Technical Paper Research on Misfire Detection Algorithms ...and affected misfire detection accuracy....

－24－

Research on Misfire Detection Algorithms for Motorcycle Engine Firing at Uneven IntervalsTechnical Paper

Key Words: misfire detection, OBD II, internal combustion engine, motorcycle, signal processing

Katsunori TASAKI＊１

1. Introduction

Exhaust gas from internal combustion engines

contains CO, HC, NOx and CO2 and has been one

cause of recent serious environment problems such

as global warming, air pollution and so on. On

the other hand, the market for internal combustion

engine automobiles and motorcycles keeps growing

because of their convenience. Therefore, i t is

necessary to take measures for environmental

issues. Many countries and regions have adopted

regulations on the amount of emissions. These

regulations have become stricter in recent years.

Many emissions reduction technologies such as

fuel injection systems with oxygen feed-back

control and a 3-way catalytic converter have been

developed also for motorcycles.

Regulations for environment protection do not

involve only restrictions on emissions but also the

introduction of OBD systems which automatically

diagnose malfunctions in the vehicle and inform

the driver of the fact. The detection targets of the

OBD II system include an emission control system

malfunction that causes deterioration in tailpipe

emissions. With this system, it is possible to detect

cases where emissions continuously exceed the

regulation limits. The OBD II system has already

been implemented in automobiles. For motorcycles,

implementation will become mandatory when Euro-

5 emission regulations are introduced in Europe and

Bharat Stage VI (BS-VI) emission regulations, in

India.

One item for which detection is mandatory in

OBD II regulations is misfire. Misfire occurs when

the engine does not fire correctly due to ignition

failure or poor combustion of the air fuel mixture,

resulting in serious deterioration of tailpipe emissions

due to discharge of unburned gas. It may also cause

deterioration of the catalytic converter. For these

reasons, the OBD II regulations require detection

and warning for misfiring occurrences which cause

worse tailpipe emissions than the specified levels

and/or pose a risk of catalytic converter erosion by

overheating.

Misfire detection technology for automobiles has

been well developed and there are many strategies

that are currently employed, using variation in

characteristics, such as crank angular velocity,

cylinder pressure or ionic current. These technologies

have proven to be effective in the detection of

misfiring in automobile engines.

However, in the case of motorcycle engines,

have the following characteristics compared with

automobile engines:

- there are unique engine variations such as

engines with uneven firing intervals; and

Research on Misfire Detection Algorithms for

Motorcycle Engine Firing at Uneven Intervals※

＊１ System Development Department, R&D Operations

※ Received 14 May 2018, Content reprinted from SAE Technical Paper 2017-32-0052, © SAE International. Reprinted with permission. Further distribution of this material is not permitted without prior permission from SAE International.

－25－

Keihin Technical Review Vol.7 (2018)

TechnicalPapers

sensors or other devices. Firstly, performance

of these algori thms was evaluated efficiently

and quantitatively, by control simulation using

various measurement data as input. The algorithm

which showed good potential in simulation was

subsequently integrated in the ECU in order to

confirm its effectiveness through vehicle testing.

Finally, this paper summarizes the prospect of

practical utilization suggests directions for future

research.

2. Testing Environment

2.1. Test Engine

In this research, a motorcycle with a water-

cooled uneven firing twin-spark V-twin engine was

used as a test vehicle. The engine was installed with

a twin-spark firing system in which two spark plugs

were equipped for each cylinder. Table 1 shows

the specifications of the engine. The engine was

developed for a high performance sports motorcycle.

2.2. Trigger Wheel / Pick-Up Sensor / Timing

Teeth

The test vehicle had a trigger wheel and a pick-

up sensor used for fuel injection and ignition control.

The trigger wheel with 22 teeth (placed at each 15

degrees of rotation with 2 continuous missing teeth)

on its periphery was mounted on an extension of the

crankshaft. A pick-up sensor was installed on the

outside of the trigger wheel. It detected the edge of

the teeth and generated a sinusoidal wave.

The positions of these devices are shown in

Figure 2 and Figure 3. The crank angular velocity

can be calculated using these devices.

In addition, the missing tooth sections correspond

to the following 2 stroke timing points;

- Cyl. #1: Suction stroke, Cyl. #2: Expansion

stroke

- there is a lower combustion stability on no-load

or low-load driving points because the engines

are designed for high-power and high-speed

performance.

As an example, Figure 1 shows the difference

in the Indicated Mean Effective Pressure (IMEP)

between a motorcycle engine and an automobile

engine when running at a steady speed of 50km/

h. In this case, the Coefficient of Variation (COV)

calculated from the IMEP of the motorcycle engine

is 13.77 and that of the automobile engine is 1.27.

This reveals that the combustion stability of the

motorcycle engine is considerably lower than that of

the automobile engine.

Moreover, as a complete vehicle, a less expensive

system is required and there is only limited space

for the installation of hardware.

Furthermore, the system needs to detect misfiring

at a higher engine speed than that for automobiles.

Therefore, it is considerably difficult to adapt misfire

detection methods for automobiles to motorcycles.

