Accelerometer signal to diagnose combustion

Accelerometer signal for combustion diagnosis in Diesel engines A.P. Carlucci, F.F. Chiara, D. Laforgia Department of Engineering for Innovation University of Salento (Italy)

Abstract The combustion development in a Diesel engine is strictly dependent on the injection parameters like the number of injections, the fuel quantity, the injection timing and pressure. The combustion ignition and development, on the other hand, affects the engine block vibration response, in a peculiar way, discernible by the vibration due to the rotation of the engine. The idea of the present work is to show the monitoring capability of an accelerometer signal installed on the engine block. A series of tests, varying the injected fuel quantity, the injection pressure and the number of injections have been performed measuring the in-cylinder pressure and the vibration signal. Classical Fourier transform and time frequency methods have been applied to the engine block vibration and to the pressure signals, in order to show their correlation. 1. Introduction Monitoring and diagnostic techniques based on non intrusive methods are an effective way to optimize the operation of a machine, acting on the maintenance when it is needed. These techniques are largely applied to rotating machines like turbo-pumps, compressors, turbines but they do not have direct applications on reciprocating internal combustion engines. This is due to the more complex dynamics characterizing these machines, as well as to more complex phenomena to model and consider like the combustion development. On the other hand, the quality of the combustion phenomenon strictly affects the emission levels, which, on the basis of the current and future emission legislations, are going to be very stringent and demanding in terms of on board monitoring. The proper operation of the injection system is crucial to meet the target imposed by the need of high efficiency and low emissions. Therefore, a non intrusive way to diagnose its correct operation, while the engine is running, is a viable solution to approach and solve the on board diagnosis need. Different diagnostic techniques have been suggested in the past to monitor combustion misfiring. In [1] Azzoni et al. use an indicator based on the crankshaft speed fluctuations. In particular, the engine speed signal, acquired on angular basis, have been analyzed using the Fourier transform to calculate the amplitudes of the first three orders of components of the signal. The combustion indicator, defined on the basis of the main frequency components of the angular speed signal, is then used to diagnose the occurrence of misfiring events. Ball et al. [2]-[3] use the environmental noise to calculate an indicator based on wavelet transforms. Different malfunctions of the engine have been simulated and, some of them, using the aforementioned technique, like compression ratio or injection pressure variations, have been recognized on the signal. Several authors have exploited the use of non intrusive diagnostic techniques applied to internal combustion engines. Currently, one of the most effective way to diagnose the occurrence of the knock phenomenon in spark ignition engines (SI) is based on the frequency analysis of the block vibration signal. Different techniques have been suggested in the past, but two main approaches can

be recognized: the first one is based on crankshaft speed fluctuations [4]-[6] , while the second, as already observed, is based on the frequency analysis of the block vibration signal [7]-[15]. Knock phenomenon is characterized by a localized auto-ignition in a portion of the combustion chamber in which the flame front is not arrived yet. This portion of the mixture is called end-gas. When the temperature and pressure in the end-gas reach and overcome a threshold, the mixture auto-ignites and another flame front develops from it causing very high pressure levels. The energy developed by the knock ignition can cause too high stresses to the combustion chamber elements causing them to break. The knocking ignition causes the development of a steep fronted wave, travelling within the combustion chamber, from the ignition point to the chamber walls, causing the resonation of the chamber and the vibration of the walls. In order to recognize the occurrence of this undesired phenomenon, the block vibration signal is acquired and filtered in a band around the first nominal chamber resonance frequency. The energy content of this frequency band is then compared to a threshold. Exceeding this threshold means a knocking combustion [7]. The indicator is used as feedback signal to act on the spark timing to immediately stop the phenomenon. Other methodologies, not yet applied for on-board applications, use more sophisticated indicators based on time-frequency distributions. In [8] a comparison between diagnosis techniques based on classical Fourier transforms and time-frequency distributions is presented, showing that this last one is more effective in different operating conditions. Differently from gasoline engines, in which the knocking combustion is an anomaly to be avoided, in Diesel engines, knock is the way flames propagate from the ignition nuclei to the rest of the combustion chamber. In particular, the ignition of the premixed portion of the charge, formed in the time interval between the injection and the autoignition (ignition delay), results in a highly non uniform energy release which acts on the combustion chamber dynamics causing its resonance. Large pressure oscillations around the mean pressure value, at defined frequencies, are therefore strictly linked to the combustion ignition and their amplitude can be related to the subsequent combustion development. A deeper understanding of the combustion chamber resonance phenomenon can give several information on the influence of different injection patterns, with the related combustion development, as Blunsdon and Dent show in [16]. In this work KIVA code has been used to simulate the influence of different injection strategies on in-cylinder pressure oscillations. In Diesel engines the combustion development is particularly dependent on the injection parameters, like the number of injection performed per cycle, the fuel quantity and the injection pressure. Each of these parameters acts in a peculiar way on the combustion development, modifying the in-cylinder pressure shape. The direct consequence of the variation of the injection parameters is the modification of the energy distributions of the surface vibration, as a response to the pressure shape change. The aim of the present work is, therefore, to use the vibration signal frequency content to extract useful understanding about the combustion development in order to relate it to the variation of the injection parameters. The classical Fourier approach, in the present work, has been integrated with a more sophisticated technique based on the use of time frequency distributions to calculate the instantaneous frequency contents of the pressure and accelerometer signals. 2. The experimental setup and the tested conditions The tests were carried out on a four cylinder DI FIAT 1929 cm3 TDID 154 D 1.000 Diesel engine, turbocharged with a GARRETT TD 2502 and equipped with a common rail (Bosch 1350 bar) injection system (Table 1). The acceleration of the cylinder block was measured by means of a

