is application specific, and is the same for all processors for a particular load L. for a load part , is the transmission time from p0 to pk, akCk, is the time it takes to perform the requisite process-ing on ak, and dakCk is the time it takes pk to transmit the results back to p0. sa and sc are two permutation of order m that represent the allocation and collection sequences respectively sa [k] and sc [k] denote the processor number that occurs at index . sa [k] and sc [k] are two lookup functions that return the index of the proces-sor k in the allocation and collection sequences. Purpose of scheduling is to find the sequence pair (sa , sc), and a[1...k] that minimize total processing time. The total pro-cessing time is started from the time of load distribution until receiving the last process from master processors. Result collection phase begins only after the entire load fraction has been processed, and is ready for transmission back to the source. This is known as a block based system model, since each phase forms a block on the time line Fig. 2.

As sa and sc are determined, we can find a[1...k] with linear programming as below:

In the above formulation, for a pair (sa, sc), (1) imposes the no-overlap constraint. The single- port com-

munication model is enforced by (2). The fact that the entire load is distributed among the processors is ensured by (3). This is known as the normalization equation. The non-negativity of the decision variables is ensured by constraint (4) [22]. By using branch and bound algorithm to find sa[1...m], sc[1...m] and a[1...k]. There is (m!) Possible permutations each of sa and sc, and the linear program has to be evaluated (m!)2 times to determine the globally optimal solution.

3. Branch and bound algorithm for solving DLS problem Branch and bound algorithm is one of the trees and

graphs traversal and exploring methods. Branch and bound algorithm is performed like below:

• Tree traversal,• heuristic function,• pruning branches.At the beginning the root node is selected, once the

root is selected its children will be created. After that heuristic function will work on all children and compare their answers. Then it will select the child who had the best result and it repeats this action until the result is found. We probably can find many answers for DLT, bout Branch and bound algorithm ended when the first answer is found. Branch and bound algorithm Travers tree as BFS and use heuristic functions for pruning branches. In Fig. 3 we display how to extend nodes.

Fig. 3. Extending node in branch and bound

In our tendered algorithm (Branch & Bound Copt), first the selected processor and its father are located in allocations list, then total slaves are located in allocation list by the best C with E between them, after that we call heuristic function with this data.

4. Computational experiments In experiments, we compared efficiency of Branch &

Bound Copt algorithm by Branch & Bound LifoC, Sport, LifoC and Genetic Algorithms. We performed our Tests

Fig. 1. A heterogeneous star network

Fig. 2. Schedule for M=3


Articles 49

by Amd Athelon Dual 3.0 Ghz with 2 Gigabyte RAM in Matlab environment. To display a heterogeneous system we consider 25 different cases of C and E. For every 25 cases, m value of C and E produced randomly. In all tests, we calculated time of process for each algorithm. If Topt shows us the time of process for optimal algorithm and Tv shows the time of process for other algorithms, the percentage of relative error (DTv ) was calculated as for-mulation (5).

Since we produce 25 different cases of heterogeneous system, the average of relative error percentage is calcu-lating as formulation (6).

In order to consider the effects of & parameter in mention algorithm, the result time obtains experiments which have been done for M=4,5 and δ = 0.1,0.2,...1,

and the average of relative errors percentage has been shown in Fig (4, 5). In these figs, we see average error percentage of Genetic algorithm, Sport, LifoC, Branch and Bound LifoC and Branch & Bound Copt for 4 and 5 slave computers.

As displayed in Fig. 4, when we have 4 slaves com-puter, Branch & Bound Copt algorithm in much δ value has lowest average of relative error percentage. Consid-ering the running time being less in Branch and Bound algorithm, we can introduced it as the best algorithm. With respect to the efficiency of Branch & Bound Copt algorithm, Branch & Bound LifoC algorithm and Genetic algorithm rather than the other two, we compare them in Fig. 5.

For m = 5 and δ = 0.7, The Run time& average of relative errors percentage for all of algorithm has been shown in Table 1.

5. ConclusionIn this paper, a new heuristic algorithm, Branch and

Bound Copt, for the scheduling of divisible loads on het-erogeneous systems and considering the Result collec-tion phase is presented. A large number of simulations are performed and it is found that Branch and Bound Copt consistently delivers near optimal performance.

As future work, an algorithm with similar performance, but with better cost characteristics than Branch and Bound Copt needs to be found. Another important area would be to extend the results to multi-level processor trees.

AUTHORS

Farzad Norouzi Fard* – Mechatronic Department Is-lamic Azad University – South Tehran Branch,Tehran, Iran, [email protected]

Sasan mohammadi – Mechatronic Department Islamic Azad University – South Tehran Branch,Tehran, Iran, [email protected]

Peyman Parvizi – Mechatronic Department Islamic Azad University – South Tehran Branch,Tehran, Iran, [email protected]. 5. Average of relative error percent for m = 4,5

Table. 1. Run time & average of relative error percentage for m=5 & δ=0.7

Algorithm Run time Average of relative error percentage

Optimal algorithm 182.6719 0

Branch & Bound Copt algorithm

0.2125 0.000283476

Branch & Bound LifoC algorithm

0.2 0.000299117

Genetic algorithm 30.5712 0.000637334

LifoC algorithm 0.0125 0.0039602808

FifoC algorithm 0.015 0.074704891

Sport algorithm 0.0025 0.183.05

Fig. 4. Average of relative error percent for m=4,5


Articles50

References

[1] Bharadwaj V., Ghose D., Mani V., Robertazzi T.G., Scheduling Divisible Loads in Parallel And Dis-tributed Systems, IEEE Computer Society Press, Los Alamitos CA, 1996.

[2] Lee C.H., Shin K.G., “Optimal task assignment in homogeneous networks”, IEEE Trans. Parallel Distrib. Syst., vol. 8, no. 2, Feb. 1997, pp.119-129.

[3] Barlas G.D., “Collection-aware optimum sequenc-ing of operations and closed-form solutions for the distribution of divisible load on arbitrary processor trees”, IEEE Trans. Parallel Distrib. Syst., vol. 9, no. 5, May 1998, pp.429-441.

[4] Agrawal R., Jagadish H.V., “Partitioning Tech-niques for Large-Grained Parallelism”, IEEE Transactions on Computers, vol. 37, no. 12, 1988, pp. 1627-1634.

[5] Cheng Y.-C., Robertazzi T.G., “Distributed compu-tation with communication delay”, IEEE Transac-tions on Aerospace and Electronic Systems, vol. 24, 1988, pp. 700-712.

[6] Robertazzi T.G., “Ten reasons to use divisible load theory”, IEEE Computer, vol. 36, 2003, pp. 63–68.

[7] Bharadwaj V., Ghose, D., Robertazzi, T. G., “Di-visible Load Theory: A New Paradigm for Load Scheduling in Distributed Systems”, Cluster Com-puting, vol. 6, no. 1, Jan. 2003, pp. 7-17.

[8] Jingxi J., Bharadwaj V., Ghose D., “Adaptive Load Distribution Strategies for Divisible Load Process-ing on Resource Unaware Multilevel Tree Net-works”, IEEE Transactions on Computers, vol. 65, no. 7, 2007, pp. 99-1005.

[9] Li X., Bharadwaj V., KO C. C., “Distributed Image Processing on a Network of Workstations”, Inter-national J. Computers and Applications, vol. 25, no. 2, 2003, pp. 1-10.

[10] Blazewicz J., Drozdowski M., Markiewicz M., “Divisible Task Scheduling: Concept and Verifica-tion”, Parallel Computing, vol. 25, 1990, pp. 87-98.

