发动机燃烧优化4

400 Commonwealth Drive, Warrendale, PA 15096-0001 U.S.A. Tel: (724) 776-4841 Fax: (724) 776-5760 Web:
SAE TECHNICAL PAPER SERIES
2006-01-0400
Analysis of Combustion Knock Metrics in
Spark-Ignition Engines
Jeffrey D. Naber and Jason R. Blough
Michigan T echnological University
Dave Frankowski, Monroe Goble and John E. Szpytman
Motorola Inc, Automotive
Reprinted From: Electronic Engine Controls 2006
(SP-2003)
2006 SAE World Congress
Detroit, Michigan
April 3-6, 2006
The Engineering Meetings Board has approved this paper for publication. It has successfully completed SAE's peer review process under the supervision of the session organizer. This process requires a minimum of three (3) reviews by industry experts.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of SAE.
For permission and licensing requests contact:
SAE Permissions
400 Commonwealth Drive
Warrendale, PA 15096-0001-USA
Email:permissions@
Tel:724-772-4028
Fax:724-776-3036
For multiple print copies contact:
SAE Customer Service
Tel:877-606-7323 (inside USA and Canada)
Tel:724-776-4970 (outside USA)
Fax:724-776-0790
Email:CustomerService@
ISSN 0148-7191
Copyright © 2006 SAE International
Positions and opinions advanced in this paper are those of the author(s) and not necessarily those of SAE. The author is solely responsible for the content of the paper. A process is available by which discussions will be printed with the paper if it is published in SAE Transactions.
Persons wishing to submit papers to be considered for presentation or publication by SAE should send the manuscript or a 300 word abstract to Secretary, Engineering Meetings Board, SAE.
Printed in USA
2006-01-0400
Analysis of Combustion Knock Metrics
in Spark-Ignition Engines
Michigan Technological University
Dave Frankowski, Monroe Goble and John E. Szpytman
Motorola Inc, Automotive ABSTRACT
Combustion knock detection and control in internal com-bustion engines continues to be an important feature in engine management systems. In spark-ignition engine applications, the frequency of occurrence of combustion knock and its intensity are controlled through a closed-looped feedback system to maintain knock at levels that do not cause engine damage or objectionable audible noise.
Many methods for determination of the feedback signal for combustion knock in spark-ignition internal combus-tion engines have been employed with the most com-mon technique being measurement of engine vibration using an accelerometer. With this technique single or multiple piezoelectric accelerometers are mounted on the engine and vibrations resulting from combustion knock and other sources are converted to electrical sig-nals. These signals are input to the engine control unit and are processed to determine the signal strength dur-ing a period of crank-angle when combustion knock is expected. As the accelerometer detects a number of sources of vibrations in addition to the desired vibration from knock, the signal quality varies significantly from engine to engine, cylinder to cylinder, and over the op-erating conditions of the engine.
To evaluate the effectiveness and accuracy of knock detection via accelerometers, a reference system is commonly employed. One of the most common refer-ence metrics is the signal strength of the combustion pressure over the appropriate frequency range as meas-ured with in-cylinder pressure transducers.
This analysis examines both cylinder pressure and ac-celerometer based knock intensity metrics, where the pressure based knock intensity metric is used as the reference measure. D istributions of the knock metrics over a number of engine cycles for various engine speeds, loads, cam timings, and knock levels are meas-ured and fit to a log-normal model distribution. The log- normal model is shown to provide a good fit to the measured distribution and also captures the characteris-tics of the distribution to include skewness and peak-ness. In addition the accelerometer intensity metric is correlated to the reference pressure intensity metric. The result of this correlation provides the coefficient of de-termination, which is used as a measure of the acceler-ometer intensity metric’s ability to indicate knock. The effects of the distribution of the pressure intensity metric on the coefficient of determination are examined by ana-lyzing subsets of the distribution
INTRODUCTION
Accurate detection of combustion knock in spark-ignition engines continues to be an important application in en-gine management systems. Accurate detection and con-trol ensures that the engine is operating at its best fuel economy and minimizes exhaust temperatures caused by unnecessary retard of spark timing. The majority of applications use engine mounted accelerometers to measure vibrations transmitted from the in-cylinder pressure oscillations that result from combustion knock to the engine structure. Similar techniques of using ac-celerometers to detect combustion processes are also being investigated in compression ignition engines for injector and fuel control and on-board-diagnostics. Ac-celerometers may also offer methods of combustion feedback in homogeneous charge compression ignition (HCCI) and other promising modes of low temperature combustion.
