Injection diagnosis through common-rail pressure measurement

http://pid.sagepub.com/Engineering

Engineers, Part D: Journal of Automobile Proceedings of the Institution of Mechanical

http://pid.sagepub.com/content/220/3/347The online version of this article can be found at:

 DOI: 10.1243/09544070JAUTO34

2006 220: 347Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile EngineeringF Payri, J M Luján, C Guardiola and G Rizzoni

Injection diagnosis through common-rail pressure measurement  

Published by:

http://www.sagepublications.com

On behalf of: 

  Institution of Mechanical Engineers

can be found at:Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile EngineeringAdditional services and information for

  http://pid.sagepub.com/cgi/alertsEmail Alerts:

http://pid.sagepub.com/subscriptionsSubscriptions:  

http://www.sagepub.com/journalsReprints.navReprints:  

http://www.sagepub.com/journalsPermissions.navPermissions:  

http://pid.sagepub.com/content/220/3/347.refs.htmlCitations:  

What is This? 

- Mar 1, 2006Version of Record >>

by guest on September 6, 2012pid.sagepub.comDownloaded from

347

Injection diagnosis through common-rail pressuremeasurementF Payri1, J M Lujan1, C Guardiola*1, and G Rizzoni2

1CMT – Motores Termicos, Universidad Politecnica de Valencia, Valencia, Spain2Ohio State University, Ohio, USA

The manuscript was received on 31 January 2005 and was accepted after revision for publication on 28 October 2005.

DOI: 10.1243/09544070JAUTO34

Abstract: Modern diesel common-rail injection systems supply fuel from a high-pressurevessel. The injection event causes an instantaneous drop in the rail pressure, as the storedmass is diminished. Pressure variations are also affected by the dynamics of the high-pressurepump that supplies fuel to the rail to compensate for the emptying process due to the injection.This paper proposes the possibility of diagnosing the injection process from measurement ofthe rail pressure. Different data treatment techniques are explored and evaluated in this paperto propose an effective method for the diagnosis of common-rail injection systems.

Keywords: diesel engines, common rail, failure detection, injection failure, misfire

1 INTRODUCTION ignition (HCCI) engines need a perfectly controlledmixture formation during the compression stroke.With a common-rail system, this goal may beModern diesel engines have significantly improvedachieved by performing several injections of smalltheir performance by means of electronic injection.quantities of fuel.Today, common-rail systems are well positioned

With such stringent requirements, minor errors inboth in the light- and heavy-duty vehicle markets.hole diameter, unavoidable owing to manufacturingCommercial common-rail engines have injectorsvariation and to the accumulation of deposits [9],with hole diameters varying from 140 to 200 mm andresult in decreased engine performance and increasedinjection pressures up to 140 MPa [1, 2]. Commonemissions. Another important factor is the dynamicrail systems have shown a clear tendency towardsresponse of the needle lift mechanism [10, 11]. Inincreasing injection pressure and decreasing holeextreme cases, an excessive delay in the opening ofdiameters (which for investigation purposes can bethe injector can cause the injection not to take place.as small as 40 mm [3]).This problem is especially important for injectionsCommon-rail systems allow the usage of veryof short duration, such as the post-injection, theflexible injection laws. To reduce combustion noise,pilot-injection and small injections used to enhanceit is normal in automotive engines to use a pilotmixture formation and promote quasi-homogeneousinjection, which consists of injecting a very small fuelcharge conditions.quantity 10–30° before the main injection [4, 5].

Two conclusions can be drawn from theseAlso, post-injections may be used to reduce smokeintroductory remarks.emissions [6]. Further, it is also possible to split the

injection into several smaller ones, resulting in 1. The number of injections in the new generation ofimproved efficiency and reduced pollutant formation diesel engines is quickly increasing. Moreover, as[7, 8]. Finally, new homogeneous charge compression the number of injections increases, the injections

become smaller, since the total mass of injected* Corresponding author: CMT – Motores Termicos, Universidad fuel must be maintained.Politecnica de Valencia, PO Box 22012 E-46071, Valencia, Spain. 2. Both performance and emissions can be seriously

affected by small faults in the injection system.email: [email protected]

JAUTO34 © IMechE 2006 Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from

348 F Payri, J M Lujan, C Guardiola, and G Rizzoni

Since the injection system is critical for emission are controlled by a PC connected to the ECU via asuitable communication link (ETK port). The railcontrol, it is necessary to develop algorithms that

check, detect, and correct, if possible, the effects of pressure could be independently controlled. Inaddition to the production pressure sensor used byits malfunction [12, 13]. The objective of the present

work is to develop a method for injection diagnosis the ECU for control purposes, the common rail wasalso instrumented with a research pressure sensorin common-rail diesel engines.

During the last two decades, several authors have (Kistler 4067A2000 sensor with a 4618A amplifier).Injection current and fuel mass flow were alsodeveloped algorithms for misfire detection, usually by

means of instantaneous crankshaft rotational speed measured for the purpose of locating the injections.Data were acquired by means of a Yokogawameasurement [14–19]; the present work is focused

on detecting the fault (‘a misfire is happening’) and acquisition system sampling at 100 kHz. A diagramof the installation is shown in Fig. 1.its possible cause (‘is the injection process correct?’).