This paper presents the results of our study into

misfire detection algorithms focusing on the uneven

firing of a motorcycle V-twin engine. Some of the

proposed algorithms use variation characteristics in

crank angular velocity for detection. Utilization of

these algorithms would not lead to any increase in

cost, but only limited installation space is necessary

since there are no requirements for any additional

Fig. 1 Difference in IMEP between motorcycle engine and automobile engine

IMEP

[bar

]

21.81.61.41.2

10.80.60.40.2

00 100

Cycle [-]

Motorcycle Engineat NE=3,400 rpm, PM=37.0 kPa

Automobile Engineat NE=1,300 rpm, PM=37.3 kPa

200 300

IMEP

[bar

]

87.87.67.47.2

76.86.66.46.2

60 100

Cycle [-]200 300

※ Content reprinted from SAE Technical Paper 2017-32-0052, © SAE International. Reprinted with permission. Further distribution of this material is not permitted without prior permission from SAE International.

－26－

Research on Misfire Detection Algorithms for Motorcycle Engine Firing at Uneven Intervals

- Cyl. #1 : Expansion stroke, Cyl. #2 : Suction

stroke

Figure 4 shows the relationship between the

missing tooth sections and the stroke timing of each

cylinder.

2.3. Misfiring Method

Misfi re generat ion sof tware was designed

and integrated in the ECU in order to generate

intentional misfire. This software generates an

ignition cut-off at equal intervals by presetting the

target cylinder, the total number of ignition events

and the number of times for misfire generation. For

example, when cylinder #1 is set to generate one

misfire for each ten combustion cycles, nine ignitions

and one ignition cut-off are repeated alternately.

The test engine adopted a twin-spark firing

system. Therefore, when misfire was generated,

ignition cut-off was executed for both plugs on the

cylinder.

Table 1 Specification of Test Engine

Fig. 2 Mounted position of trigger wheel

Displaced Volume 1,301 ccNumber of Cylinders 2 cylindersEngine Architecture V-type (Angle of the V : 75 deg)Phase Difference Cyl. #1 to Cyl. #2 : 435 degStroke 71 mmBore 108 mmCompression Ratio 13 : 1Valve Train Layout DOHC 4 valvesCooling Type Water cooledMaximum Output 118 kw / 8,750 rpmMaximum Torque 140 Nm / 6,750 rpm

Fig. 3 Timing teeth

Pick-Up Sensor

Timing Teeth(Periphery of

Trigger Wheel)

Crank ShaftAxis

Pick-Up Sensor

Timing Teeth(Periphery of

Trigger Wheel)

Fig. 4 Relationship between missing tooth sections and stroke timing

TDC / BDC

Engine stroke

Output ofPick-Up sensor

Exhaust Compression Exhaust

#2 Cyl.

#1 Cyl.

#1 Cyl.Compression

TDC

ExhaustCompressionSuction Compression

Missing toothsection

(45 deg)

Missing toothsection

(45 deg)

#2 Cyl.Compression

TDC

ExpansionSuction

SuctionExpansion

3. Pre-testing Results

In this research, misfire detection algorithms

using crank angular velocity focused on the fact

that when misfire occurs, no combustion energy

is generated, and thus the crank angular velocity

decelerates. Incidentally, the teeth intervals and

missing teeth of the trigger wheel influenced the

linearity of the crank angular velocity calculation

and affected misfire detection accuracy. The aim

of this research was to create an algorithm misfire

detection with a trigger wheel missing 2 teeth at 15

degree intervals as installed in the test vehicle.

In addition, in conventional misfire detection

algorithms using crank angular velocity, digital

filtering calculations have been often used to extract

characteristic frequency components at misfiring or

to remove components unnecessary for detection.

In order to design such filtering calculations, it is


－27－


TechnicalPapers

Fig. 5 Difference of IMEP value and crank angular velocity variation

necessary to check the frequency components of

angular velocity by normal combustion or misfiring.

Prior to the construction and evaluation of misfire

detection algorithms, the following two items were

pretested.

1. Confirmation of lack of combustion and crank

angular velocity variation

2. Frequency analysis of crank angular velocity

3.1. Confirmation of Lack of Combustion and

Crank Angular Velocity Variation

Purpose:

The purpose of this first stage of testing is to

confirm lack of combustion by misfire generation

and to also confirm that decrease of crank angular

velocity due to lack of combustion can be observed

at 15-degree intervals using the trigger wheel

equipped with missing teeth and the pick-up sensor.

Method:

With the test vehicle secured to a chassis

dynamometer, misfire was generated by software.

The lack of combustion was confirmed from analysis

of the IMEP and angular velocity variation was

calculated from the output of the pick-up sensor. The

same test process was carried out for each cylinder.

Result:

Due to misfiring, IMEP decreased greatly. From

this result, it was evident that lack of combustion

was occurring by misfiring. Additionally, at the

same time with IMEP decrease, a decrease in crank

angular velocity was also observed. Figure 5 shows

the difference of IMEP value and crank angular

velocity variation at NE=1,400rpm and 9,500rpm.

Conclusion/Decision:

A lack of combustion occurred due to misfiring.

At that time, decrease of crank angular velocity

was confirmed by calculations using the trigger

wheel with two missing teeth at 15-degree intervals.

