1 Reliability Block Diagram Modeling – A Comparison of Three Software Packages Aron Brall, SRS...
-
Upload
phoebe-mckenzie -
Category
Documents
-
view
218 -
download
1
Transcript of 1 Reliability Block Diagram Modeling – A Comparison of Three Software Packages Aron Brall, SRS...
1
Reliability Block Diagram Modeling – A Comparison of Three Software Packages
Aron Brall, SRS Technologies, Mission Support DivisionWilliam Hagen, Ford Motor Company, Powertrain Manufacturing Engineering Hung Tran, SRS Technologies, Mission Support Division
22007 RAMS – Brall, Hagen, Tran
THE SOFTWARE PACKAGES - 1 ARINC RAPTOR 7.0.07
From RAPTOR web site: “Raptor is a software tool that simulates the operations
of any system.” “Sophisticated Monte Carlo simulation algorithms are
used to achieve these results.” Our Take:
Pure Monte Carlo simulation tool to solve reliability block diagrams.
32007 RAMS – Brall, Hagen, Tran
THE SOFTWARE PACKAGES - 2 Reliasoft BlockSim 6.5.2
From BlockSim web site: “Flexible Reliability Block Diagram (RBD) creation.” “Exact reliability results/plots and optimum reliability
allocation.” “Repairable system analysis via simulation (reliability,
maintainability, availability) plus throughput, life cycle cost and related analyses.”
Our Take: Monte Carlo simulation with algorithms used to speed
the processing time. Also provides analytical calculation of reliability.
42007 RAMS – Brall, Hagen, Tran
THE SOFTWARE PACKAGES - 3 Relex Reliability Block Diagram
From Relex web site: “At the core of Relex RBD is a highly intelligent
computational engine.” “First, each diagram is analyzed to determine the best
approach for problem solving using pure analytical solutions, simulation, or a combination of both.”
“Once a methodology is determined, the powerful Relex RBD calculations are engaged to produce fast, accurate results.”
Our Take: Relex RBD appears to be a hybrid tool that uses algorithms
and simulation in varying combinations to solve reliability block diagrams.
52007 RAMS – Brall, Hagen, Tran
Why Compare Reliability Software
Analysts (especially new analysts) tend to report reliability software results as exact values
Engineering judgment, caution and experience are being supplanted by software analysis
Error checking is often absent Number of runs; confidence limits; garbage in, garbage out all
impact value of software analysis
62007 RAMS – Brall, Hagen, Tran
One Block Model
Block Parameter
Block Probability Distribution
Parameter 1 Parameter 2
Failure Distribution
a Weibull Shape 1.5 Scale 1000
Repair Distribution
a Lognormal Mu 5 Sigma 0.5
72007 RAMS – Brall, Hagen, Tran
Simple Model BlockName
FailureDistribution
Parameter1
Parameter2
a Weibull Shape 1.5 Scale 1000
b Normal Mean 250 Std Dev 50
c Exponential 10000 0
d Lognormal Mu 6 Sigma 2
e Weibull Shape 1.5 Scale 2300
f Normal Mean 250 Std Dev 50
g Exponential 10000 0
h Lognormal Mu 8 Sigma 1
i Weibull Shape 1.5 Scale 1000
j Normal Mean 250 Std Dev 50
k Exponential 10000 0
l Lognormal Mu 8 Sigma 3
m Weibull Shape 2.0 Scale 1000
n Weibull Shape 3.0 Scale 1000
o Weibull Shape 4.0 Scale 1000
p Weibull Shape 0.5 Scale 1000
q Weibull Shape 0.4 Scale 1000
cMTBF: 10000
Qty: 1R: 0.99005
dFailure: Log Normal
Mu: 6Sigma: 2
Qty: 1R: 0.757228
eFailure: WeibullChar. Life: 2300Shape Fact.: 1.5
t0: 0Qty: 1
R: 0.990975
fFailure: NormalMean: 250StdDev: 50
Qty: 1R: 0.99865
gMTBF: 10000
Qty: 1R: 0.99005
nFailure: WeibullChar. Life: 1000Shape Fact.: 3
t0: 0Qty: 1
R: 0.999
hFailure: Log Normal
Mu: 8Sigma: 1
Qty: 1R: 0.999657
iFailure: WeibullChar. Life: 1000Shape Fact.: 1.5
t0: 0Qty: 1
R: 0.968872
qFailure: WeibullChar. Life: 1000Shape Fact.: 0.4
t0: 0Qty: 1
R: 0.67159
jFailure: NormalMean: 250StdDev: 50
Qty: 1R: 0.99865
bFailure: NormalMean: 250StdDev: 50
Qty: 1R: 0.99865
lFailure: Log Normal
Mu: 8Sigma: 3
Qty: 1R: 0.871101
aFailure: WeibullChar. Life: 1000Shape Fact.: 1.5
t0: 0Qty: 1
R: 0.968872
pFailure: WeibullChar. Life: 1000Shape Fact.: 0.5
t0: 0Qty: 1
R: 0.728893
kMTBF: 10000
Qty: 1R: 0.99005
mFailure: WeibullChar. Life: 1000Shape Fact.: 2
t0: 0Qty: 1
R: 0.99005
oFailure: WeibullChar. Life: 1000Shape Fact.: 4
t0: 0Qty: 1
R: 0.9999
1::1 1::1
1::1
1::2
1::21::1
3::6
Start End1::1
82007 RAMS – Brall, Hagen, Tran
Large Model
92007 RAMS – Brall, Hagen, Tran
Complex Model
102007 RAMS – Brall, Hagen, Tran
Results of SimulationsModel Parameter Trials or
RunsTime (hours)
Raptor BlockSim Relex
One Block Reliability 1,000 1,000 0.3797 0.3663 0.365One Block Availability 1,000 1,000 0.8927 0.8894 0.843Simple Reliability 1,000 100 0.983 0.977 0.978Simple Availability 1,000 100 0.9955 0.9892 0.978Simple System Failures 1,000 100 0.017 0.023 Not ReportedLarge Reliability 10,000 61,362 0.7024 0.737 0.6914Large Reliability 1,000 61,362 0.718 0.729 0.707Large Availability 1,000 61,362 0.858 0.861 0.691Large Availability 10,000 61,362 0.847 0.865 0.6866Large MTTFF: (Hours) 10,000 61,362 144,775.99 201,679.13 146,321.53Complex Reliability 10,000 100 0.1313 0.1315 0.0988Complex Availability 10,000 100 0.3877 0.3741 0.3333Complex MTBF (MTBDE)(Hrs) 10,000 100 36.2732 39.3565 33.92Complex MTTR (MDT)(Hrs.) 10,000 100 68.3853 62.7677 74.51
Software PackageModel Data
112007 RAMS – Brall, Hagen, Tran
What Do the Results Tell Us
If precision is required, it isn’t there One to two significant figure agreement at best between packages Confidence limits are necessary for data
Some parameters are either defined differently, or calculated using such diverse algorithms or methodologies that they aren’t comparable