[11] Chan S., Bharadwaj V., Ghose D., “Large Matrix-Vector Products on Distributed Bus Networks with Communication Delays Using the Divisible Load Paradigm: Performance and Simulation”, Math. And Computers in Simulation, vol. 58, 2001, pp.71-92.

[12] Altilar D., Paker Y., “Optimal Scheduling Algo-rithms for Communication Constrained Parallel Processing”. In: Proc. 8th Int’l Euro-Par Conf., 2002, pp. 197-206.

[13] Bharadwaj V., Ghose D., Mani V., “Multi-install-ment Load Distribution in Tree Networks with De-lays”, IEEE Transactions on Aerospace and Elec-tronic Systems, vol. 31, 1995, pp. 555-567.

[14] Li X., Bharadwaj V., Ko C. C., “Processing divis-ible loads on single-level tree networks with buffer constraints”, IEEE Transactions on Aerospace and Electronic Systems, vol. 36, 2000, pp. 1298-1308.

[15] Drozdowski M., Wolniewicz P., “Optimum divis-ible load scheduling on heterogeneous stars with limited memory”, European Journal of Operation-al Research, vol. 172, 2006, pp. 545-559.

[16] Rosenberg A. L., “Sharing Partitionable Workloads in Heterogeneous NOWs: Greedier Is not Better”, IEEE International Conf. on Cluster Computing, Newport Beach, CA, Oct. 2001, pp. 124-131.

[17] Beaumont O., Marchal L., Rehn V., Robert Y., “FIFO Scheduling of Divisible Loads with Return Messages Under the One Port Model”. In: Proc. Heterogeneous Computing Workshop HCW’06, April 2006.

[18] Ghatpande A., Nakazato, H., Watanabe, H., Beau-mont, O., “Divisible Load Scheduling with Result Collection on Heterogeneous Systems”. In: Proc. Heterogeneous Computing Workshop (HCP 2008), April 2008.

[19] Ghatpande A., Nakazato H., Beaumont O., Wata-nabe H., “SPORT: An Algorithm for Divisible Load Scheduling With Result Collection on Heteroge-neous Systems”, IEICE Transactions on Communi-cations, vol. E91-B, no. 8, August 2008.

[20] Ghatpande A., Nakazato H., Beaumont O., Wata-nabe H., “Analysis of Divisible Load Scheduling with Result Collection on Heterogeneous Sys-tems”, IEICE Transactions on Communications, vol. E91-B, no. 7, July 2008.

[21] Suresh S., Mani V., Omkar S. N., Kim H. J., “Divis-ible Load Scheduling in Distributed Systems with Buffer Constraints: Genetic Algorithm and Linear Programming Approach”, International Journal of Parallel, Emergent and Distributed Systems, vol. 21, no. 5, Oct. 2006, pp. 303-321.

[22] Vanderbei R. J., Linear Programming: Foundations and Extensions, 2nd Ed., International Series in Op-erations Research & Management, vol. 37, Kluwer Academic Publishers, 2001.


Articles 51

Surface Characterization by Accurate Measurement and Image Processing Systems on Machined Surfaces of

Precision Cutting Tools

Numan M. Durakbasa, Ismail Bogrekci, Pınar Demircioglu, Gökcen Bas, Aslı Gunay



Articles26

Surface ToPograPhical Investigations By Accurate Measurements And Image Processing TechniQues

Numan M. Durakbasa, Ismail Bogrekci, Pınar Demircioglu, Gökcen Bas, Aslı Gunay


Abstract: High precise measurement techniques and surface

structure analysis are required in advanced fields of in-terchangeable manufacturing and precision engineering. This study presents the characterization of the surface roughness of the machined milling cutters by exper-imental precision measurements and the image pro-cessing tool. The data obtained are compared to assess the surface characterization parameters and computa-tional data in terms of precision, accuracy, sensitivity, repeatability and resolution.

In the experimental measurement phase, the roughness measurements and surface topography characterization were performed in the nanotechnology laboratory using the stylus profilometry and digital microscopy. The compu-tational phase was performed using an image processing toolbox with precise evaluation of the roughness for the machined metal surfaces of the end mill cutting tool. The surface parameter database is established ex-hibiting an advantage over the traditional method. This study reveals a comparison methodology of the end mill surface parameters using both stylus readings and im-age processing software for widely used end mill cutting tools that have considerable effect on characterization of sensitive manufacturing surface of millings.

Keywords: surface roughness, end milling, image processing, precision machining

1. IntroductionThe experimental precision measurements are one

of the key methods for characterization of the rough-ness conducted by nanometrology devices to serve the manufacturing industry. However, the characterization of the surface of the machined metal milling is often chal-lenging due to its complex surface structure, its effect on the end-product and predefined limits in accordance with the standards.

Surface roughness is a measure of the texture of a sur-face. Roughness is evaluated by means of some param-eters: Ra, the area between the roughness profile and its central line, or the integral of the absolute value of the roughness profile height over the sampling length, and Rz, average maximum height of roughness profile are the most preferred parameters measured commonly by a stylus profilometer [1] in surface roughness evaluation, as the most of the surface finish standards in the world are written for these profilometers. As a complemen-tary of the stylus method, the digital microscopy makes

a notable contribution to the development of the field of dimensional measurement. The other important property of the digital microscopes is to monitor high quality re-corded images easily, since the optical image is projected directly on the charge-coupled device (CCD) in a digital camera.

This study focuses on establishing a methodology for the surface roughness characterization by managing an evaluation process after comparison of both experi-mental measurements utilizing the stylus method, the digital microscopy and also image process technique implementing an image recognition algorithm. The sam-ples used for this study were chosen according to the fac-tors of the shape, size, material, process parameters, the stiffness, macro-geometry, coating specifications deter-mining the mechanical properties of the machined metal millings. The uncoated and coated machined cutting mill samples were observed, measured and their images were processed to serve a solution to the problems of two basic complications in manufacturing tools; operational surface abuse and life time. Roughness of the cutting mills’ surface is one of the most important considerations affecting the desired surface quality and the functional behavior of the part moreover preventing these compli-cations. The aim of this application was to see the differ-ences between the end mill cutting tools with and without coatings, when measuring roughness of the end mill sur-faces, because such surfaces are being measured after a series of machining processes in general.

The evaluation process flow as represented in below Fig. 1 consists of two main phases that reveals the de-tailed scientific, collaborative research process. The two main phases, namely the measurement phase and the computational phase were managed by using the profes-sional process management software toolbox (iGrafx). The process management contributed to the research study exhibiting an advantage over traditional methods through real-time process control, analysis, reporting and standardization.

2. MaterialThe coating technology called “Physical Vapor De-

position (PVD)” is conducted by a process of collecting the cathode AlTiN and an additional element’s atomized and afterwards vaporized materials. This process takes place at 500 ºC and then the cutting mill is covered under vacuum. The coating process is performed by many layers of nano scale coatings with an average thickness of 1 – 3 µm.

The uncoated and coated cutting mill samples have the similar surface structure except the coating layer observed


Articles52


Articles 27

in the colour of Anthracite. However the precise measure-ments to obtain surface roughness parameters have in-dicated that the coating process caused the same surface of the cutting mill a higher density of material with an ir-regular geometry. The irregularities caused by the coating process were observed using nanometrology methods to serve the already stated two basic problems in manufac-turing tools; operational surface abuse and life time.

In this study, two samples from the machined cut-ting mills with the same geometrical characteristics were analyzed in order to assess surface roughness of both the uncoated and coated mills shown in Fig. 2.