Although this tendency was found to be weaker

1,400 rpm 9,500 rpm

Cyl

. #1

Mis

firin

gC

yl. #

2 M

isfir

ing

-40

-30

-20

-10

0

10

20

30

40

0 90 180 270

Cra

nk A

ngul

ar V

eloc

ity (r

ad/s

)

Crank Angle (deg)

-40

-30

-20

-10

0

10

20

30

40

0 90 180 270

Rel

ativ

e C

rank

Ang

ular

Vel

ocity

(r

ad/s

)

Crank Angle (deg)

Cyl. #2 Comp.TDC Cyl. #2 Comp.TDC

-40

-30

-20

-10

0

10

20

30

40

0 90 180 270 360

Rel

ativ

e C

rank

Ang

ular

Vel

ocity

(r

ad/s

)

Crank Angle (deg)

Cyl. #1 Comp.TDC-40

-30

-20

-10

0

10

20

30

40

0 90 180 270 360

Rel

ativ

e C

rank

Ang

ular

Vel

ocity

(r

ad/s

)

Crank Angle (deg)

Cyl. #1 Comp.TDC

Normal combustionIMEP = 1.03

MisfiringIMEP= -0.56


MisfiringIMEP = -0.71





when the NE was higher, as a result of these tests,

misfire detection using the trigger wheel and pick-up

sensor was judged possible for this test engine.

3.2. Frequency Analysis of Crank Angular

Velocity

Purpose:

The purpose of this analysis was to check

frequency components of angular veloci ty in

test engine for reference in the design of misfire

detect ion algori thms and also to confirm the

characterist ics of an uneven fir ing engine by

comparing with frequency analysis results from an

even firing engine.

Method:


dynamo, angular velocity calculated by the output of

the pick-up sensor was analyzed using a Fast Fourier

Transform (FFT) algorithm in post-processing. Also,

the analysis results were compared with results from

a comparison engine (even firing) which had been

preliminarily tested. Table 2 shows the specifications

of the comparison engine.

Results:

As a result of the FFT analysis of angular

velocity of the test engine, i t was found that


－28－


occurred when misfiring. Results of the FFT analysis

is shown in Figure 7.

Conclusion/ Decision;

In the test engine, the 0.5th order of crankshaft

rotation was the basic frequency component in both

normal combustion and misfiring. In addition, a

distinct frequency component occurred when there

was misfiring in the comparison engine (even firing),

but this did not occur in the test engine (uneven

firing). The reason is assumed to be that the crank

angular velocity variation patterns from combustion

for each cylinder are different.

Table 2 Specification of comparison engine

Fig. 6 Result of FFT analysis (test engine)

Fig. 7 Result of FFT (Comparison engine)

Normal combustion

Frequency (1/deg)

#1 Cyl. Misfiring

Pow

er S

pect

rum

0 0.005

120

100

80

60

40

20

0

Pow

er S

pect

rum

120

100

80

60

40

20

0

0.01 0.015 0.02 0.025 0.03

Frequency (1/deg)0 0.005 0.01 0.015 0.02 0.025 0.03

0.5th order

0.5th order

1st order

1st order

#2 Cyl. Misfiring

Pow

er S

pect

rum

120

100

80

60

40

20

0

Frequency (1/deg)0 0.005 0.01 0.015 0.02 0.025 0.03

0.5th order

1st order

Normal combustion

Frequency (1/deg)

50% Misfiring

Pow

er S

pect

rum

0 0.005

60

50

40

30

20

10

0

Pow

er S

pect

rum

60

50

40

30

20

10

0

0.01 0.015 0.02 0.025 0.03

Frequency (1/deg)0 0.005 0.01 0.015 0.02 0.025 0.03

0.5th order

0.5th order

0.25th order

Displaced Volume 109.1 ccNumber of Cylinders 1 cylinderStroke 55.6 mmBore 50.0 mmCompression Ratio 9 : 1Valve Train Layout OHC 4 valvesCooling Type Air cooledMaximum Output 6.4 kw / 7,500 rpm Maximum Torque 9.36 Nm / 5,500 rpm

the 0.5th order of crankshaft rotation is the basic

frequency component. Additionally, even if either

of the cylinders was misfiring, there was no distinct

frequency component. The result of FFT analysis is

shown in Figure 6.

O n t h e o t h e r h a n d , t h e r e s u l t s f r o m t h e

comparison single cylinder engine show that the 0.5th

order of crank-shaft rotation is the basic frequency

component. Additionally, a distinct frequency

component at the 0.25th order of crankshaft rotation

4. Detection Algorithms

In this research, four algorithms were evaluated

in parallel and compared for accuracy. These

algorithms were composed of four parts, and the

input used is the time between the edges of pulses

calculated from the output of the pick-up sensor.

Misfire detection was performed by comparing

the detection index parameter value calculated from

each algorithm with a preset threshold value. Figure

8 shows the construction of each algorithm. The


－29－


TechnicalPapers

a decrease in misfire detection accuracy. In order

to prevent such a decrease, a moving average

calculation, a kind of an LPF calculation, was

adopted. Since moving average calculation can be

applied to the number of data, crank angle, and not

to time, it is possible to remove an intended high

frequency component regardless of the speed of the

crank angular velocity. Incidentally, each detection

index parameter calculation of Part 4 was premised

by monitoring the variation pattern of crank angular

velocity due to combustion. Therefore, it is not

preferable to remove components due to combustion

by moving average calculation. As a result of the

pre-testing, the component due to combustion was a

component at the 0.5th order of crankshaft rotation.