Errors in modeling or application of the software can go undiscovered when only one software package and one analyst are used
The complexity of large models and different issues with each software interface opens up many opportunities for human failure
Checking a model for errors can be more time intensive than creating the original model
122007 RAMS – Brall, Hagen, Tran
Cautions - 1 Use of a single model, especially a highly complex model, to
demonstrate compliance with a requirement is error prone and risky Many times the results of these simulations are used to demonstrate
compliance with a specified reliability or availability requirement. A result that would show a Reliability of 0.85 when the
requirement was 0.90 might cause redesign, request for waiver, or other action to address the shortfall.
The shortfall may be due to the parameters used for the simulation, the algorithms used by the software, a lack of understanding of how long to simulate, how many independent random number streams to use, and/or how many runs to use.
Analytical solutions for highly complex models are based on approximations.
132007 RAMS – Brall, Hagen, Tran
Cautions - 2
The programs do not necessarily describe variables in the same manner.
i.e.When using the Lognormal distribution, there was a difference in terminology between Raptor and BlockSim.
Raptor allows the Lognormal to be entered as Mean and Std Dev. or Mu and Sigma.
BlockSim only uses Mean and Std. Dev., but this is the same as Raptor’s Mu and Sigma.
A novice could waste a great deal of time clarifying what needs to be entered as data.
142007 RAMS – Brall, Hagen, Tran
Cautions - 3
Modeling special cases can be difficult because of the way the programs handle standby (which was in our models) and phasing (which was not in our models).
Output parameters were not consistently labeled. The user should understand the difference between MTTF, MTTFF, MTBDE, and MTBF for reliability and MDT and MTTR for maintainability.
152007 RAMS – Brall, Hagen, Tran
Cautions - 4
The products provide reliability and availability results with various adjectives such as “mean”, “point”, “conditional”, etc.
A review of the literature provided with the packages is necessary to understand these terms and relate them to those found in specifications, handbooks, references, and texts.
It is a serious issue that there doesn’t appear to be standard and/or consistent terminology and notation from one program to another as well as to standard literature in the field.
162007 RAMS – Brall, Hagen, Tran
Cautions - 5
Flexibility Each package has tabs, checkboxes, preferences, defaults, multiple
random number streams, selectable seeds for random numbers, etc to facilitate the modeling, analysis, and simulation process.
Flexibility can provide huge pitfalls to the analyst. Care in modeling, and use of support services provided by the
software supplier is a good practice. Numerous runs and reruns may be necessary due to idiosyncrasies of
the software, Beware of errors in modeling, confusion of parameter definition, etc. Problems compound as a variety of failure distributions are
intermixed with a similar grouping of repair distributions. As a model becomes more complex, simulation becomes mandatory
172007 RAMS – Brall, Hagen, Tran
Observations - 1 The models can run quickly even on old Pentium II PCs, or they can
take hours to run. Length of simulation time, number of runs, and failure rate of the
system can all contribute to lengthening of simulation time. One of the models took in excess of 1 hour on a 3 GHz Pentium
IV. Convergence of the results is heavily dependent on how consistent
the block failure rates are. For example, one block with an MTBF of 1000 hours, can double
or triple simulation time. The display during simulation on some of the packages shows the
general trend, but there can be a lot of outliers. One model failed to converge on one of the packages – again this
may have been due to a subtle preference selection (or non-selection).
182007 RAMS – Brall, Hagen, Tran
Observations - 2 The display of Availability and or Reliability during simulation can
be useful for seeing how the simulation is behaving. For most models, this rapidly stabilizes to the first decimal place,
then the second decimal place tends to bounce around. Usually you get the first 2 significant figures in a hundred runs.
We have the impression that most of the user interfaces were designed by software designers, working with R&M engineers.
The problem is that we seem to have gotten what an R&M engineer would tell someone never having used the product.
For example, it’s really annoying that you have to double click and work through tabs to put data into blocks in the block diagrams; the alternative is to use the Item Properties Table, which doesn't let you create blocks and in some cases change probability distributions.
192007 RAMS – Brall, Hagen, Tran
Recommendations When demonstrating compliance to a requirement is required
Model system using one of the following approaches to reduce human error
Have one analyst model in two different software packages Software methodologies are sufficiently different to
avoid repeating errors Have second analyst perform detailed audit of model and
data entry Have two analysts independently model and enter data
Compare results Results should agree within +/- 3 Standard Errors of the
Mean Make detailed notes of assumptions, methods, simulation values,
etc. to provide an audit trail