piezoelectric accelerometer, KISTLER K SHEAR 8704B100, powered by a single channel coupler KISTLER type 5118B2. The accelerometer (a1 in Figure 1) was firmly fixed by means of a threaded pin, on the side wall of the block close to cylinder No. 4 whose pressure was acquired. Table 1. Engine and fuel injection system specifications.

ENGINE SPECIFICATIONS Total number of cylinders Total displacement Maximum power Maximum torque Minimum speed Maximum speed Bore Stroke Compression ratio Inlet valve opening Inlet valve closure Exhaust valve opening Exhaust valve closure

4 1930 69 200 950 4950 82.6 90 19.8 ± 0.8 0° 32° 32° 0°

cm3 kW Nm rpm rpm mm mm BTDC ABDC BBDC ATDC

FUEL INJECTION EQUIPEMENT SPECIFICATIONS Injector type Injection pressure Nozzle type Number of nozzle holes Hole diameter Hole L/D ratio Included spray angle

Electro-hydraulically controlled Variable up to 135 MPa VCO 5 0.18 mm 6 145°

Figure 1. Experimental setup scheme.

The in-cylinder pressure was acquired using a piezoelectric transducer, KISTLER type 6053, connected by a charge amplifier to the acquisition board. The voltage signals of the accelerometer and the in-cylinder pressure signal were sampled with an acquisition board NI type 4472, with a peak acquisition frequency of 100 kHz per channel and an anti-aliasing filter. The sampling frequency used for all the aforementioned signals was 50 kHz. The accelerometer and pressure transducer main features are summarized in Table 2. Table 2. Accelerometer and pressure transducers specifications

ACCELEROMETER SPECIFICATIONS Sensor type Measuring range Sensitivity Natural frequency Transverse sensitivity

[g] [mV/g] [kHz] [%]

piezoelectric  100 50 50 1.5

PRESSURE TRANSDUCER SPECIFICATIONS Sensor type Measuring range Sensitivity Natural frequency Linearity

[bar] [pC/bar] [kHz] [%FSO]

piezoelectric 0 - 250 -19 130 <  0.5

The tested operating conditions are summarized in Table 3. All the tests refers to the operating condition defined by an engine speed of 1400 rev/min and an output torque of 60 Nm. Table 3. Experimental plan