Fig. 2. The coated (a) and uncoated (b) cutting end mill samples

Fig. 1. The evaluation process flow for characterization of surface roughness

EvaluationProcess

The Measurement

Phase

(Nanometrology

Laboratory)

The Computational Phase

(Image Processing)

Start

Cutting MillSample Choice

- Uncoated-AlTiN+Adds1

Contact Stylus Type

Profilometer

StartImage captured

by the Digital Microscopy

ComputerProgram

Data

DigitalMicroscopy

Image

Processing

Technique -Line Scanning-Fast Fourier Transform

Display and

Record

RoughnessParameters

&Surface

Topography

Surface RoughnessCharacterization

Comparison End

3. The Experimental Measurement PhaseTwo high-precision cutting tools having cylindrical

handlings of different textures, with and without coating were investigated by means of the evaluation of the roughness measurements as well as the analysis of the images captured from high-resolution digital microscope with the help of image processing techniques called Line Scanning and Fast Fourier Transform (FFT). The analyses of the surfaces of the cutting tools with different geometry, material as well as the surfaces of uncoated cutting tool with the same geometry and material and its coated counterpart were examined.

3.1. Contact Stylus Type ProfilometerThe contact roughness measurement of the machined

metal surfaces was performed by the Form Talysurf Intra 50 profilograph [2] with μltra software (FTS Iμ) illus-trated in Fig. 3 according to the ISO 4287 [3] by mapping the readings taken in a direction perpendicular to the direction of lay by calculating of the parameters Ra and Rz from a standard spectrum of roughness. Table 1 represents the specifications of the contact stylus profilometer.

Table 1. The specifications of contact stylus type pro-filometer

Measurement Method

Spatial resolution Z Resolution Range Z

Stylus Profilometer (SP) 1-2 μm 3 -16 nm 0.2 - 1

mm

(a) (b)


Articles 53


Articles28

Fig. 3. Schematic diagram illustrating Form Talysurf In-tra 50 profilograph during the contact measurement of end mill sample used in this study

3.2. The Digital MicroscopyThe Keyence VHX-1000 digital microscope illustrated

in Fig. 4 is a high resolution CCD camera based system with a high intensity halogen lamp and image processing capabilities that integrates observation, recording, and measurement functions [4]. Table 2 indicates the specifi-cations of this digital microscope in detail.

4. The Computational PhaseThe captured images were processed and analysed

using Matlab image processing toolbox. The techniques carried out in this study were Line scanning and Fast Fourier Transform (FFT).

4.1. Line ScanningIn this study, the true color images were binarised.

The size of images was 1000 by 552. DN (Digital Num-ber) values of binarised images were collected from the selected lines using line scanning technique used in Fig. 5 [5]. These selected lines were taken from the points at 300, 600 and 900 composed of 552 pixels on y axis.

x=300 x=600 x=900

y=552

4.2. Fast Fourier Transform (FFT) The true color images were reduced to 8 bit grey level

images and grey images were sharpened by 20 pixels of radius and Gaussian blurry. Processed images were con-verted from spatial domain to frequency domain using Matlab Image Processing Toolbox (2D FFT).

5. Results5.1. Measurement Results

The roughness measurements were carried out with Taylor Hobson Form Talysurf Intra 50 profilograph. The roughness data taken from the stylus profilometer were processed in TalySurf Intra software. In the measurements of the stylus profilometer, 60 mm stylus arm length, 2 µm radius conisphere diamond stylus tip size and 1 mN force (speed=1 mm/s) were selected [6,7,8,3]. The complicated digital based systems provide great image resolution as indicated in Table 2. The evaluation processes have been carried out in a 0.2µm interval by a software (Keyence

VHX1000) developed for imaging pur-poses. This investigation was performed with a 20x objectives, as 20x objective has smaller deviations in the measure-ment results. Actually, the instrument has from 20x to 500x and 500x to 5000x objectives. This property enables the in-strument to get different views of surface structure. For the coated and uncoated end mills, a standard high-pass Gaussian filter with a long-wavelength cutoff of 0.8 mm was used and sampling length as 1.6 mm according to the ISO standards were chosen [3]. The analyses were then performed for all specimens for several roughness parameters but Ra parameter gave us much of the idea on the surface topography.

Table 2. The specifications of the digital microscopy [4]

Model VH-Z20R VH-Z500R

Magnification 20x 30x 100x 200x 500x 1000x

Field of view (mm)

Horizontal 15,24 10,16 3,05 1,52 610 305

Vertical 11,40 7,60 2,28 1,14 457 229

Diagonal 19,05 12,70 3,81 1,91 762 381

Depth of field (mm)2

34 15,5 1,6 0,44 - -

Evaluation distance (mm) 25,5 4.5

Fig. 4. Schematic diagram illustrating the Keyence VHX-1000 Digital Microscopy

y=552

y=1 x=300 x=600 x=900

Fig. 5. Line Scanning Representation [5]


Articles54


Articles 29

As observed in Fig. 6, the results for the end milling cutting tool samples represent smooth surface topography in terms of Ra values. Ra values are around 293 nm for uncoated end mill sample and 362 nm for coated end mill sample. The aim of the study was to investigate the surfaces in terms of coating characteristics by measuring their roughness after their manufacturing processes.

5.2. Image Processing Results In this study, the images taken fromuncoated (Fig.7a) and coated (Fig.7b)endmills were analysed using linescanning and 2D FFT image processingtechniques. The measurement results fromthe stylus profilometer were compared with those from the image procesiing techniques.

(a)

(b)Fig. 7. The images taken from a)

uncoated and b) coated end milling cutting tools

5.2.1. Line Scanning The application of line scanning im-age processing method were carried out with the points at 300,600 and

Fig. 6. Ra values belonging to the coated and uncoated end mill cutting tool

Fig. 8. The number of pulses taken from the end mill samples with and without coated

End Mill Sample with Coated

End Mill Sample withoutCoated

umber of pulses

x=300 147 158156 153138 168

Mean Values 147 160

Std. Dev. 9 8

Fig. 9. The signal taken from the image of uncoated end milling cutting tool

Fig. 10. The signal taken from the image of coated end milling cutting tool

900 composed of 552 pixels on y axis. The results of line scanning as an image processing technique from Figs. 9 and 10 indicate that pulse numbers and pulse widths are different for each surface. The number of pulses is shown in Fig. 8 respectively for coated and uncoated surfaces of the end mill samples taken from different regions.

The standard deviations of the analysed lines are 9 and 8 respectively for the coated and uncoated cutting end mill samples. The features of pulses taken from all regions represented consistency.

Samples

Number ofPulses

x=300x=600

x=900

Mean ValuesStd. Dev.

End Millwith Coated

End Millwithout Coated

147

156

138

158

153

163

147 160

9 8


Articles 55


Articles 29

As observed in Fig. 6, the results for the end milling cutting tool samples represent smooth surface topography in terms of Ra values. Ra values are around 293 nm for uncoated end mill sample and 362 nm for coated end mill sample. The aim of the study was to investigate the surfaces in terms of coating characteristics by measuring their roughness after their manufacturing processes.

5.2. Image Processing Results In this study, the images taken fromuncoated (Fig.7a) and coated (Fig.7b)endmills were analysed using linescanning and 2D FFT image processingtechniques. The measurement results fromthe stylus profilometer were compared with those from the image procesiing techniques.