Combustion knock in engines has received a great deal of study over the years and just a cursory examination of the literature will find many excellent investigations in-cluding the early work of the National Advisory Commit-tee for Aeronautics (NACA) [1-6] to determine its origin and characteristics. Obert [7] and Heywood [8] give syn-opses of knock in engines including their sensitivities to engine conditions and fuels and impact on performance and engine deterioration. D raper [1] and others [9-11] Jeffrey D. Naber and Jason R. Blough
Copyright © 2006 SAE International
have studied the oscillation and frequency content of knock using analytical, experimental and numerical methods. More recent discussions on knock sensors, signal processing and knock control as applied in cur-rent engine management systems can be found in refer-ences [12, 13, 14].
More to the thrust of this investigation, Xiaofeng et al [15] reviews a number of knock intensity metrics from cylinder pressure measurement. Arrigoni et al [17] dis-cusses a method designated the Integral of the Modulus of the Pressure Oscillation (IMPO) which has a similar foundation to the pressure intensity metric utilized in this analysis. Brunt et al [18] also provides a detailed analy-sis of knock intensity determination from cylinder pres-sure. In their work, the primary knock intensity metric is based upon a band-passed filtered peak pressure ob-served during a crank-angle period when combustion knock is expected to occur. They discuss the variability in knock and quantify the variability with cumulative dis-tribution functions. They do not however examine the probability distribution of knock nor do they discuss in detail the non-normal distribution aspects of knock in-tensity as investigated in this work. Based upon large sample sizes and analysis of sub-samples, Brunt et al. recommend a sample size of 1000 cycles of combustion data to accurately capture the knock intensity metric. Sinnerstad [16] examines the statistical nature of com-bustion knock in an SI engine as measured from the band-pass peak knock cylinder pressure. Sinnerstad utilizes a Gamma distribution to characterize the distri-bution of the peak pressures then applies a method of maximum likelihood estimation to determine the distribu-tion parameters. The distribution is then used to esti-mate the confidence interval of the mean peak knock intensity for a given sample size. There is however, only limited data presented the goodness of the distributions fit is not quantified. The statistical characterization utili-zation is focused on confidence interval estimation.
The works [15-18] examine the pressure based knock intensity metrics. There is however, significant interest in the accelerometer based knock intensity metrics as it is the knock accelerometer sensor that is used for the indi-cation of knock in the majority of engine management systems in use today.
D ues et al. [19] discuss the types of knock sensors available and provide recommendations to sensor loca-tions, but does not discuss quantitative methods of se-lecting locations. Scholl et al. [20] compare pressure and accelerometer signal metrics. Their comparison is based upon the energy of the signal in a crank-angle window. They point out that there are a number of sensitivities including frequency shifts due to charge temperature, air-fuel ratio, and exhaust gas recirculation. They do not provide a statistical analysis of the observed knock. Finally Bengisu [21] and Shi [22] discuss finite element methods to select accelerometer sensor locations. Their results depend upon the accuracy in which they have included the forcing functions for the induced pressure oscillations and other engine vibrations that deteriorate the sensor signals. Shi [22] employs a technique to in-clude random excitation for the noise sources. These techniques are helpful in determining potential sensor locations and minimizing the number of locations to test, but as of yet have not in most cases replaced direct measurement techniques with induced knock on a run-ning engine.
In this analysis, a signal processing method of band-pass filtering, rectifying, windowing, and integrating is applied to determine a measure of knock intensity (AI) from the accelerometer signal. This signal processing has been encoded into a number of Application Specific Integrated Circuits (ASIC’s) which are incorporated into Engine Control Units (ECU’s) for real-time signal proc-essing to provide feedback of the combustion knock level in the engine.
These accelerometer based indicators are deteriorated when other engine components’ vibrations introduce noise including those produced by the closing of valves. The deterioration of the signal is especially evident in V-8 and other high cylinder count engines at high engine speeds. This increased noise can lead to limited detec-tion capability, and thus it is important to be able to quantify the AI metrics response to knock. To quantify the accelerometer based knock intensity metric, a refer-ence metric must be utilized. A common measure of knock intensity in the test cell is obtained from in-cylinder pressure measurement utilizing a similar signal processing approach (band-pass filtering, rectifying, and integrating) to determine the pressure based knock in-tensity (PI).