Test 1 2 3 4 5 6

AiP [DCABTDC] 19 19 19 19

EtP [s] 150 150 200 200

Ai [DCABTDC] 0 0 0 0 0 0

pinj [MPa] 60 100 60 100 60 100

Two different injection strategies have been analyzed. In particular, tests 1 and 2 in Table 3 refer to a single injection performed at 0 degree of crank angle before top dead center (DCABTDC), while tests from 3 to 6 refer to a double injection scheme in which the first injection, namely pilot injection, is performed at 19 DCABTDC and the main injection at 0 DCABTDC. In Table 3 AiP is the pilot injection timing, ETP is the duration of the injector’s electrovalve opening related to the pilot injection, Ai is the main injection timing and finally pinj is the injection pressure. 3. The main features of the pressure signal as stimulus for the engine block vibration response Diesel engine combustion can be schematically divided in three stages [17]: the first one, namely ignition delay, in which the fuel injected in the combustion chamber accumulates and spreads without autoigniting; the second one, in which a sudden development of the combustion of the

premixed charge takes place, namely premixed combustion; and the third one, in which the combustion of the remaining portion of the fuel injected develops in heterogeneous phase, namely diffusive combustion. The harshness, which is the distinctive aspect of diesel combustion, is caused by the sudden development of the combustion of the premixed charge formed during the ignition delay; increasing the ignition delay, the portion of the fuel burnt in the premixed phase increases as well. In the past, great attention was addressed to study a method to characterize the time diagram of the in-cylinder pressure to predict its effects on block vibration and engine noise. Some of these methods consider the peak value or alternatively the first or second derivative of the in-cylinder pressure. As a matter of fact, the in-cylinder pressure peak value clearly affects the low-frequency components, while the highest frequency harmonics are directly related to the shape of the pressure history. In particular, the sudden detachment from the motored (without combustion) pressure curve, due to the early stage of the combustion process, produces a steep fronted wave that, spreading inside the combustion chamber, causes charge and pressure oscillations, source of highfrequency peaks on the in-cylinder pressure spectrum. In Figure 2 the Power Spectral Density (PSD) function of the pressure signal is showed. The PSD of the pressure signal p(t) was calculated according to equation (1): 

PSD p ( f ) 

 Ep(t )  p(t   ) e

 j 2f

d

(1)



The PSD can be read as a measure of the rate of variation of the root mean square (RMS) of the data varying the frequency. Therefore, it gives useful informations on the distribution of the power of the signal in the Fourier space.

Figure 2. Power Spectral Density function of the in-cylinder pressure signal for test 6 in Table 3.

Analyzing Figure 2, it can be observed that, beyond the initial part of the diagram, showing a nearly straight trend, due to the high density of the components related to the firing frequency, the highfrequency peaks, due to the resonant oscillations of the charge, are clearly detectable. In particular three evident peaks can be detected in the range between 5 and 15 kHz. These peaks are directly

related to the propagation of the pressure wave caused by the combustion ignition and they are the main source for the engine block vibration. 4. Preliminary treatment of the signals Both the accelerometer and the in-cylinder pressure signals were processed, in order to exclusively isolate all the interval of 180 DCA around the top dead center (TDC) corresponding to the end of the compression stroke of the cylinder No. 4. This angular windowing is needed to isolate the vibration signal components only, due to the combustion on the cylinder in which the in-cylinder pressure was detected, leaving out the contribution of the combustion in the adjacent cylinders. In particular, Figure 3 presents, in arbitrary units (A.U.), all sampled signals after the time windowing operation, while Figure 4 underlines that in the aforementioned angle interval, the combustion inside the cylinder No. 4 is the only one that takes place. Amplitude [A.U.]

time [s]

Figure 3. Accelerometer and in-cylinder pressure signal after the windowing operation

OVO(3) p_inj(4)

IVO(1) OVC(1)

IVC(2)

m_inj(4)

crank angle [deg] 90

32 23/19

0 3/0

-32

-90

TDC(4) Figure 4. Theoretical occurrences of inlet valve opening (IVO) and closure (IVC), output valve opening (OVO) and closure (OVC), pilot injection (p_inj) and main injection (m_inj) start for the cylinder in round brackets.

Therefore, it is possible to recognize the purely mechanical events distinguishing them from those caused by the combustion process. Although the signals were acquired with a temporal basis, the good operating stability of the engine made it possible to convert them to the angular domain. In this way, it was possible to move all the analysis from time to angle domain. 5. Time frequency characterization of pressure and accelerometer signal