Therefore, in this research, by adopting a moving

average calculation of six data, which constitutes a

90-degree section, only values below that of the 4th

order of the component of rotation were taken into

account, and thus high frequency components were

removed.

The moving average value of crank angular

velocity (ωMAn) was calculated by equation 3.

61

MAn Σ=ω CRKnω0

i = –5 (3)

This calculation was applied to all algorithms.

As an example, Figure 9 shows the calculation

results of ωCRKn and ωMAn for misfiring of the cylinder

#1 at NE=5,000rpm. The high frequency components

were removed while leaving the characteristic

frequency components due to combustion.

4.3. Part 3: Identifying Calculation

In an uneven firing engine, the variation patterns

of crank angular velocity show large differences

for each cyl inder. This ca lcula t ion a imed a t

identifying the characterizing portion of combustion

corresponding to the crank angular velocity variation

detail specification of each algorithm is described

later in this paper.

4.1. Part1: Crank Angular Velocity Calculation

Normally, the pick-up sensor would generate a

sinusoidal wave rising every 15 degrees. When the

time taken to rotate 15 degrees is t, the average

c rank angula r ve loc i ty ω 15 ( rad / s ) would be

calculated by equation 1.

12t15 =ωπ

(1)

In the missing tooth section of the trigger wheel,

a sinusoidal wave was generated at 45-degree

intervals. The crank angular velocity ω 45 (rad/s) at

that time was calculated by equation 2, and this

value was generated three times continuously.

4t45 =ωπ

(2)

Calculated crank angular velocity was generated

in the next part as continuous data (ω CRKn). This

means that continuous 24 data per crank rotation

was the output for the next part.

This calculation was applied to all algorithms.

4.2. Part 2: Low Pass Filter (LPF) Calculation

Unnecessary high frequency components were

included in the crank angular velocity as calculated

in Part 1, and these components may have caused

Fig. 8 Construction of algorithms

DifferenceCalculation

AccumulatingCalculation

Crank Angular Velocity Calculation

Time between crank pulse edge (ms)



#1 Calibrated #2 Calibrated

Input

Misfire Detection Algorithms

Algorithm 1 Algorithm 4Algorithm 3Algorithm 2

LPF Calculation

RelativeAngular Velocity

Calculation


Calculation


Calculation

Identifying Calculation

Part 1

Part 2

Part 3

Part 4


－30－


for each cylinder and formulating it so that the

combustion by each cylinder could be regarded as

equivalent. With this calculation, we were able to

bring the variation pattern closer to that of an even

firing engine.

In this calculation, the following six points, P1 to

P6, were extracted as combustion characteristics for

one cycle (720-degree crank angle). The compression

TDC points were extracted as the reference points

for the start of combustion, and the points during

the expansion stroke and the exhaust stroke were

extracted as points where a large difference occurs

between normal combustion and misfiring.

P1: Cyl. #1 Compression TDC

P2: Cyl. #1 During expansion stroke

P3: Cyl. #1 During exhaust stroke

P4: Cyl. #2 Compression TDC

P5: Cyl. #2 During expansion stroke

P6: Cyl. #2 During exhaust stroke

The ωMAn value at these six points was defined as

a continuous signal of ω idn.

Figure 10 shows ω MAn and extraction points

with cylinder #1 misfiring at NE=5,000rpm as an

example.

This calculation was applied to algorithm 1 and

algorithm 2.

4.4. P a r t 4: D e t e c t i o n I n d e x P a r a m e t e r

Calculation

Algorithm 1

In Part 3, in order to regard the combustion

of each cylinder as equivalent, the ω idn value was

calculated by formulating the crank angular velocity.

In this algorithm, the value is compared with the

angular velocity at the same stroke with the previous

ignit ion cylinder to calculate detection index

parameters. That is, since the previous combustion

and the current combustion are compared, the

detection index parameter value in the case of

misfiring is greatly reduced.

The detect ion index parameters using this

algorithm (DIP1) were calculated by equation 4.

DIP1 = idn –ω idn–3ω (4)

Furthermore, the ω idn values calculated at the

timing of P1 to P3 were used for cylinder #1 and

in the same at the timing of P4 to P6 are used for

misfire detection in cylinder #2.

Algorithm 2

Firs t ly, the re la t ive angula r ve loc i ty was

calculated based on the compression TDC of each

cylinder for the ω idn value in Part 3. For cylinder #1,

the extracted angular velocity at P1 (ω idP1) is used

as a reference, the relative angular velocity at P2

(ω RidP2) is calculated by equation 5 and the same at

P3 (ω RidP3) is calculated by equation 6.

Fig. 9 Calculation results of ωCRKn and ωMAn

Cra

nk A

ngul

ar V

eloc

ity (r

ad/s

)

545

550

540

535

530

525

520

515

510

505

500

Number of Data200 250 300 350 400 450 500

Misfiring on #1 Cyl.