In this study the accelerometer based knock intensity (AI) is correlated to the pressure based knock intensity utilizing linear regression analysis and determining the Coefficient of Determination (COD). The COD is a valu-able evaluation metric and can be used at multiple stages in engine development and calibration. D uring engine development, it is desired to design in and select the best location or locations for placement of the accel-erometer(s) to enable knock detection over the widest range of conditions possible. This is done by maximiz-ing the COD over the engine operating conditions of in-terest for knock detection.
This procedure can be performed via a sensor location study where a number of accelerometers are placed on the engine at locations defined by Finite Element Analy-sis (FEA) and by the judgment of experienced design and application engineers. As the experimental methods are expensive and come at a point in the design cycle where the opportunity for changes is often limited, the results should be as quantitative as possible to enable the engine team to make the correct choices for produc-tion.
EXPERIMENTAL SETUP
The testing for this study were conducted in an engine dynamometer at Motorola’s applications engineering facility in D earborn, MI. The engine dynamometer test cell has a 320 kW (450 hp) DC motoring dynamometer interfaced and controlled through an AVL Puma V dy-namometer controller. The test cell has speed, torque and transient control capability with engine monitoring as well as measurement and logging of all the standard engine parameters including torque, fuel consumption,pressures, temperatures and emissions.
The engine tested was a V6 3.0L gasolines port-fuel–injected,spark-ignited engine with intake and exhaust cam timing control (phase relative to crank). While the engine had both intake and exhaust variable cam timing (TCV) devices for phase control; in this study only the intake valve timing was changed with the range of intake valve closing (IVC) phase tested from 216 to 246q ATD C.The exhaust cam phase was fixed with an ex-haust valve closing (EVC) phase of 11.5q ATDC. Other specifications for the engine are shown in Table 1.Non-knocking test were conducted with Indolene with a Re-search Octane Number (RON) of 96.5 and a Motor Oc-tane Number (MON) of 87.5. The knock tests were con-ducted with Amoco Regular RON of 91.8 and MON of 84.9.
Table 1: Engine Specifications.
Figure 1shows a schem including the locations of the in-cylinder pressure trans-ducers (shown as PT in the diagram) and the locations of the three accelerometers (A1, A2, A3) on the engine. In addition the figure sh instrumentation utilized and selected portions of the wiring.
Kistler type 6125b piezoelectric cylinder pressure trans-ducers were mounted in the cylinder head adjacent to the spark plugs through 4valves to measure the in-cylinder pressure. The sen-sors were mounted flush with the combustion chamber in order to avoid pipe oscillations. The pressure trans-ducer signals were input to Kistler type 5010charge amplifiers to convert the charge to voltage.
atic of the experimental setup ows the the valve cover and between the
EMS Calibration Tool
Figure 1:Schematic of engine including locations of accelerometers and test instrumentation.
There were three accelerometers located on the engine as shown in Figure 1. Sensors A1 and A2were instru-ment grade Brüel & Kjær Type 4371piezoelectric accel-erometers. These were located midway down the block with sensor A1 located between cylinders C3 and C5 and sensor A2located between cylinders C1 and C3 as shown in the figure. The outputs of these instrument grade accelerometers were also input into Kistler type 5010charge amplifiers to convert their output to voltage. Prior to being installed on the engine,the accelerometer and charge amplifiers were calibrated as a pair using a Brüel &Kjær Type 4291 hand calibrator with an ampli-tude of 10 m/s 2 rms.
The third accelerometer, A3, is the production sensor in the production location,which is on top of the engine in the valley between the cylinder banks biased to the left hand side between cylinders 2and 4.The production sensor does not require an amplifier; its output was fed directly into the data acquisition system.
An optical encoder was attached to the crank-shaft at the front of the engine. The encoder outputs two pulse trains, the ‘z ’ pulse (once per revolution) and the ‘a ’pulse (75 times per revolution).
The voltages from the charge amplifiers from the 6 pres-sure transducers, the 2 instrument grade accelerome-ters, the voltage from the production accelerometer and the two encoder signals were input into a Nicolet Tech-nologies OD-200 Odyssey data acquisition system. This system has 16 differential simultaneously sampled 14 bit resolution A/D’s with 50 kHz anti-aliasing filters. Each channel was digitized at a sample rate of 200,000 sam-ples per second. Data was acquired for a minimum of 10
ed to a 0-720q crank-angle vector with cycle phase determined by
between the ‘a’ pulses. Utilizing this method the crank-angle resolution is de-
An Engine Control Unit (ECU) programmed with an En-
sure signal from each cylinder was routed through a band-pass filter with a bandwidth
hos-phor Oscilloscope (D PO) was utilized to visualize the and fuel were set back to the nominal val-ues. When all cylinders were checked at a given speed and load, the individual spark and fuel settings for the
bility of the accelerometer based knock intensity.
orner frequency of one er was set at 5 kHz while the other was set at 27 kHz. he two uppermost plots in Figures 2 and 3 show the cylinder pressure with the 27 kHz low pass filter applied
is due to pressure oscillations in the cylinder caused by
seconds for each test.