For that, the instantaneous measurement of the railpressure will be used to reveal whether the injection 2.2 Engine test benchhas taken place.

To emulate real operating conditions, a four-cylinderAs the rail is a high-pressure vessel that supplies

common-rail passenger car engine was used. Thefuel to the injectors, each time an injector opens, the

engine was installed in a dynamic test cell withrail pressure experiences an instantaneous drop.

an electric dynamometer, as shown in Fig. 2. TheAlso, the dynamics of the high-pressure pump that

ECU controlled the injection pressure and the startsupplies fuel to the rail to compensate for the

of injection. Also, the existence and characteristicsemptying process due to the injection will affect

of the pilot injection were determined by the ECU.the instantaneous rail pressure [20]. A method for

Changes in the engine operating point, definedisolating the variations due to the injections from

by the engine speed and the accelerator pedalthose due to the fuel supply from the pump is pre-

position, result in the ECU automatically calculatingsented in this paper. The ability to discriminate

and applying new injection settings. To simulatebetween different injections in the same cylinder is

injection failures, one of the injectors was completelyalso studied. Further, it is shown that it is possible

disconnected in one of the tests.to diagnose the injection quality on the basis ofmeasurement of the common-rail instantaneouspressure obtained with a production pressure sensor.Finally, the method is shown to be useful even withsmall injections (such as pilot injections).

2 EXPERIMENTAL SET-UP

Two different installations were used. The first is acommon-rail test bench where the injection systemcan be tested separately from the engine. After thisinitial characterization, a direct injection dieselpassenger car engine with a commercial common-rail system is used. In this section, both installationsare described.

2.1 Common-rail test bench

A Bosch CP1 common rail is used for the first experi-ment. The high-pressure pump is driven by a speed-controlled electric motor. The motor shaft powersa replica of the engine flywheel which is used toprovide the crankshaft position signal to the fuelinjection system electronic control unit (ECU). Asingle injector is driven by the ECU. The number

Fig. 1 Common-rail test benchof injections and their duration and starting point

JAUTO34 © IMechE 2006Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from

349Injection diagnosis through common-rail pressure measurement

and computation are well understood [21]. However,joint time–frequency analysis techniques are com-putationally expensive and consequently it is not easyto apply these methods directly in online diagnosisalgorithms. In the present work, they are only usedin order to reveal interesting features of the measuredsignals and to show the most relevant harmonics.

A second approach uses the discrete Fouriertransform (DFT) to analyse the experimental data.The DFT allows discrimination of the signal contentsat different frequencies. As reciprocating combustionengines behave periodically with respect to crank-shaft angular position, this tool may be well suitedto the problem.

The result of the analysis conducted using theDFT was finally used to design ideal filters. The idealfilter is a zero-phase filter resulting from the selectionof some terms of the DFT transformation and the

Fig. 2 Engine test bench application of the inverse discrete Fourier transform(IDFT). This type of filtering can only be used offline,as an entire segment of the signal is required for

The common-rail system was a Bosch CP2, and calculating the DFT, thus resulting in an algorithmthe pressure was measured with the production that is not causal [22]; however, the quality of thesensor used for the rail pressure control (Bosch RDS). smoothing obtained from such filtering amply com-Its gain was 35.7 mV/MPa. This sensor was preferred pensates for this inconvenience. Furthermore, offlineto a research-grade pressure sensor since in this way calculation must be understood here in the sensethe diagnosis technique is tested for production con- that the result is calculated after each engine cycle.ditions, proving that no more expensive sensor is This delay does not affect online use for failureneeded. An oscilloscope was used to measure the diagnosis, providing that the diagnostic decision cansensor voltage. An angular reference was obtained by be delayed until the end of the cycle. Thus, thisrecording the injection current of cylinder 1. Engine method does not allow corrective intervention to bespeed, rail mean pressure, and fuel mass flow were applied during a given cycle. It is, however, question-also measured for each test. The sampling frequency able whether such intervention would be safe andwas 50 kHz. desirable – the one-cycle delay penalty is clearly not

a serious shortcoming of this approach.

3 SIGNAL ANALYSIS AND PROCESSING

Signal processing is a major focus of the present 4 RESULTS AND DISCUSSIONwork. The rail pressure signal contains both low- andhigh-frequency information. Low-frequency signal 4.1 Common-rail test benchcomponents are related to the filling and emptying

The first experiments were performed on theof the rail. Higher-frequency information is relatedcommon-rail test bench presented in section 2.1. Ato the injection events. Different data treatmentsummary of the experiments is presented in Table 1.techniques are explored and evaluated in this paperThe main objective of these tests was to determineto propose an effective method for the diagnosis ofwhether the rail pressure variations due to the fuelinjection systems.injection(s) were measurable. A first experiment wasFor the first approach, a time–frequency study isconducted using the Kistler research-grade pressureused by means of the spectrogram. The spectrogramsensor to obtain distortion-free measurements; theis based on the short-time Fourier transform (STFT)sampling frequency was selected to be 100 kHz toand is capable of showing local details that mayavoid aliasing effects due to the rail natural frequency,not be evident when analysed only in the time orexpected to be in the range of 2250 Hz for longi-frequency domain. The spectrogram has been widely

used for non-stationary analysis, and its interpretation tudinal propagation waves (see below). Figure 3(a)