MAnωCRKnω

Fig. 10 ωMAn and extraction points

MA

n (ra

d/s)

530

535

525

520

515

510

505

Number of Data230 250240 270260 280 300 310290 320 340330

#1 Cyl. MisfiringP5 P5

P4P4

P1P1P6

P6

P2

P2

P3

P3ω


－31－


TechnicalPapers

= –idP2ωRidP2ω idP1ω (5)


In the same manner, for cyl inder # 2, the

extracted angular velocity at P4 (ω idP4) was used as

a reference, the relative angular velocity at P5 (ω RidP5)

was calculated by equation 7 and the same at P6

(ω RidP6) was calculated by equation 8.



Then, relative angular velocities based on the

compression TDC timing of each cylinder were

integrated so as to be used as misfire detection index

parameters. The misfire detection index parameter

(DIP21) for cylinder #1 was calculated by equation 9

and the same for cylinder #2 (DIP22) was calculated

by equation 10.

+RidP2 RidP3ω ωDIP21 = (9)

+RidP5 RidP6ω ωDIP22 = (10)

Algorithm 3

For this algorithm, the ωMAn value calculated in

Part 2 was used as is. At first, the relative angular

velocity was calculated based on the compression

TDC of each cylinder for the ω MAn value. The

extracted angular velocities at the cylinder #1

compression TDC (ω MA1CT) and at the cylinder #2

compression TDC (ωMA2CT) were used as references,

and the relative angular velocities (ω RMAn) were

calculated by equations 11 and 12.

If 1CT + 1 ≤ n < 2CT: –RMAn MA1CTω MAnω ω=

(11)

If 2CT + 1 ≤ n < 1CT: –RMAn MA2CTω MAnω ω=

(12)

The ω RMAn value was integrated in the section

up to the point before the compression TDC of the

cylinder firing next, which was used as the detection

index parameter. The misfire detect ion index

parameter for cylinder #1 (DIP31) was calculated by

equation 13, and the same for cylinder #2 (DIP32)

was calculated by equation 14.

RMAnΣ ω2CT–1

i = 1CT+1

DIP31 = (13)

RMAnΣ ω1CT–1

i = 2CT+1

DIP32 = (14)

The schema of these calculations is shown in

Figure 11.

Algorithm 4

The basic logic of this algorithm was the same

as algorithm 3. Only the integration section of the

ω RMAn value was changed so as to be 180 degrees

from compression TDC of each cylinder. The reason

for that was that the ω RMAn value at the section

before the compression TDC of the next firing

cylinder included many disturbance components

such as friction that cannot be attributed to the

combustion state, and large deviations in the angular

velocities.

The misfire detect ion index parameter for

Fig. 11 Calculation schema of algorithm 3

MA

n (ra

d/s)

530

535

525

520

515

510

Number of Data280 300290 320310 330 340 350 370360

#1 Cyl. Misfiring

Negativevalue

Negativevalue

Positivevalue

Positivevalue

Positivevalueω

DIP31

DIP31

DIP32

DIP32


－32－


cylinder #1 (DIP41) was calculated by equation 15

and the same for cylinder #2 (DIP42) was calculated

by equation 16.

RMAnΣ ω1CT+12

i = 1CT+1

DIP41 = (15)

RMAnΣ ω2CT+12

i = 2CT+1

DIP42 = (16)

The schema of these calculations is shown in

Figure 12.

Fig. 12 Calculation schema of algorithm 4

MAn

(rad

/s)

530

535

525

520

515

510

Number of Data280 300290 320310 330 340 350 370360

#1 Cyl. Misfiring

Negativevalue

Positivevalue

Positivevalue

Positivevalueω

DIP41

DIP41

DIP42

DIP42

5. Results of Simulations

5.1. Evaluation Method

5.1.1. Evaluation Flow

The signal output from the pick-up sensor was

entered into an external measurement instrument and

measured. After that, the measured value was entered

into a data analysis simulation model on a computer.

There were two parts of the simulation model. The

first part calculated the crank angular velocity from

the pick-up sensor output signal, and the second part

applied the four misfire detection algorithms at the

same time in parallel. For the next step, detection

performance was compared and evaluated based on

the behavior of the parameters calculated by each

detection algorithm. An outline of the evaluation

flow is shown in Figure 13.

One th ing that must be ment ioned is that

sampling rate for the output of pick-up sensor is

substantially influenced by the linearity of crank

angular velocity calculation as well as the tooth

intervals and missing teeth of the trigger wheel.

Subsequently, these factors affect the accuracy of

the misfire detection.

In this evaluation, in order to acquire analysis

results closer to a real system, the recording rate

of the measurement instruments for pick-up sensor

output was set to 1 MHz rate which conforms to the

read interval of the ECU.

5.1.2. Index for Evaluation

The Signal-to-Noise ratio (SNR) was used as an

index for comparing misfire detection performance

for each algorithm. SNR was defined as the power

ratio of the signal and background noise. In this

evaluation, the detection index parameters calculated

by the detection algorithm at misfiring were set

as signal, and the parameters calculated at normal

combustion were set as noise. In addition, assuming

that the detection parameter value exhibits a normal

distribution, the SNR was calculated by equation 17.