The data was extracted from the data acquisition system and converted to MathWorks Matlab ® format where the signal processing was completed. As part of this proc-essing, the encoder signals were convert
aligning the cylinder #1 pressure to the correct cycle. The crank-angle is interpolated between encoder pulses utilizing a specially developed low pass filter tuned to the engine speed and then spline fit
pendent upon engine speed. At the speeds tested, the resolution was [0.045 0.075 0.105] degrees at [1500, 2500, 3500] rpm respectively. Several internal studies have shown this method to be accurate to better than ¼ of a crank-angle degree.
With the time and crank-angle specified at each data point in the test, the analysis methods utilize combined time and angle based signal processing and logic.
gine Management System (EMS) control strategy and a dynamometer calibration was used with the correspond-ing calibration tool. This instrumentation is also shown as part of Figure 1. This system enabled the operator to control the load via the electronic throttle, fuel quantity and timing, spark timing, and cam phasing among other parameters.
To set test conditions, the dynamometer controller was adjusted to the desired engine speed set-point and throt-tle position was adjusted to the desired load while the intake valve timing was adjusted to the desired position. To set the knock level in the test, spark timing was ad-justed on an individual cylinder basis. To observe the peak knock level, the pres
from 5 kHz to 20 kHz, and then monitored for level of knock on an oscilloscope as shown in Figure 1. To en-hance the monitoring of the peak level, a D igital P variation in amplitude of the knock. The target level of knock was 100 – 150 kPa peak (15 - 22 psi). Starting from a nominal fuel and spark timing just below border-line knock, the spark was advanced and the fuel pulse width adjusted on each cylinder individually to obtain the target level of knock visually on the oscilloscope. The fuel and spark settings were then documented and the cylinder spark desired knock level for each cylinder were entered into the ECU calibration. After entering the individual spark and fueling parameters and waiting a short stabilization period, the cylinder pressure and accelerometer data were captured on the data acquisition system while the engine data was captured on the dynamometer control-ler.
This procedure was repeated for the test matrix listed in Appendix A. The test included non-knocking and knock-ing conditions at engine speeds of 1500, 2500, and 3500 rpm. In addition, multiple loads and IVC timings were investigated. Typically the engine speed would be tested to red-line; however, as this engine was to be used in further development and calibration work, the engine speed was limited to 3500 rpm.
SIGNAL PROCESSING FOR KNOCK INTENSITY
The collected pressure and accelerometer data was analyzed using custom analysis software. The data was first extracted and converted to Matlab ® format. The pressure and accelerometer signals were then filtered, rectified and integrated during specified crank-angle windows to produce knock intensity metrics. In this analysis the cylinder-pressure based knock intensity is used as the quantitative knock reference to judge the capa
Sample pressure data as a function of crank-angle for cylinder 4 from knocking test 54 (see Appendix A) at an engine speed of 2500 rpm, a normalized load of 1.0 and an IVC cam timing of 236q are shown in Figures 2 and 3. They represent the cylinder pressures under a non-knocking (1st percentile1, Figure 2) and knocking (99th percentile, Figure 3) condition. For each figure, two low pass filters were created. The c
filt
T
cylinder pressure without filtering (red solid line) and the
(blue dashed dotted line). The middle plot shows a zoom in of the data just after TDC, where knock is expected.
The bottom plot in Figures 2 and 3 shows the cylinder pressure filtered with a 5 kHz to 27 kHz band-pass filter. With the application of the band-pass filter the low fre-quency component (mean pressure in the cylinder) of the pressure due to compression, heat release, and ex-pansion is removed. The remaining pressure signal
high rates of combustion (combustion rumble) and com-bustion knock.
1The percentile refers to the pressure intensity knock metric which is detailed below.
Figure 2: Cylinder pressure (C4) for a non-knocking combustion event in test 54.