(a)

(b)Fig. 7. The images taken from a)

uncoated and b) coated end milling cutting tools

5.2.1. Line Scanning The application of line scanning im-age processing method were carried out with the points at 300,600 and

Fig. 6. Ra values belonging to the coated and uncoated end mill cutting tool

Fig. 8. The number of pulses taken from the end mill samples with and without coated

End Mill Sample with Coated

End Mill Sample withoutCoated

umber of pulses

x=300 147 158156 153138 168

Mean Values 147 160

Std. Dev. 9 8

Fig. 9. The signal taken from the image of uncoated end milling cutting tool

Fig. 10. The signal taken from the image of coated end milling cutting tool

900 composed of 552 pixels on y axis. The results of line scanning as an image processing technique from Figs. 9 and 10 indicate that pulse numbers and pulse widths are different for each surface. The number of pulses is shown in Fig. 8 respectively for coated and uncoated surfaces of the end mill samples taken from different regions.

The standard deviations of the analysed lines are 9 and 8 respectively for the coated and uncoated cutting end mill samples. The features of pulses taken from all regions represented consistency.

Samples

Number ofPulses

x=300x=600

x=900

Mean ValuesStd. Dev.

End Millwith Coated

End Millwithout Coated

147

156

138

158

153

163

147 160

9 8


Articles30

As observed in signal valleys, the width of pulse increases with surface roughness. The number of black pixels is higher in rough surfaces. This confirms that pulse width at valleys increase with increase in surface roughness.

5.2.2. Fast Fourier Transform (FFT) FFT analyses results are shown in Fig. 11 for un-coated and coated tools.

(a)

(b)

Fig. 11. The FFT images taken from a) uncoated and b) coated end milling cutting tool

Although Ra values from stylus profilometer for uncoated and coated cutting tools are not so distinguishable (Ra = 0,293 and Ra = 0,362, respectively), FFT images shows the difference. The diameter of white blob in the uncoated FFT image was found higher than that of coated FFT image.

6. Conclusions and Future WorkThis study was performed by the researchers of dif-

ferent academic organizations through a process management toolbox delivering process improvement and commu-nication enhancement. The methodology to define the surface roughness characterization of the end milling surface parameters using both the stylus readings and an image processing technique exhibited an advantage for more precise and accurate results.

The computer vision algorithm developed in the methodology presented a considerable potential in the determination of the surface roughness parameters that performed as a noncontact measurement technique in nanometrology. The evaluation process of the surface roughness parameters was carried out by using the image processing technique called Line Scanning and 2D Fast Fourier indicated results supporting detailed results.

The continuation of this study is to be the surface in-vestigation of the cutting tools after a series of process and their coating performances and issues such as the op-timization of the surfaces of the tools. The results will be evaluated for comparison of the surface roughness char-acterization, coating material effect and instrument life time as indicated in the process flow.

The future work of image processing is comprised of developing high quality image processing techniques to evaluate the surface roughness parameters of different imaging systems and techniques of high resolution, high digitization, and high CCD sensitivity.

AcknowledgementsThe precise measurement and evaluation processes of

this study were carried out at the Department of Inter-changeable Manufacturing and Industrial Metrology, Na-notechnology Laboratory of Vienna University of Techno-logy. The image processing analyses were performed at the Department of Mechanical Engineering of Adnan Menderes University.

AUTHORS

Numan M. Durakbasa – Vienna University of Technology (TU-Wien), Department of Interchangeable Manufacturing and Industrial Metrology (Austauschbau und Messtechnik/Produktionsmesstechnik & Qualität), Institute of Production Engineering and Laser Technology, Vienna University of Technology (TUWien), Karlspl. 13/3113, 1040 Vienna, Austria, [email protected]. tuwien.ac.at Ismail Bogrekci – Adnan Menderes University, Faculty of Engineering, Department of Mechanical Engineering, Adnan Menderes University 09010, Aydin, Turkey, [email protected] Pınar Demircioglu* – Adnan Menderes University, Faculty of Engineering, Department of Mechanical Engineering, Adnan Menderes University 09010, Aydin, Turkey, [email protected] Gökcen Bas – Vienna University of Technology (TUWien), Department of Interchangeable Man-ufacturing and Industrial Metrology (Austauschbau und Messtechnik/Produktionsmesstechnik & Qualität), Institute of Production Engineering and Laser Technology, Vienna University of Technology (TUWien), Karlspl. 13/3113, 1040 Vienna, Austria, [email protected]. tuwien.ac.at Aslı Gunay – Yildiz Technical University, Faculty of Mechanical Engineering, Department of Mechanical Engineering, Yildiz Technical University, 34010, Istanbul, Turkey, [email protected]


References[1] Leach, R. K., Haitjema, H., “Bandwidth character-

istics and comparisons of surface texture measuring instruments”, Meas. Sci. Technol., vol. 21, no. 3, 2010, pp. 1-9.

[2] http://www.taylor-hobson.com[3] EN ISO 4287:2009, Geometrical Product Specifi-

cations (GPS) - Surface texture: Profile method – Terms, definitions and surface texture parameters (ISO 4287:1997 + Cor 1: 1998 + Cor 2: 2005 + Amd 1: 2009) (includes Corrigendum AC:2008 and Amendment A1:2009).

[4] http://www.keyence.com/products/microscope/ microscope/vhx1000/vhx1000.php

[5] Bogrekci I., Godwin R. J., “Development of Image Processing Technique for Soil Tilth Sensing”, Biosystems Engineering, vol. 97, no. 3, 2007, pp. 323-331.

[6] Whitehouse D.J., “Stylus contact method for surface metrology in the ascendancy, Meas. Cont., vol. 31, no. 2, 1998, pp. 48-50.

[7] Blunt L., Stout K.J., Three-dimensional Sur-face Topography, Penton Pres, London, ISBN 1857180267, 2000, p. 320.

[8] Demircioglu P., Durakbasa M.N., Investigations on Machined Metal Surfaces through the Stylus Type and Optical 3D Instruments and their Mathematical Modeling with the Help of Statistical Techniques. „Measurement”, vol. 44, issue 4, 2011, pp. 611-619.


Articles56

Metrology for Pressure, Temperature, Humidity and

Airspeed in the Atmosphere

Anna Szmyrka-Grzebyk, Andrea Merlone, Krzysztof Flakiewicz, El bieta Grudniewicz , Krzysztof Migała

Submitted 26th October 2011; accepted 3rd December 2011

Abstract: The Joint Research Project METEOMET – “Metrology

for Meteorology” realized in the frame of the European Metrology Research Programme (EMRP) is described in the paper. The project is focused on the traceability of measurements involved in the climate changes: surface and upper air measurements of temperature, pressure, humidity, wind speed and direction, solar irradiance and reciprocal influences between measurands. It includes development and testing of novel instruments as well as improved calibration procedures and facilities for ground based observations, including in-situ practical calibra-tions and best practice dissemination.

The project consortium is based on 18 National Metrol-ogy Institutes (NMIs), three un-funded partners and sev-eral collaborators, such as universities, research centers, meteorological organization and institutions, from Europe and other non-European countries. Istituto Nazionale di Ricerca Metrologica (INRiM) in Italy is the project co-ordinator. Three Polish organizations participate in the project: the Central Office of Measure (MG-GUM), the Institute of Low Temperature and Structure Research (INTiBS) and the Wrocław University (UWr).

Keywords: weather station, temperature, humidity, pres-sure, environment, climate change

1. IntroductionThe project “Metrology for pressure, temperature,

humidity and airspeed in the atmosphere” is realized as the Joint Research Project METEOMET “Metrology for Meteorology” in frame of the European Metrology Re-search Programme (EMRP). The EMRP is implemented by a Regional Metrology Organisation (RMO) of Europe – EURAMET e.V. (the European Association of National Metrology Institutes) [1].