Fig. 13 Outline of evaluation flow

TimingRotor

Sensor

Data AcquisitionSystem

AlgorithmSimulation Model

: Crank signal

Angular velocityCalculation Model

: Angular velocity

: Detection parameter

Vehicle

Measurement Instruments

Computer


－33－


TechnicalPapers

+nσ mσSNR = AVEn – AVEm

(17)

AVEn : Average of detection parameter with

combustion

AVEm : Average of detection parameter with

misfiring

σ n : Deviation of detection parameter with

combustion

σ m : Deviation of detection parameter with

misfiring

The higher the SNR value, the easier it is to

distinguish between the distribution of normal

combustion and that of misfiring. This facilitates

the setting of a misfire detection threshold for

satisfactory misfire detection performance. For

example, when the SNR=3.0, the detection accuracy

is calculated at approximately 99.8%.

5.2. Evaluation Targets


dynamometer, data from running tests were recorded

for the following two cases and the data generated

by the simulation were evaluated:

- Steady state: By fixed throttle angle; and

- Acceleration state: By wide throttle opening.

During the simulation, misfires were generated

at even intervals at the rate of once every ten

combustion cycles in one cylinder. The same tests

were conducted for both cylinders on the test engine

in a warmed-up state (TW ≥ 80 degC). Detection

index parameters at a non-intentional misfiring cycles

were classified as noise, and those at intentional

misfiring cycles were classified as valid signals for

analysis.

5.2.1. Measurement Points for Steady State Tests

Data was measured a t several load points

inside the detection area as defined by the OBD II

regulations of the European Union.

Testing points for cylinder #1 are shown in

Figure 14 as an example. The mark indicates the

gear for each driving cycle.

5.3. Evaluation Results

5.3.1. Steady State Test Results

The misfire detection SNR values for cylinder #1

at each load point are shown in Table 3 and those

for cylinder #2 are shown in Table 4. The algorithms

which achieved the highest SNR at each load point

are highlighted in green.

From these results, it was found that algorithm 4

Fig. 14 Cyl. #1 testing points for steady state tests

PM (k

Pa)

70

65

80

75

60

55

50

45

40

35

30

NE (rpm)

Detection Area3rd Gear6th Gear

Neutral4th Gear

1st Gear5th Gear

0 20001000 40003000 5000 6000 7000 100008000 9000

Fig. 15 ωMAn and DIP41 with Cyl. #1 misfiring in steady state

#1 Cyl. Misfiring points

DIP

41 (–

)

100

150

50

0

-50

Number of Cycle0 10 50403020 60 70 9080 100

530

540

520

510

500

Number of Data0 200015001000500 2500 3000 40003500 4500

MA

n (ra

d/s)

ω


－34－


5.3.2. Acceleration State Test Results

In this test, the 3rd gear was used to accelerate

with a wide-open throttle. For this acceleration, NE

was increased from 3,000rpm to 10,000rpm in 4.0

seconds.

The misfi re de tec t ion SNR va lues for the

acceleration state are shown in Table 5.

From these results, it was found that the three

algorithms excluding algorithm 1 have a high

detection performance for both cylinders, and this

performance was judged to be sufficient for practical

use.

Figure 16 shows the behavior of ωMAn and DIP41

with intentional misfiring of cylinder #1 for the

acceleration state. The DIP41 value drops sharply

according to the misfiring points in the same manner

as the results for the steady state tests.

Table 3 SNR values for Cyl. #1

Table 4 SNR values for Cyl. #2

Algorithm 1 Algorithm 2 Algorithm 3 Algorithm 4

1400 N 4.51 4.52 4.60 4.202000 N 5.80 6.07 6.10 5.382000 5 2.16 2.78 2.94 4.682000 6 1.00 1.52 1.61 2.663000 N 9.50 10.81 11.15 9.813000 5 1.52 2.25 2.27 4.393000 6 3.10 5.09 5.14 7.844000 1 4.50 5.84 5.56 6.454000 5 2.37 3.72 3.64 6.694000 6 1.69 2.84 2.84 5.315000 1 5.58 7.24 6.87 7.905000 5 2.77 4.16 4.13 6.565000 6 3.15 4.81 4.82 6.176000 1 6.92 8.29 8.79 9.516000 5 3.28 4.76 4.78 6.956000 6 3.58 4.86 5.23 6.047000 1 8.01 8.51 9.45 9.647000 5 3.77 5.27 5.51 8.527000 6 2.80 3.71 4.36 5.998000 1 8.35 10.04 9.52 11.118000 4 4.50 5.60 6.12 8.848000 5 3.95 5.37 5.75 8.899000 1 8.22 8.74 8.56 8.889000 4 4.83 6.25 6.85 9.259000 5 4.45 6.14 6.66 10.479500 1 8.74 9.15 8.32 8.779500 3 5.73 7.45 8.16 8.379500

40.9538.0540.7544.3933.9538.3544.0437.0140.3048.4340.2145.1162.6543.4155.6673.6346.6165.4779.8549.8161.7074.4753.0169.7278.9555.0060.0474.84 4 5.50 7.09 7.59 9.60

GearS/N [ratio] by each algorithm

NE [rpm] PM [kPa]