Comparing the pressure signals in Figures 2 and 3 for the non-knocking and knocking cycles within this test, the evidence of knock is quite clearly defined by the large signal increase in the 5 to 27 kHz band. For these two cycles the peak in the band-pass pressure is 24 and 277 kPa for the non-knocking and knocking cycles re-spectively (rms = 3.5 & 21.1 kPa). In addition, in Figure 3 it is seen that the oscillating band-pass pressure signal attributed to knock begins shortly after 10q ATD reaching several peaks in the 12.5 to 25q range, after which the pressure oscillations decay. These are well known char-acteristics of knock and are the basis for most knock detection methods including those employed in this work. This process of band-pass filtering is used throughout this analysis to extract the frequency content of interest for both the pressure and accelerometer sig-nals.
Figure 3: Cylinder pressure (C4) for a knocking com-bustion event in test 54.
Frequency Analysis
In Figure 4, the frequency spectrum of the cylinder pres-sure for cylinder 4 in test 54 for the 99th percentile knock combustion event is shown. Similarly in Figure 5 the fre-quency spectrum for accelerometer 1 during the same time is shown. The time “zero” corresponds to the crank-angle of 12.5q in Figure 3, and for this test at an engine speed of 2500 rpm, 1ms equals 15q. In these figures an intensity scale is provided on the right hand side. As the amplitude of the signal in the frequency band designated on the y-axis increases, the color in the figure transitions from dark blue (lowest intensity) to dark red (highest in-tensity). In Figure 4 a high-pass filter of 2.5 kHz was used prior to determining this spectral image for this pressure signal. This was done to remove the large DC components due to the pumping and combustion proc-esses.
Time (ms)
-2-1
123
4
5
051015
202530Figure 4:Cylinder pressure frequency spectrum plot from a combustion event in test 54.
Figure 4 shows that the primary frequencies in the pres-sure spectrum (those with colors of red to dark red) are 6-8 kHz and 12-15 kHz with smaller contributions in the frequency ranges of 18-20 kHz and 22.5 - 25 kHz.These signals remain quite strong for 1ms.Following this the signal amplitudes decay and the frequencies begin to decrease. The decay results from the non-isentropic propagation and reflection of the waves in the combustion chamber, the non-ideal shape of the com-bustion chamber,and the expansion of the combustion chamber as the piston begins to move down. The reduc-tion in frequency is also a result of the expansion of the combustion chamber and the reduced gas temperature (the frequency is proportional to speed of sound of the gas which is proportional to the square root of tempera-ture, see Appendix B). Evidence of the oscillations con-tinue to 4ms (60q ) after the initial oscillations.
Figure 5 shows the corresponding frequency spectrum r accelerometer 1for the same combustion event. The of the cylinder,it is not surprising that this mode is not detected effectively.
fo primary frequency observed on the accelerometer is in the 10-12 kHz range with smaller contributions in the frequency ranges of 5-7.5 kHz, 14.5-16.5 kHz, and 21-23 kHz. The primary frequency in the accelerometer does not correlate with frequencies observed in the pressure sensor; however, this is not uncommon. Recall that the pressure sensor is located in a single position in the combustion chamber and thus does not detect all modes of oscillation with the same effectiveness [18].This is in contrast to the accelerometer, which detects the vibrations on the engine induced by the pressure waves impacting all parts of the combustion chamber. Examining the estimated frequency modes in Appendix B, the 10-12 kHz frequency correlates with the (2,0) circumferential mode at an estimated frequency of 10.9kHz. As the pressure transducer is located near the center Time (ms)
-2
-10
12345
05
10
15
20
25
30
Figure 5:Accelerometer frequency spectrum plot from a combustion event in test 54.
As shown in Figure 5, the accelerometer signal is de-layed by approximately 0.33 ms relative to the pressure signal. In addition the amplitude of the accelerometer signal is extended in time as compared to the pressure signal with relatively high amplitudes occurring out to 2ms (compared to 1ms for the pressure); in addition, significant signal amplitude is observed out to 5 ms.The frequency decay observed in the pressure signal is not apparent in the accelerometer signal. The signal ob-served in the 5–7 kHz range in the time interval from -2 to 0 ms is background vibration.
Figure 6shows the frequency content of the combustion events for the pressure signal (top plot) and accelerome-ter 1 (lower plot) over ranges of knock-intensity 2. These figures were created by windowing and computing the FFT for each combustion event. A cosine-tapered win-dow with a 0.5 ratio of taper to constant sections was utilized.The window started 0.75 ms prior to observing ignal to 1.5ms after the increase.6.5-7.5,19and 24kHz.These frequencies
the rapid increase in s This window has shown to capture the initial burst of energy well without including significant noise that can occur for longer window periods. In addition the taper at the start and end of the window ensures that end points have a zero amplitude and a zero derivative, minimizing leakage in the FFT. The amplitudes of the FFT were av-eraged over the combustion events in the range of knock-intensities listed (e.g, 0 – 5th percentile).