EURAMET coordinates the cooperation of National Metrology Institutes of Europe in fields like research in metrology, traceability of measurements to the SI units, international recognition of national measurement stan-dards and related Calibration and Measurement Capabili-ties (CMC) of its members. Through knowledge transfer and cooperation among its members EURAMET facilitates the development of the national metrology infrastructures. As well EURAMET is responsible for the elaboration and execution of the EMRP.

The EMRP is based on Article 185 of the Lisbon Trea-ty. It provides the opportunity for the user community and other stakeholders to directly suggest topics that the me-trology community should address with its resources. The

EMRP supports the collaboration of European metrology institutes, industrial organisations and academia through Joint Research Projects (JRPs).

This JRP METEOMET is focused on the traceability of measurements involved in the climate changes: surface and upper air measurements of temperature, pressure, humidity, wind speed and direction, solar irradiance and reciprocal influences between measurands. It responds to the need of new stable and comparable measurement standards, proto-cols, sensors and calibration procedures, data-fusion and uncertainty-evaluation methods, to enhance data reliability and to reduce uncertainties in climate models. It includes development and testing of novel instruments as well as improved calibration procedures and facilities for ground based observations, including in-situ practical calibrations and best practice dissemination. The development of novel instruments for the measurement of water vapour, the most important gas in the atmosphere and a key player in climate change, is a scientifically and technically relevant part of the project.

The project, started on 1 October 2011, is coordinated by Istituto Nazionale di Ricerca Metrologica (INRiM) in Torino (Italy). 18 European National Metrology Institutes and three Universities (un-founded partners) are involved to the project. 29 collaborators from all Europe representing meteorology organizations and other non-European coun-tries, institutes and instrument companies have declared their interest in the JRP participation. Three Polish organi-zations participate in the project:

– the Central Office of Measure (GUM), – the Institute of Low Temperature and Structure Re-

search (INTiBS), where the national temperature standard for a low temperature range is maintained,

– the Wrocław University (UWr) – the Department of Climatology and Atmosphere Protection.

2. The project aimWhy the project is needed? Because recent decades

have seen notable changes in global and European climate. The World Meteorological Organization (WMO) and the Bureau International des Poids et Mesures (BIPM) have established that many of the principal challenges faced by climate science are indeed measurement challenges. In 2010 the WMO signed a document MRA – “Mutual Rec-ognition Arrangement” elaborated by the BIPM and signed by majority of the NMIs in 1999 [2].

Development of homogeneous climate observations and data sets are basic aims of the project.

This project will respond to some principal needs:• Ensuring a defined traceability to the national stan-

dards for meteorological observations. Definition of a mea-


Articles 57

surement protocol in accordance with WMO guidance.Routine calibration procedures for most of the mea-

surements are not usually adopted but are necessary to maintain a high level of confidence in the quality of the data.

• Climate measurements uncertainty evaluation.The development of more accurate climate models to

reduce uncertainties in existing climate change scenarios is a fundamental task.

• Calibration of weather measurement stations and reference radiosondes.

Stations must be equipped with calibrated sensors for improving the reliability of measurements.

• Improved humidity sensors and calibration methods.Water vapour is the most important greenhouse gas in

the earth’s atmosphere and a key component for several physical and chemical processes. Therefore the humidity of air in terms of volume concentration of water is a key parameter to be measured for understanding the climate processes all over the world.

• Cross-linking of the weather and climate monitoring sites to establishing a well-modelled and cooperating (dis-tributed) European smart monitoring network.

Metrological cross linking of the weather and climate monitoring sites is an indispensable prerequisite to ex-plore the whole performance in terms of accuracy, reliabil-ity, traceability, comparability and cost efficiency. Only a complete, well-modelled and cooperating sensor network allows for low-uncertainty and full-area covering weather and climate-change modelling.

• Robustness of the historical temperature measurement data.

Published historical data often lack of a clear statement on the measurement technique and sensors, surrounding environment change, uncertainty budgets and traceability to standards and temperature scales of the different periods.

• Improve availability of data and promote their useEasy and rapid access to information achieved in dif-

ferent European data centres is needed for: modelling, data interpretation, quality control, network selection and net-work/system performance monitoring.

• Improve communication and co-operation between scientific community.

Communication between all the National Institutes or local Agencies interested to climate observations should be improved in order to realize a wide scale monitoring system which overcomes the nowadays braking up.

• Improve measurements program for better geographi-cal and temporal coverage and for near real time monitoring capability.

In Europe the expansion of measurement programmes is necessary to provide adequate global coverage, to in-crease the number and quality of weather stations with par-ticular attention to areas more sensitive to climate change.

3. The project structure schemeThe project structure scheme is presented in Fig.1. It

well reflects those aspects in its work packages (WPs) and tasks organisation.

WP1: Upper air measurements: sensors and techniques

Aim of this work package is to investigate the possibil-

ity for traceable absolute humidity sensors based on tunable diode laser absorption spectroscopy (TDLAS). TDLAS hy-grometers have big advantages in meteorological humidity measurements. They can be extremely fast but also small and lightweight.

The objectives of this WP can be summarized in the following list:

• realisation of a traceable TDLAS hygrometer,• evaluation of the strength and broadening of the spec-

tral absorption lines of the water molecule,• to improve comparability between existing sensor

technologies through:a) develop special calibration protocols to produce

quantitative information about the consistency of data ob-tained with different sensor technologies,

b) develop a new transportable humidity generator to enable on-site calibration of field hygrometers,

c) develop a new “fast humidity calibration system” for establishing traceability to radiosonde based measurements,

d) realize an international field humidity sensor inter-comparison campaign with traceable on-site calibration using a humidity generator as transfer standard. This inter-comparison is jointly organized by institutions involving state of the art as well as recently developed instruments from the major groups active in this field and the German National Standard. This activity will be closely coordinat-ed with the COST – Action ES0604: “Atmospheric Water Vapour in the Climate System (WaVaCS)”, the SPARC (Stratospheric Processes And their Role in Climate) water vapour initiative (a project of the World Climate Research Program), and the GRUAN of the GCOS a joint undertak-ing of the WMO.

The work package is coordinated by the Physikalisch- Technische Bundesanstalt (PTB), Germany.

WP2: Novel methods, instruments and meas-urements for climate parameters

The aim of this work package is to develop novel methods and instruments for the measurement of tem-perature, humidity, and pressure in lower and upper atmo-sphere, and to obtain new data to improve the accuracy of

Fig.1. The METEOMET project structure scheme


Articles58

the saturation water vapour equations in the temperature range between -80 °C and +100 °C

The WP is split into five tasks:2.1: Water vapour formulae improvement,2.2: Novel methods and instruments for atmospheric

humidity measurement,2.3: Novel atmospheric multi-sensors,2.4: GPS and Galileo based measurements,2.5: Development of accurate laboratory calibration

facilities and procedures for air temperature sensors.The development of novel instruments and measure-

ment methods, in the framework of a metrological re-search project oriented towards climate and meteorology, is necessary to answer to the needs expressed by the me-teorological community, in terms of improvement of the accuracy for the measurement of atmospheric parameters.

CNAM, the National Metrology Institute in France, is a lider of the work package.

In this work package, CNAM, CETIAT (Centre Tech-nique des Industries Aérauliques et Thermiques from France), and MG - GUM from Poland will realize mea-surements of saturated water vapour pressure over water and ice in the temperature range between -80 °C and +100 °C, where vapour pressures lie between 50 mPa and 101 kPa. Two different devices will be used, to detect and quantify corrections related to possible systematic effects, and define their impact on the final uncertainty budget.