Algorithm 1 Algorithm 4Algorithm 3Algorithm 2

1400 N 4.62 4.48 4.63 4.242000 N 5.54 5.47 5.68 5.142000 5 2.82 4.32 4.26 4.712000 6 2.55 3.82 3.89 4.453000 N 9.35 9.38 9.70 8.723000 5 4.26 9.21 9.41 9.833000 6 3.55 4.58 4.31 6.394000 1 5.86 7.88 8.14 7.634000 5 4.09 6.36 6.39 6.994000 6 3.37 7.51 7.54 8.185000 1 7.65 11.00 11.55 11.135000 5 3.82 6.78 6.72 7.355000 6 3.42 6.69 6.56 7.326000 1 6.16 7.70 8.20 7.636000 5 3.47 6.30 6.00 7.116000 6 3.23 7.29 7.29 7.697000 1 7.32 10.30 10.51 11.217000 5 3.85 7.46 7.16 8.567000 6 2.93 5.54 5.46 5.918000 1 7.39 9.37 9.29 9.048000 4 4.51 7.52 7.12 8.308000 5 4.18 7.10 7.03 7.999000 1 9.18 10.22 9.33 9.589000 4 5.50 9.56 9.03 10.239000 5 4.44 7.43 7.14 8.559500 1 7.73 8.53 7.89 8.829500 3 6.74 9.95 10.42 9.529500

39.6936.9940.6042.9932.6237.6243.6936.9040.5248.3939.6246.4063.2042.1354.9774.6345.0164.8579.9949.2960.8974.5852.9969.7979.4954.1860.4574.88 4 5.14 6.92 7.53 7.04

NE [rpm] PM [kPa] GearS/N [ratio] by each algorithm

had a high detection performance for both cylinders,

and i t was judged that this performance was

sufficient for practical use.

Figure 15 shows behavior of ωMAn and DIP41 with

intentional misfiring on cylinder #1 at NE = 5,000

rpm with PM = 45.11 kPa. The DIP41 value drops

sharply according to the misfiring points.

Table 5 SNR values for Cyl. #1 and #2

Algorithm 1 Algorithm 2 Algorithm 3 Algorithm 4

#1 2.66 3.41 3.48 3.56

#2 4.03 4.18 4.19 3.62

3000to

10000

NE [rpm] Gear MisfiringCyl.

S/N [ratio] by each algorithm

3

Fig. 16 ωMAn and DIP41 with Cyl. #1 misfiring in acceleration state

#1 Cyl. Misfiring points

DIP

41 (–

)

400

800

600

200

0

-200

Number of Cycle0 15010050 200

800

1200

1000

600

400

200

0

Number of Data0 8000600040002000 10000

MA

n (ra

d/s)

ω


－35－


TechnicalPapers

6. Result of Vehicle Tests

6.1. Evaluation Method

Based on the results of the simulations, algorithm

4 was judged to have the highest potential, and

thus it was entered into the ECU. During these

evaluations, misfire was detected when the detection

index parameter value was lower than the detection

threshold, a value which, for prototype purposes, is

a constant value for the all load points.


dynamometer, data from running tests of the vehicle

was recorded from the ECU and the detection

accuracy of the a lgor i thm was subsequent ly

evaluated.

6.2. Evaluation Targets

The test data for each of the three phases in the

World Motorcycle Test Cycle (WMTC) Class 3-2

test cycle (Figure 17) with the engine in a warmed-

up state (TW ≥ 80 degC) were measured separately.

In each phase, the detected misfiring rate by

our algorithm was compared with the intentional

misfiring rate in the detection area. In addition, the

value calculated by dividing the sum of the number

of undetectable, misdetection or natural misfires

by the number of cycles in the detection area was

defined as the error rate, which was also evaluated.

The schema of the evaluation targets is shown in

Figure 18.

Fig. 17 WMTC Class3-2 test cycle Fig. 18 Schema of evaluation targets

Veh

icle

Spe

ed (k

m/h

)

40

140

120

100

80

60

20

0

Time (s)0 800600200 400 1200 1400 1600 18001000

Phase 3Phase 2Phase 1Cycles in detection area

I-D: UndetectableI: Intentional misfire

I ∩ D: Correct detection

for intentional misfire

D-I: Misdetection or natural misfire

D: Detected misfire

During the vehicle testing, misfire was generated

at even intervals a t a ra te of once every ten

combustion cycles in one cylinder in the same

manner as the earlier simulations. The same testing

was conducted for each cylinder.

6.3. Evaluation Results

The evaluation results for cylinder #1 misfiring at

each phase are shown in Table 6, and results for the

same for cylinder #2 are shown in Table 7.

In the case o f cy l inder #1, a lmos t a l l o f

intentional misfire could be detected. Although there

were a noticeable number of misdetection or natural

misfires, the number of undetectable misfires was

small, allowing for an error rate of less than 1% in

all phases.

In the case of cylinder #2, although the results

were inferior to those of cylinder #1, almost all of

the intentional misfires could be detected. Although

there were a noticeable number of undetectable

misfires, the number of misdetection or natural

misfires was small, allowing for an error rate of less

than 0.5% in all phases.

As mentioned above, the detection thresholds

for each cylinder were set at a constant value

for all load points for prototype purposes. From

these results, it was evident that by optimizing the

thresholds based on the number of undetectable and

of misdetection misfires, detection accuracy can still

be improved further.