The information on the in-cylinder pressure signal and therefore on the block vibration signal, obtained through the classical Fourier transform, can be integrated using a more appropriate timefrequency approach. In fact the combustion phenomenon is highly non stationary due to the variation of the resonant frequencies of the combustion chamber, because of the variation of both the chamber volume and temperature, during the expansion stroke. Therefore the time-frequency approach allows the description of the frequency content variation of the signals with the time. If a simplified cylindrical chamber with acoustically hard walls, filled with a homogeneous gas is considered and a simple model of excitation is simulated, the related natural frequencies can be calculated. In the most general case, the real vibrations of the cylindrical chamber results from the superposition of infinite characteristic mode, transversal or longitudinal, depending on the direction of propagation of the pressure wave. In the real case, however, the theoretical approach can be applied only approximately, because of the geometric complexity of the combustion chamber and the unhomogeneity of the charge distribution. In general, a pressure signal can be described by a multi-component signal modulated in both frequency and amplitude [12]: P

y (t )   Ap e

t

cos(  p ( )d   p )

d pt

p 1

(2)

t0

where t0 is the reference instant of time , Ap and p are the initial amplitude and phase of the p-th resonance, dp is the damping constant and p is the instantaneous angular frequency. Cohen in [18] showed that, for a mono-component signal whose instantaneous frequency varies linearly with time, only the Wigner distribution can properly reproduce the expected shape of the instantaneous power concentration and the related instantaneous frequency, once verified the condition that the amplitude variation inside the windowing function is much slower than the phase variation. In this case, the instantaneous frequency is given by:

 (t )  arg max WD yy (t , ) 

(3)





where WDyy(t,) is the Wigner or the pseudo-Wigner distribution of the analytic signal y(t), the last being defined as: 







WD yy (t , )   w( ) w( ) y(t  ) y * (t  )e  j  d 2 2 2 2

(4)

with w(t) the windowing function. For a multi-component signal, as an engine signal described as shown in (2), a distribution D(t,) has to be defined, having the property of giving the instantaneous power of each of the frequency components according to the following relation:

D(t ,  ) 

P

WD y y p 1

p

p

(t ,  )

(4)

where WDy p y p (t ,  ) is the pseudo-Wigner distribution of the component yp(t). The analysis of a multi-component signal should be able to determine the number of the components carried out by the signal, to discriminate between the components and the cross-terms, to resolve components that are very close in the time-frequency plane and finally it should give an estimate of the instantaneous frequency of each components. All these points can be satisfied only

in the case of a highly concentrated time-frequency distribution free or almost free from cross terms. Unfortunately these two conditions are difficult to be reached simultaneously, even if the research efforts in this direction are very intensive. One of the classes of time-frequency distribution that has been proposed to analyze non-stationary signals is the quadratic (or Cohen’s) one. Different distributions of this class has been used to analyze various real life problems (radar, sonar, automotive, etc...) but none of them can reach all the previous objectives simultaneously. An efficient method leading to a distribution satisfying the important property defined in (4), is based on the S-distribution, used in the present work to apply the S-method [19]. The S-distribution, is based on the scaled variant of the L-Wigner Distribution defined, in its pseudo-form, by: SDL (t ,  )   wL ( ) y [ L ] (t  

 2L

) y [ L ]* (t 

 2L

)e  j  d

(5)

in which y[L](t) is obtained by y(t) multiplying its phase by L and keeping the amplitude unchanged, while wL() is the windowing function. This kind of distribution is able to achieve high concentration around the instantaneous frequency, satisfying, at the same time, the energy condition, the time marginal and, for asymptotic signal, the frequency marginal property. The direct realization of the S-method is based on the MPWD(t,) (Modified Pseudo-Wigner Distribution): MPWD(t ,  ) 

1

P( )  STFT (t ,    )  STFT 

*

(t ,    )d

(6)

where P() is a frequency windowing function aimed at decreasing the effect of the cross terms and STFT is the Short Time Fourier Transform. Stanković and Böhme in [20] applied the S-method for knock detection in a spark ignition engine, analysing simulated and experimental engine signals. 6.1 Effect of the tested parameters The in-cylinder pressure history clearly reveal the start of the combustion process, showing also amplitudes varying according to the peak of the first derivative of the pressure. On the other hand the variation of the injection parameters, such as the number of injections, the fuel quantity injected during each injection, and the injection pressure, strictly affects the combustion evolution. Using a pressure sensor for condition monitoring would be very interesting, but its use in real applications is limited by the fact that it is very intrusive, and its cost is still too high. The block vibration signal, detectable by common and cheaper accelerometers, can be seen as a time-varying filtered version of the pressure signal, therefore the real challenge is to recognize, on those signals, the frequency components directly related to in-cylinder pressure in order to operate a continuous engine operation monitoring. The effect of each of the parameters varying in Table 3 has been analysed through the aforementioned S-method, in order to show the relation between the in-cylinder pressure fluctuations and the block vibration signal. Since the frequency components of interests for both the accelerometer and pressure signal are located in a frequency band above 3 kHz, the analysis has been performed on the high pass filtered portion of the signals. In order to relate the quantities derived form the analysis of the pressure and accelerometer signal, the Heat Release Rate (HRR), calculated from the low pass filtered portion of the pressure signal