MG – GUM has a water vapour pressure cell made of stainless steel, realized as a compact thick-wall saturator and installed in a thermostat – calibration bath. The sys-tem is designed to minimize temperature gradients and water contamination. GUM will carry out measurements of saturated water vapour pressure over water and ice in the temperature range in the whole mentioned temperature range and will perform tests at temperatures down to -90 °C as well.

CETIAT, GUM and INRIM will compare results of water vapour pressure measurements obtained from the three experiments, and, where possible, will merge data and propose a new equation for the water vapour pres-sure curve.

WP3: Traceable measurement methods and protocols for ground based meteorological observations

The biggest group of participants are involved in real-ization of the work package since the activities can widely be distributed and carried on separately in order to gain in terms of time and resources. The wide participation of several nations also brings the advantage of establishing or enforcing numerous relationships between NMIs and meteorological services and climate research groups over Europe. The tree Polish organizations are included in the WP as well.

The aim of this work package is to develop traceable measurements methods and protocols for temperature, hu-midity, pressure and airspeed ground-based measurements needed for climate studies and meteorological long-term and wide scale observations. Weather stations-based mea-surements are the main subject for this WP.

Six tasks are planned in the WP:3.1: Definition of state of art in European countries of

weather stations performance, use, calibration and trace-ability,

3.2: Evaluation of the effect of solar radiance and age-ing on weather stations,

3.3: Development of a method for establishing trace-ability in wind speed measurements,

3.4: Development of a laboratory calibration facility and procedures for the combined, simultaneous, calibra-tion of temperature, humidity and pressure sensors in weather stations,

3.5: Construction of a facility for in situ traceable cali-bration of weather stations also for special purposes and under extreme environmental conditions (high mountains, poles),

3.6: Protocols for quality assessment of ground-based measurement and software validation of in situ weather stations.

The Polish institutes will collaborate in the major-ity of these tasks. They will create database of European weather stations: sensors, design, calibration practices, traceability routes and report on the available models.

The aim of the task 3.4 is the development of a facility for the combined, simultaneous, calibration of tempera-ture, humidity and pressure sensors in weather stations. This device will allow the study of the impact of interfer-ing quantities on individual calibration curves of T, H, P sensors. INTiBS and GUM will perform measurements of temperature, humidity and pressure for sensors used on weather stations against appropriate reference standards, in laboratory. The three parameters ranges will be defined according to the real conditions the weather stations can be exposed to, and will be: (-50 ÷ 50) °C, (10 ÷ 98) % Rh and (80 ÷ 110) kPa. Some of general aspects of mutual in-fluences of temperature, humidity and pressure will be in-vestigated. Analysis of achieved results with comparison to routine in situ measurements performed by the Wro-claw University. The UWr will provide the basis for the definition of calibration procedures for weather stations.

During the task 3.5 realization INRiM will study and

Fig. 2. The Wrocław University weather station apparatus


Articles 59

manufacture a reduced dimension facility for the in situ calibration of weather stations. It will allow simultane-ous calibration of temperature, humidity and pressure sensors, covering the whole expected range according to the stations locations. It will feature temperature control within 0.05 °C between -20 °C and 50 °C, pressure con-trol within 100 Pa between 75 kPa and 110 kPa, humidity calibration by comparison at 1.5 % uncertainty between 5 % and 95 % Rh. It will be a special chamber made in such reduced dimensions that it can be easy transported. INTiBS and UWr will organise, manage and take a care of the field works for the testing in extreme conditions of weather stations calibrated using the INRiM facility. Two sites are proposed: the research station in the Western Sudetes (Mt. Szrenica, 1365 m being a very moist and windy place with high frequency of rime/icing) and the station on the Spitsbergen (Svalbard) with its logistics and infrastructure that give an easy access to a natural arctic environment.

Fig. 3. The weather station apparatus in the Western Sudetes

The Wrocław University has a permanent access to the Wester Sudets station, by whole year, and to the Spitsber-gen station in the period when observation and measure-ments are possible.

The aim of the task 3.6 is the development of proto-cols for quality assessment to improve the reliability of ground-based measurement. After an assessment of the needs, the validation of software for in situ weather sta-tions will be performed.

The work package is coordinated by INRiM – Italy.

WP4: Harmonisation of data. Assessment of the his-torical temperature data, data fusion

The work package is coordinated by the Czech Me-trology Institute (CMI – Cesky Metrologicky Institut) in Brno.

The aim of this work package is to investigate sources of uncertainty in historical temperature data, include them into the uncertainty budget and correct the input to the climate models thus enhancing climate change detection, prediction and adaptation assessments. A novel software based model for the harmonisation of data under such a metrological approach will be developed.

A further objective of this WP is to enforce the whole relevance of the JRP impact. After the collection of the

data and assessment of methods has been performed and when such a metrological analysis will be adopted, the re-quirement of standardised and traceable methods for col-lecting temperature data over wide scales and long terms will be strengthened. The main result will be the reduced variability of data made available for similar analysis to be performed also in the future. Harmonization uncertainties, statistical A Type and B Type uncertainties will then be included in the temperature trends evaluations.

WP5: Creating impactThe technical work in this project will deliver advanc-

es in measurement and calibration to a highly interna-tional science area of critical importance to future global sustainability. Therefore, it is crucial that the outputs are disseminated widely to meteorology organisations, and other stakeholders. The geographical dispersion of us-ers requires suitable dissemination mechanisms: web-based knowledge transfer and training will be used, as will decentralised knowledge transfer mechanisms, with each project participant liaising with local (national) me-teorology organisations and practitioners. Presentations during the period of the project will provide short-term high-impact dissemination, while publications and com-puter-aided learning applications will provide long-term dissemination.

The objectives of the impact work package are to:• provide links, participation, and knowledge transfer

to the “end user” community; including national and inter-national climate and meteorology institutes and organisa-tions, measurement users and instrument companies,

• feed into the development of key standards and protocols through appropriate climate and meteorology bodies,

• develop a coherent approach at the European level in this field of metrology.

The new equipment, facilities, measurement methods and primary standards developed by the project will be disseminated in order to facilitate best practice amongst meteorology organisations, and European NMIs. In all dissemination, there will be an emphasis on promoting measurement traceability and realistic understanding of measurement uncertainty.

The knowledge developed in the project will be trans-ferred via:

• website dissemination of project progress and out-puts, web-based training material and public access web page containing real time data including measurement uncertainty (T, H, P) from some calibrated weather sta-tions in Europe,

• input into WMO and national working groups de-veloping new best practice and documentary standards in this area,

• stakeholder workshop meetings organized at Euro-pean and local level, as opportunities for proposal, agree-ment and adoption of strategies,

• publications – reports, good practice guides, and sci-entific papers.

Training will be organised, to transfer improved met-rological practices and procedures, targeted to meteoro-logical instrument users, and those analyzing meteoro-logical data. Training, using web-based material, and via training courses held in at least 5 countries/regions across


Articles60

Europe will be organized in Poland too.

WP6: JRP management and coordinationThe large number of NMIs/DIs (Designated Institute

to EURAMET) participating in this JRP together with unfunded JRP-Partners makes the management an im-portant aspect of this project, in order to assure a fruitful coordination of scientific activities and budget control. In addition a number of collaborators are linked to this JRP. The coordination structure for this project will be based on the JRP-Coordinator, a Scientific Advisory Group and the WP leaders. A project Scientific Supervisor will assist the JRP-Coordinator in preparing the reports.

In order to optimize the coordination and the informa-tion flux from and to the management and the participants, and to better reflect the main objectives of this JRP, the number of technical work packages has been limited to four only. Work package leadership is assigned to key scientists, experts in the fields of the WP tasks, thus in-creasing the scientific value to those roles. WP leaders are at the same time responsible of the coordination for progress of the individual work packages and take part in the management and coordination decisions.