－36－


Table 6 Detection accuracy of misfiring in Cyl. #1

Number of Cyclesin Detection Area(times)

Number ofIntentional Misfiring(times)

IntentionalMisfiringRate(%)

Numberof Correct Detection(times)

Numberof Undetectable(times)

Numberof Misdetectionor Natural Misfiring(times)

System DetectedRate(%)

Error Rate(%)

10.72 10.19

479373

0 0 2

0.94 0.85 0.29

7,790 10,969 17,084

9.63 9.87 9.93

10.56

750 1,083 1,696

750 1,083 1,694

Phase 1 Phase 2 Phase 3

Number of Cyclesin Detection Area(times)

Number ofIntentional Misfiring(times)

IntentionalMisfiringRate(%)

Numberof Correct Detection(times)

Numberof Undetectable(times)

Numberof Misdetectionor Natural Misfiring(times)

System DetectedRate(%)

Error Rate(%)

9.63 9.70

61212

0 33 55

0.20 0.43 0.36

6,124 10,461 17,026

9.16 9.83 9.99

9.36

561 1,028 1,701

561 995 1,646

Phase 1 Phase 2 Phase 3

Table 7 Detection accuracy of misfiring in Cyl. #2

7. Summary

Misfire detection algorithms using crank angular

velocity fluctuation which can be applied to uneven

firing V-twin engines was proposed. For calculation

of the crank angular velocity, output from pick-up

sensor with a trigger wheel with two missing teeth

at 15-degree intervals was used.

For proposed algorithms, the misfire detection

accuracy was evaluated and compared by control

simulation using various measurement data as input.

Subsequently, the algorithm which showed higher

potential in simulation was integrated in an ECU

and it was confirmed by vehicle testing that misfire

detection was almost possible.

In the future, further studies especially focused

on the following issues are needed in order to

establish a viable misfire detection system for

motorcycles:

- measures to set the optimum detection threshold;

- detection performance for different misfire

occurrence patterns;

- measures for abnormalities in one plug in a twin

spark engine;

- de t ec t i on pe r fo rmance when d i s t u rbance

influences such as rider operation and travel on

rough surfaces; and

- application to different engine types.

References

(1) O f f i c i a l J o u r n a l o f E u r o p e a n U n i o n ,

“REGULATION (EU) No 168/2013 OF THE

EUROPEAN PARLIAMENT AND OF THE

COUNCIL of 15 January 2013”

(2) O f f i c i a l J o u r n a l o f E u r o p e a n U n i o n ,

“COMMISSION DELEGATED REGULATION

(EU) No 44/2014 of 21 November 2013”

(3) H i royasu , H . , “Easy unde r s t and In t e rna l


－37－


TechnicalPapers

combus t ion eng ines ,” Nis s in pub l i sh Co .

1996.4.30 6th Edition. (In Japanese), ISBN

4-8173-0075-0/0053-0

(4) K a n e k o , Y. , “ F u n d a m e n t a l s o f I n t e r n a l

Combustion Engine,” Sankaido Co. 1997.1.10

First Edition. (In Japanese), ISBN4-381-10097-2

C3053.

(5) Ko b a t a k e , H . , H a m a d a , N . , Ta m u r a , Y. ,

“Introduction to Signal Processing,” Corona

publ ishing Co. 2009.5.25 2nd Edi t ion. ( In

Japanese), ISBN4-339-03365-6.

(6) Hsien-Chi, T., Bo-Yu, G., Ming-Hao, C., Bo-

Chiuan, C. et al., “Misfire Diagnostic Strategy

for Motorcycles,” SAE Technical Paper 2013-32-

9058, 2013, doi:10.4271/2013-32-9058.

(7) Katsunori, T., “Misfire detection system for

internal combustion engine,” Japan Patent 2017-

139025.

(8) Katsunori, T., Yuki, M., “Misfire detection system

for internal combustion engine,” Japan Patent

2017-139027.

Acknowledgments

The author would like to express his gratitude to

KTM A.G. R&D EMS team who supported with the

provision of a test engine and a vehicle.

All testing data acquisition in this research are

from the work package of the misfire detection

development project in Keihin Corporation, and the

author’s colleague, Yuki Morita contributed greatly

in his support of this project.

Remarks

It is a great honor that this paper was chosen

for the Best Presentation Award in the 23rd Small

Engine Technology Conference (SETC) in November

15-17, 2017, Jakarta.

Abbreviations

OBD on board diagnosis

ECU engine control unit

CO carbon monoxide

HC hydrocarbon

NOx nitrogen oxides

CO2 carbon dioxide

IMEP indicated mean effective pressure

COV coefficient of variation

DOHC double overhead camshaft

OHC overhead camshaft

TDC top dead center

FFT fast fourier transform

LPF low-pass filter

SNR signal-to-noise ratio

NE number of engine revolution [rpm]

PM pressure of manifold [kPa]

TW engine coolant temperature of water cooled

engine [degC]

Author

K. TASAKI

Do emission regulations and on board diagnosis

requirements make motorcycles fascinating? The

answer seems to be NO. Then, will motorcycles remain

as fascinating without them? The answer is NO.

To make it possible for motorcycles to meet present

day demand, and to sustain motorcycles as fascinating,

motorcycle enthusiasts have to fulfill their own

missions. If I can carry out a part of their missions, it

would be a great pleasure for me. (TASAKI)


Technical Paper Research on Misfire Detection Algorithms ...and affected misfire detection accuracy....

Documents

Transcript of Technical Paper Research on Misfire Detection Algorithms ...and affected misfire detection accuracy....