was computed through classic thermodynamic single-zone model. The model (7) is derived by the application of the first law of thermodynamics [17].

ROHR( ) 

dV   1 dp   V  1 d k 1 d



p

(7)

In Eq. (7) V() is the chamber volume at the crank angle , p() is the in-cylinder pressure value measured at the same crank angle and  is the specific heat ratio. 6.1.1 Effect of different injection schemes The effect of two different injection strategies on both the in-cylinder pressure fluctuations and the block vibration signal has been tested. In particular a single injection scheme was compared with a double injection scheme in which a pilot injection is performed before the main injection. In Figure 5 the time frequency analysis of the in-cylinder pressure signal (a), and of the vibration signal (b) for test 2 in Table 3. Experimental plan is presented. The structure of Figure 5 is described as follow: -1- and -4- are the Discrete Modified Pseudo-Wigner Distribution (DMPWD) of the signals, -2- and -5- the instantaneous power of the frequency components detected by the Smethod and showed in -1- and -4- by the dotted black line, -3- and -6- respectively the HRR and the high pass filtered portion of the block vibration signal. For linear frequency modulated signal components, like the ones we are analysing, the instantaneous frequency can be estimated by the local maxima of the DMPWD in which all the values under a significant threshold have been eliminated. The structure of the figures from 6 to 10 is the same as the one already described. The peak of the HRR (shown in Figure 5(a)-3-) is located at around 12 DCAATDC while the peak power of the main frequency component is located at around 18 DCAATDC. The block vibration signal shows large oscillations starting at around 12 DCAATDC as shown in Figure 5(b). The peak power of the main frequency components of the vibration signal is located at around 14 DCAATDC. Therefore the power of the accelerometer signal is concentrated exactly in the crank angle interval in which the combustion energy is released.

-1-

-4-

-2-

-5-

-3-

-6-

(a)

(b)

Figure 5. Time frequency analysis of the in-cylinder pressure fluctuations (a) and accelerometer signal (b) for test 2 in Table 3. (1) and (4) are the DMPWD superimposed with the instantaneous frequency of the signals, (2) and (5) are the instantaneous power of the frequency components, (3) is the HRR, (6) is the high pass filtered portion of the accelerometer signal.

Comparing the single injection (Figure 5) with the double injection scheme (Figure 6), it can be observed that the correlation between the start of the pressure fluctuation and the increase of the vibration signal power in the last case still exists. In fact, the start of the heat released during the pilot , located at around 8 DCABTDC, and therefore the start of the pressure oscillation, are directly related to the power increase of the vibration signal becoming evident at nearly 10 DCABTDC. Nevertheless, the more complex heat release due to the double injection scheme, causes the development of a larger number of frequency components on the vibration signal, making the detection of the correlation between the signals more difficult.

-1-

-4-

-2-

-5-

-3-

-6-

(a)

(b)


6.1.2 Effect of the injected fuel quantity In general, the fuel injected quantity depends on the combination of injection pressure and the time interval of the needle opening, which is related to the energizing time of the solenoid of the electroinjector. The fuel quantities, as a function of ETP and pinj, injected during the pilot injection are reported in Table 4. Pilot injection quantities. Table 4. Pilot injection quantities

Test

pinj [MPa]

ETP [s]

Pilot quantity [mm3/cycle]

3 4 5 6

60 100 60 100

150 150 200 200

1.3 2.8 2.0 4.0

Percentage of pilot fuel mass to total [%] 7.1 15.4 11.0 22.0

Figure 7, showed in the following, refers to test 3 in table 3. It can be observed that in this case, despite the double injection scheme, the heat release shows only one peak at around 14 DCAATDC, as can be noticed from Figure 7(a)-3-. The pilot injection quantity in this case is too low to ignite, therefore it accumulates in the combustion chamber and burns with the main injection. The strict correlation of the in-cylinder pressure fluctuations and the vibration signal

power increase is particularly evident in this case. In fact the peak power of the main frequency components of both the pressure and the vibration signals is respectively located at 17 and 18 DCAATDC, both after the peak of the HRR. -1-