The progress of the project will be monitored against the Gantt chart and the deliverable list. All JRP Partners will be required to report their scientific and technical activities, the results achieved, their contributions to de-liverable, eventual risks mitigations adopted and problems encountered in respecting the Gantt.

The aim of this task is to provide the scientific and financial reporting to EMRP, according to the guidelines. The first and final periodic reports and financial report-ing, the interim reports, the final publishable report are the subject of this task, together with the organisation of JRP meetings.

Reporting, based on information provided by all JRP-Partners, will be provided to EURAMET according to the guidance requirement in terms of content and deadlines.

4. SummaryThe need of establishing the road through a uniform

approach to the traceability for those measurements in-volved in climate studies, such as pressure, temperature humidity and airspeed in the atmosphere is a fundamental goal of the project. Weather observations and collection of data are not carried on with similar methods and proce-dures in the different European countries. A large number of participants from the 18 countries allows a better under-standing of the present situations, and constitutes a wide

forum for discussing and proposing common procedures. NMIs operating at regional level, in cooperation with me-teorological institutes can directly disseminate the results and best practices. The JRP-Consortium brings together the largest European NMIs having broad experience and expertise in metrology for pressure temperature, humid-ity and airspeed. Expertises have also been acquired in mathematics applied to data harmonisation and software validation. The various tasks are carefully assigned to the corresponding experts of the Consortium.

The development of novel sensors, techniques and facilities can be achieved thanks to the well-established scientific and metrological capabilities and activities of the NMIs involved. Testing in different environmental conditions, moreover, is possible since NMIs operating in several and different areas are included.

The JRP MeteoMET started on 1 October 2011. In the middle of October a kick-off meeting was orginised by the project Coordinator – the INRiM. It was held in the Reale Collegio Carlo Alberto di Moncalieri where is located the Historical Meteorological Observatory of SMI (Società Meteorologica Italiana-Project Partner). About 60 persons from European metrological and meteorological organiza-tions participated in the meeting. The World Meteorologi-cal Organization (WMO) was represented on the meeting as well. During the meeting the Scientific Advisory Group and a project Scientific Supervisor were elected.

The JRP MeteoMet is one of the largest projects of the EMRP.

Acknowledgements The research leading to these results has received

funding from the European Union on the basis of Deci-sion No 912/2009/EC.

AUTHoRS

Anna Szmyrka-Grzebyk* – Instytut Niskich Tempera-tur i Badań Strukturalnych PAN, 50-950 Wrocław 2, ul. Okólna 2, [email protected]

Andrea Merlone – Istituto Nazionale di Ricerca Met-rologica, 10-135 Torino, strada delle Cacce 73-91, Italy, [email protected]

Krzysztof Flakiewicz – Główny Urząd Miar, 00-139 Warszawa, ul. Elektoralna 2, [email protected]

Elżbieta Grudniewicz – Główny Urząd Miar, 00-139 Warszawa, ul. Elektoralna 2, [email protected]

Krzysztof Migała – Uniwersytet Wrocławski, Zakład Klimatologii i Ochrony Atmosfery, 51-621 Wrocław, ul. Kosiby 6/8, krzysztof.migala@ uni.wroc.pl.


References [1] www.euramet.org [2] www.bipm.org

Fig. 4. JRP MeteoMet management structure


Articles 61

Applying Reverse Engineering to Manufacture the Molds for the Interior Decorations Industry

Florin Popister, Daniela Popescu, Dan Hurgoiu, Radu Racasan


Abstract: Reverse engineering is a way of replicating complex

shapes and designs with the use of specialized hardware and software. This paper presents the methodology need-ed for obtaining the 3D virtual model starting from exist-ing models made by hand, and then editing this virtual model to manufacture molds. These are then machined using three axis CNC manufacturing equipment. The pa-per approaches some of the key challenges for reverse engineering: repairing damaged sections in the model, recreating an ordered pattern and balancing the number of points and processing time. An analysis is made to try and optimize both the scanning, processing and machin-ing processes in order to decrease the time and costs as-sociated.

Keywords: scanning, reverse engineering, mould, CNC equipment

1. Introduction Interior decorations industry is a dynamic industry

which, through the variety and multitude of products they offer, tries to keep up with customer needs. The in-creasing number of new constructions and the need for renovations or interior redesign requires that the decora-tions’ industry designs new appealing products with low cost and manufacturing time.

In order to create such products that meet the require-ments of the customers, the work starts in most cases with the artistic design. Through skill and imagination the art-ist creates a model that meets his reasoning in terms of aesthetic. But in the process of creating the model, the artist crafts by hand the geometry of the surface and the surface texture. In order to get a high quality product suit-able for the demands of the mass market it is important to create products that appeal to the general consumer and current architectural trends. This will form the basis of the design process that will later be converted into high quality casting molds in the subsequent manufacturing of the final product.

Conception is for the artist and the execution is for engineer’s “Scientists generalize, artist individualize” (Jules Renard). The science behind reverse engineering creates from the primary model envisaged by the artist, a reliable model of good quality that is used for mass production of products or for creating one of a kind pro-totypes.

Reverse engineering techniques allow for the selection of the optimal strategy for manufacturing the molds by first digitizing and artists concept and then processing

the data to obtain the CAD model. The techniques and concepts used by the artists, lead to designs that feature complex and freeform surfaces. Reverse engineering methods enables the conversion of aesthetic design to virtual form design.

The artistic model is created on an aesthetic basis but processed CAD model must be able to both reproduce the artistic concept and allow for the manufacturing of the part. Reverse engineering provides the link between art and the manufacturing industry by bridging the gap between the design and manufacturing process. Reverse engineering technique allows the use of different forms and materials to express the artistic design (clay, wood, and fiberglass), however, obtaining a virtual model, based on the artistic creation poses several challenges.

Reverse engineering methods and techniques are be-ing applied in several key areas of the industry in order to accomplish the following [1]:• Design of new components;• Reproduction of an existing component;• Recovery of a damaged or broken component;• Development of model precision;• Observation of a numerical data.

The authors of this paper propose an algorithm and experimental case studies to obtain the desired molds for the interior decorations industry, starting from hand-made models and using reverse engineering techniques. Through the strategy of reverse engineering allows the transition from the artistic concept to the parameterized one no matter how asymmetrical, abstract or free form the shapes are. Reverse engineering technology involves

Figure 1. The steps required to achieve molds from hand shaped pieces


Articles62

a complex activity theoretical foundation to go from the primary model to the virtual model – CAD. It will be created based on an algorithm controlled by reasoning based on experience and theory to identify the most ef-fective model.

2. Methodology and case studies This chapter of the paper presents a proposed method-

ology (Figure 1) for manufacturing molds starting from hand modeled designs that are digitized with revere en-gineering techniques. The case studies are based on three gypsum handmade models, two of them used for wall decorations and one for ceiling, each with different char-acteristics made by designers.

A. Handmade gypsum modelIn the first phase designers manually craft the shapes

from gypsum. These models are designed for the inte-rior decorative industry. These three models have been chosen due the complex shape (Figure 2) of the surface making it difficult to design these freeform shapes with CAD software.

a b c

Figure 2. Gypsum model made by hand

B. Evaluation of the models and surface finishThe material used for making the parts is very fragile namely gypsum. Even so, using a contact scanner (Ren-ishaw Cyclone Series II, Renishaw Ltd.) won’t damage the surface of the part because the surface is coated with a thin layer of lacquer. This was considered to be the op-timum scanning technique making it possible to capture thefinedetailscraftedbytheartist.