The upper plot of Figure 6 shows the pressure FFT am-plitudes for cylinder four for all the combustion events in test 54.We see that as the knock level increases from 0-5to 5-95and finally to the 95-100% percentiles, the am-plitudes of the FFT at the four frequency modes also increase. The frequency amplitude is highest at 13.7kHz.Other peaks in the frequency amplitude are seen at 2
Again this is in reference to the PI metric below.
were also observed in the individual combustion event shown in Figure 4. Comparing the observed frequencies in the pressure signal with the estimated mode frequen-cies in Appendix B, the measured frequency band6.5-7.5kHz correlates with the (1,0) circumferential mode at 6.6 kHz. The 13.7, 19, and 24 kHz measured frequen-cies are in good agreement with the estimated frequen-cies13.7,19.1, and 24 khz from the modes (0,1), (1,1) and (2,1). Again it is not surprising that the pressure sensor picks up these modes with the order 1 radial mode given its central location.
The lower plot of Figure 6 shows the average FFT ampli-tudes over the respective knock percentiles for acceler-ometer #2. Again as the knock percentile increases, the amplitudes increase. The modes observed include 7.5, 11.5, 13.7-16, 18.5-19, and 22.5 kHz and are consistent with the modes observed in the single combustion event shown in Figure5.Of these modes only the 11.5 kHz peak does not have a corresponding mode observed in the pressure.
From the observations and the analysis as shown in Fig-ures 2-6, the frequencies and crank-angle locations of the knock events are determined. Typically the identified frequencies do not change significantly with speed or load [20]. However at different operating conditions the relative amplitude of the modes may differ. The crank-angle locations and durations of the signals are however sensitive to engine speed and load and this variation must be accounted for when performing this analysis and using these metrics for knock indication.
There have been many metrics for determining knock intensity developed over the years; however, most rely on the two principles observed here, that the knock events result in signals
x with frequency modes that can be estimated as in Appendix B,
x that occur during specific crank-angle periods during combustion.
The metrics utilized in this analysis also make use of these characteristics.
Pressure and Accelerometer Intensity
Figures7and8 show the final processing in determining the knock intensities from the pressure and accelerome-ter signals.In Figure7 the pressure (top) signal from cylinder 4 and accelerometer (bottom) signal from A1 have been band-pass filtered with a 5-27 kHz filter and rectified. The signals for both a knocking (red trace with 0offset)and non-knocking(blue trace with offset at -100 for pressure and -50 for accelerometer plots)are shown. In Figure 8 the band-pass filtered and rectified pressure (top) and accelerometer (bottom) signals are windowed in crank-angle as shown by the green dashed line and integrated(summed)during
the window period.Figure 6:Frequency components for cylinder 4 pressure and accelerometer 1 signals during the expected knock period for test 54.
The integrated signals are shown for the knocking event by the red line and the non-knocking event by the blue line. The integrated values at the end of the integration are the pressure intensity (PI) and accelerometer inten-sity(AI)that are used throughout the remaining analysis. These knock intensities are computed by the following equations:
¦
³
¦
³
n
i
f
t
t
t
t
f
n
i
f
t
t
t
t
f
i
a
n
dt
t
a
t
t
N
AI
i
p
n
dt
t
p
t
t
N
PI
1
1
2
1
1
2
)(
1
)(
1
)
(
)(
1
)(
1
)
(
2
2
1
1
2
2
1
1
T
T
T
T
, [1]
where p f and a f are the filtered pressure and acceler-ometer signals respectively, and PI and AI are computed for every combustion cycle N.These knock intensity metrics are the average rectified amplitude of the filtered signals during the crank period T1to T2.
For these two combustion events shown in Figure8,the pressure intensities are 4.6 and 46.4 for the non-knocking and knocking events. The accelerometer in-tensities for the same two combustion events are4.2 and 27.6. These two knock metrics are then considered a data pair (PI, AI) which for these two combustion events are (4.6, 4.2) for the non-knocking condition and (46.4, 27.6) for the knocking condition. For test 54there are 210 data pairs (one pair per combustion cycle)and this set of data pairs are then fit with a linear least。