-4-

-2-

-5-

-3-

-6-

(a)

(b)


Increasing the fuel quantity in the pilot injection, as showed in Figure 8 referring to test 5 in Table 3. Experimental plan the pilot injection ignites before the main injection. This phenomenon has a direct consequence on the power distribution of the in-cylinder pressure and on the vibration signal. In fact the peak power of the components of both the signals is evidently advanced compared to the one shown in Figure 7.

-1-

-4-

-2-

-5-

-3-

-6-

(a)

(b)


6.1.3 Effect of the injection pressure

The effect of increasing the injection pressure is twofold. In fact, from one hand there is an effect on the injected fuel quantity and therefore on the potential energy that can be released during the combustion. The fuel mass flow rate m injected varies approximately according to Eq. (8):

  Ce  p   c m

(8)

where p is the difference between fuel pressure at injector nozzle and in-cylinder pressure, Ce is efflux coefficient, is the surface area of the injector holes and c is the volumetric mass of the fuel. Increasing p, the fuel mass flow rate increases as well. On the other hand, the variation of the injection pressure affects the dimension of the droplets of fuel and their total penetration inside the combustion chamber. This can be observed in Eq. (9) which describes the dependency of the droplet diameter d g from several parameters:  2 l d g  Cd   u 2  g rl

 *   

(9)

where Cd is a nearly unit constant, g is the gas volumic mass,  is the dimensionless wavelength of the spray surface fluctuations, url the relative motion between the gas and the liquid and l the surface tension of the liquid. The increase of url due to the increase of the injection pressure, reduces the mean dimension of the droplets creating better mixing conditions and faster ignition of the combustion process. Comparing Figure 9 to Figure 5, referring respectively to test 1 and 2 in Table 3. Experimental plan , the effect of the injection pressure increase can be observed in the single injection scheme. In particular, the increase on the HRR peak from 70 to 105 J/deg causes a related increase of the peak power of the main frequency components of the vibration signal, which is three times the previous value. Since only one injection is performed there is only one main heat release.This is directly related to the shape of the instantaneous power of the main frequency components showing only one peak related to the location of the HRR peak.

-1-

-4-

-2-

-5-

-3-

-6-

(a)

(b)


Using a double injection scheme, the accelerometer signal still follows the HRR rate shape and trend. In fact comparing Figure 8 and Figure 10, referring respectively to test 5 and 6 in Table 3, the HRR of both the pilot and the main injection can be recognized on the power distribution of the frequency components of the vibration signal. Both the peak power trend and the shape of the power distribution of the block vibration signal are well correlated with the measured HRR. Therefore the effect of this parameter can be clearly recognized on the vibration signal.

-1-

-4-

-2-

-5-

-3-

-6-

(a)

(b)


7. Conclusions The effect of injected fuel quantity and injection pressure, as well as the effect of using two different injection schemes has been inquired on the in-cylinder pressure fluctuations and on the block vibration signal of a common rail Diesel engine. The analysis was performed using algorithms based on a time frequency distribution, in order to detect and insulate the instantaneous frequency and power distribution of the aforementioned signals. The variation of the injection strategy using two different injection schemes, the first with a single and the second with a double injection, shows a direct influence on the vibration signal. In particular the distribution of the power of the instantaneous frequency components of the vibration signal follows the HRR, in terms of shape and trend. The vibration signal is obviously delayed compared to the HRR. In fact, a certain amount of time is needed for the propagation of the pressure wave within the combustion chamber to the wall, where the accelerometer is installed. The effects of the injected fuel quantity and the injection pressure increase were recognized as well on the vibration signal, both leading to an increase of the instantaneous power carried by the frequency components of both the pressure and vibration signal. The angular location of the peaks of the instantaneous power of the vibration signal were also in accordance with the related peaks of the HRR. References