C. Scanning strategyBecause the material from which the pieces are made is gypsum, and the small details on each of the models a different scanning strategy was applied for each indi-vidual model.Forthefirstmodel(Figure2a)theauthorsusedacontactscanning technique with the following parameters:• Stylus – ball type – Ø2 mm• Distance between two scanned points – 0.1 mm• Distance between two scanning lines – 0.2 mm• Scanning speed – 600 mmIn the second case study (Figure 2b), for the ceiling tiles model, the parameters are:• Stylus – ball type – Ø2 mm• Distance between two scanned points – 0.2 mm• Distance between two scanning lines – 0.2 mm• Speed – 750 mmThe characteristics of the scanning process for the last case study (Figure 2c) which is the wall decorative model are:• Stylus – ball type – Ø2 mm• Distance between two points – 0.1 mm

• Distance between two scanning lines – 0.3 mm• Speed – 650 mm

D. DigitizationIn the digitization phase a contact scanning equipment, Renishaw Cyclone Series 2 CMM scanner was used. In this way the information resulted after scanning is and ordered point cloud of the scanned part (Figure 3). The digitized surface is processed using a dedicated software CopyCAD (Delcam Ltd.) [2] to compute the triangula-tion and then the 3D surface of the model. This process also eliminates any unwanted points resulted from the scanning process.

Figure 3. Point cloud of each part model

E. Reconstruction and correctionThe imperfections that have to be inspected and cor-

rected because the pieces are shaped by hand can be: form deviation of the margins for parts that are meant to join together, flatness of planar surfaces or consistent repetition of regular patterns that may be lost during the artists creation. These details can be repaired [3] or cor-rected once the CAD model has been generated using software that allows for editing of the surfaces.

Planar surfaces (Figure 4 b) in certain areas of gypsum models are very difficult to create by hand. Furthermore the joining inner edges of the molds (Figure 4 a) that are essential in the manufacturing of molds for mass produc-tion of parts are difficult craft manually. As mentioned before applying scanning strategies to produce 3D mod-el then allows for the editing of these joining edges to straighten them.

a b

Figure 4. Model with extraction edges a) and flatness b) issue

In case of degradation and damage of a used mold for making polystyrene ceiling covers, the initial model was remodeled manually using gypsum. After several at-tempts to obtain the correct mold depth and increase the existing depth of the figures on the model (Figure 5 a) by hand, a dedicated CAD software enabled the scaling of the 3D model in one direction, Z axis (Figure 5 b). The values used for the scaling process highlighted certain features on the model. In this way the 3D model was cre-ated at an appropriate depth for the texture on polysty-rene to be produced correctly.


Articles 63

a bFigure 5. Difference between the a) original and b) scaled model

Another example of presents a way of overcoming the problem of nonuniform joining edges[4] (Figure 6 a) of a certain part(Figure 6 b). The radius of curvature which in the case of manual crafting cannot be created perfectly. There is the possibility of joining problems between parts due to non uniform shapes. These problems can be cor-rected when the CAD model is edited using a dedicated CAD software CATIA V5, Imagine and Shape Module [5] where parameterized joining edges are created recon-structed that are later used in the CAM software generat-ing the manufacturing code.

a bFigure 6. Boundary edges a) of the boards and the way of joining b) them tougher

F. Manufacturing the moldsThe manufacturing process was achieved using Pow-

erMill (Delcam Ltd.) CAM software [6], and a Vertical 3-axis Milling Center. The manufacturing process was simulated in the CAM software in order to avoid any problems in the machining processes such as: collisions between the tool and the work piece, proper selection of the tools used during the manufacturing phase to ensure that the resulted 3D model is as accurate as possible. After finishing the manufacturing process, in each case study the resulted mold (Figure 7 b) was analyzed to compare the finished product to the initial 3D model (Figure 7 a).

is feasible to use the scaling method to recreate certain worn out areas.

The analysis was performed in the case of the third where the resulted mold (Figure 9 b) was correct ma-chined in accordance with the 3D model (Figure 9 a)

a b

Figure 9. The third case study a) 3D mold and b) the machined version

3. ConclusionModern scanning and manufacturing methods are the

main driving forces behind the design of new complex and freeform surfaced parts. Manufacturers want to cre-ate mass produced parts with few losses and lower manu-facturing costs. In the case of the interior decoration in-dustry there are a vast number of models that need to be produced each requiring specialized injection or casting molds that are expensive to design and manufacture.

Imagination of the designers and requirements from clients have no limits when achieving models with com-plex forms, which due to time, resources or cost cannot be designed within CAD software. The creation process starts with designers that craft by hand the majority of the models. Thus reverse engineering techniques in this area is very important in terms of correcting or reconstruction of certain parts of the models and transform them into a 3D model.

The models created manually by designers will always have slight imperfections in details, form or surface de-viations. Appling reverse engineering techniques these imperfections can be very easily repaired.

A major benefit of applying reverse engineering tech-niques in the interior decorations industry is that after the 3D scanning process the parts can be reconstructed or corrected. In order to obtain a larger sized model with the same features of the original a scaling process in one or more axes is required.

There is also the possibility of combining different de-

a bFigure 7. Resulted mould a) 3D and b) the machined

In the second case study the shapes on the initial mould were damaged and not so visible due to wear. The result-ed virtual 3D mold (Figure 8 a) after scaling the model at the necessary depth, was compared to the manufactured mold (Figure 8 b) and the results obtained proved that it

a bFigure 8. The reconstructed mould a) 3D and b) the ma-chined version for the second case study


Articles64

tails such as flowers, geometric shapes or other freeform surfaces from a 3D model to another in order to obtain a new design. Obtaining the 3D model of a handmade modeled part is the first step towards achieving the de-sired model by modifying and editing the virtual model with the help of dedicated CAD software.

In the case of mass produced parts the 3D model is imported into a CAM software in order to generate the NC file needed to manufacture the molds on CNC Ma-chining Centers.

As further area of research for reverse engineering techniques the authors are investigating the reconstruc-tion and restoration of missing parts from different mon-uments or important decorations from old historic build-ings that have been damaged or are missing.

ACKNOWLEDGEMENTS This paper was supported by the project “Doc-

toral studies in engineering sciences for developing the knowledge based society-SIDOC” contract no. POSDRU/88/1.5/S/60078, project co-funded from Euro-pean Social Fund through Sectorial Operational Program Human Resources 2007-2013.

AUTHORS

*Popişter Florin, Technical University of Cluj-Napoca, bd. Muncii 103-105, [email protected]

Popescu Daniela, Technical University of Cluj-Napoca, bd. Muncii 103-105, [email protected]

Hurgoiu Dan , Technical University of Cluj-Napoca, bd. Muncii 103-105, [email protected]

Răcăşan Radu , Technical University of Cluj-Napoca, bd. Muncii 103-105, [email protected]


References

[1] Eyup Bagci., Reverse engineering applications for recovery of broken or worn parts and re-manufactur-ing: Three case studies. „Advances in Engineering Software”, 40 (2009), p. 407-418.

[2] http://www.copycad.com/general/copycad.asp[3] G. Uçoluk and I.H. Toroslu, “Automatic reconstruc-

tion of broken 3-D surface objects,” Comput. Graph., vol. 23, no. 4, pp. 573-582, 1999

[4] Yilmaz Ceken, “Three dimensional object recon-struction from boundary data,” METU, Ankara 1995

[5] http://www.3ds.com/products/catia/solutions/shape-design/

[6] http://www.powermill.com/general/hsm.asp.

Jamris 2012 vol 6 no 3

Documents

Transcript of Jamris 2012 vol 6 no 3