[1] P. Azzoni, G. Cantoni, G. Minelli, D. Moro, G. Rizzoni, M. Ceccarani, S. Mazzetti, Measurement of engine misfire in the Lamborghini 533 V-12 engine using crankshaft speed fluctuations. Society of Automotive Engineers Paper No. 950837, 1995. [2] F. Gu, W. Li, A. D. Ball, A. Y. T. Leung, The condition monitoring of diesel engines using acoustic measurements. Part 1: Acoustic characteristics of the engine and representation of the acoustic signals. Society of Automotive Engineers Paper No. 2000-01-0730, 2000. [3] D. Ball, F. Gu, W. Li, The condition monitoring of diesel engines using acoustic measurements. Part 2: Fault detection and diagnosis. Society of Automotive Engineers Paper No. 2000-01-0368, 2000. [4] G. Mauer, On-line Performance Diagnostics For Internal Combustion Engines. Proceedings of 1992 International Conference Industrial Electronics, Control, Instrumentation and Automation, vol. 3, pp. 1460-1465, 1992. [5] A.M. Alkhateeb and M. Das, A Robust Algorithm for Identifying Different Types of Engine Misfires. Proceedings of the 45th Midwest Symposium on Circuits and Systems, vol. 3, pp. 465-468, 2002. [6] D. Taraza, N.A. Henein and W. Bryzic, The Frequency Analysis of the Crankshaft’s Speed Variation: a Reliable Tool for Diesel Engine Diagnosis. Transactions of the ASME, vol. 123, pp. 428-432, 2001. [7] G. deBottom, J. Ben-Ari and E. Sher, Vibration Monitoring as a Predictive Maintenance Tool for Reciprocating Engines. Proc. Instn Mech. Engrs, vol. 214 Part D, pp. 895-903, 2000. [8] B. Samimy and G. Rizzoni, Engine Knock Analysis and Detection Using Time-Frequency Analysis. SAE paper 960618, 1996. [9] G. Rizzoni and X.C. Chen, Detection of Internal Combustion Engine Knock Using Time-Frequency Distributions. Proceedings of the 36th Midwest Symposium on Circuits and Systems, vol. 1, pp. 360-363, 1993. [10] B. Samimy and G. Rizzoni, Time-Frequency Analysis for Improved Detection of Internal Combustion Engine Knock. Proceedings of the IEEE-SP International Symposium on Time-Frequency and Time-Scale Analysis, 1994. [11] B. Samimy and G. Rizzoni, Mechanical Signature Analysis Using Time-Frequency Signal Processing: Application to Internal Combustion Engine Knock Detection. Proceedings of the IEEE, vol. 84, pp. 1330-1343, 1996. [12] F. Böhme and S. Carsten-Behrens, Applying Time-Frequency Methods to Pressure and Structure-Borne Sound for Combustion Diagnosis. Sixth International Symposium on Signal Processing and its Applications”, vol. 1, pp. 256259, 2001. [13] F. Böhme and J. F. König, Application of Cyclostationary and Time Frequency Signal Analysis to Car Engine Diagnosis. Proceedings of the IEEE-ICASSP, vol. 4, pp. 149-152, October 1994. [14] F. Böhme and J. F. König, Statistical Processing of Car Engine Signals for Combustion Diagnosis. Proceedings of seventh SP Workshop on Statistical Signal and Array Processing, pp. 369-374, Quebec 1994. [15] D. Konig, Application of Time-Frequency Analysis for Optimum non Equidistant Sampling of Automotive Signals Captured at Knock. Proceedings of IEEE International Conference on Acoustics, Speech, and Signal Processing, vol. 5, pp. 2746-2749, 1996. [16] C. A Blunsdon and J.C. Dent. “Modelling the Source of Combustion Noise in a Direct-Injection Diesel Engine using CFD”, SAE paper 941898, 1994. [17]

J. B. Heywood Internal combustion engine fundamentals McGraw-Hill Publishing Company, New York, 1988.

[18]

L. Cohen. “ Time Frequency Analysis”. Prentice Hall PTR, New Jersey, US.

[19] L. Stanković. ”S-class of Time-Frequency Distributions”, IEEE Proceedings on Visual Image and Signal Processing, vol. 144, No.2 pp. 57-64,1997. [20] L. Stanković, F. Böhme. “Time-frequency Analysis of Multiple Resonances in Combustion Engine Signals”, Signal Processing, vol. 79, pp. 15-28, 1999.