JAUTO34 © IMechE 2006 Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from

350 F Payri, J M Lujan, C Guardiola, and G Rizzoni

Table 1 Operating conditions for the tests performed in the common-rail test bench (BTDC – beforetop dead centre)

Rail pressure Pilot i. timing Main i. timing Pilot i. duration Main i. duration Engine speed#test [MPa] [°BTDC] [°BTDC] [ms] [ms] [rpm]

1 80 – 0 – 1000 20002 80 15 0 250 1000 2000

Further, the rail natural frequency can be isolated.For a rail with characteristic length l and with speedof sound a, the natural frequency of the rail andits harmonics are expected to be located at thefrequencies

fr(k)=ka

2l

Given the physical length of the rail (0.3 m), andestimating the speed of sound in the rail (approxi-mately 1300 m/s for the working pressure [23]), thefirst harmonic can be located at approximately2250 Hz. The result agrees with the experimentaldata, as can be seen in Fig. 4.

Since the injection frequency and the high-pressurepump rotational speed depend on the engine speed,the location of the peaks in the spectrum associatedwith these phenomena will also be dependent on theengine speed. If the data acquisition is performed

Fig. 3 Time evolution of the rail pressure for the in the crankshaft angle domain, the peaks will becommon-rail test bench tests: (a) without pilot located at fixed positions (orders). On the other hand,injection; (b) with pilot injection. Injections

the peaks of the rail natural frequencies are relatedhave been marked with arrowsto the speed of sound and are further pressure andtemperature dependent, since the sound speedvaries slightly with these factors. No interference

shows the results of this first experiment. The high- between these signal components is expected, givenpressure pump effect on the rail, which appears as a that the locations of the low-frequency engine speedwave at 3 times the cycle frequency, can be clearly dependent peaks and the high-frequency pressureseen. After the injection pulse, the rail pressure dependent peaks are sufficiently distant.experiences a significant drop. This pressure drop The spectrum of the pressure signal does notcould be a useful diagnostic feature of the injection clearly reveal local injection events. Thus, time–process. Absence of such a drop when the ECU has frequency analysis was performed by means ofcommanded an injection signal would be a clear spectrograms. This analysis will help develop criteriaindication of a malfunction. for the isolation of the rail pressure drop due to local

Figure 4(a) depicts the spectrum of the pressure injection events. Figure 5(a) shows the spectrogramsignal. Since injection is performed in only one of of this first experiment. In this figure it is easy tothe injectors, the spectral peaks associated with the distinguish when the injection has occurred, asinjection are located at the frequencies the spectrogram reveals an excitation in a wide

frequency band from low frequencies up to 2 kHz.fi(k)=k f0, k=1, 2, …

This excitation is shown as a narrow black regionwhere f

0is the engine cycle frequency. The harmonics

in the level plot, indicating significant increase indue to the dynamics of the high-pressure pump are

energy contents in all the affected frequencies.located at the frequencies

A crucial question remains to be clarified: althoughthe previous results have shown that the isolation offp(k)=3k f

0, k=1, 2, …

JAUTO34 © IMechE 2006Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from

351Injection diagnosis through common-rail pressure measurement

Fig. 4 Spectra of the rail pressure for the common-rail test bench tests: (a) without pilot injec-tion; (b) with pilot injection

a single injection is possible, it is not clear that such engine cycle. This situation is rather different fromreal operating conditions, where injections are per-an strategy can deal with two injections performed

in a short time period. Problems in the detection formed in all the cylinders (note that the differentinjections will be done at different rail pressures,are expected, especially when the effect of the

first injection is small compared with that of the as the number of cylinders and the pump strokesper cycle are primes between them). Thus, as thesecond. This is the case for a pilot injection just a

few degrees before the main injection. To study emptying and filling process is a main factor in railpressure evolution, it is not possible to study thethis phenomenon, a second experiment with two

injections was performed. problem without performing the injection in all of thecylinders. The next section is devoted to analysingFigures 3(b) and 4(b) show the time evolution and

the spectrum of the pressure signal respectively. In data obtained under realistic operating conditions.the case of the time domain plot [Fig. 3(b)], a smallerdrop is observed before the pressure drop due to the

4.2 Engine test benchmain injection. The spectral plot of Fig. 4(b) does notshow any appreciable difference between the single Several experiments were run on the engine test bed,

varying the rail pressure, the engine speed, and theand pilot–main injection cases. Figure 5(b) shows thespectrogram of the signal. In this case, the area with duration of the main injection and pilot injection. A

condition with no pilot injection was also measured.high-frequency contents is divided into two differentparts: a thin black line caused by the pilot injection Finally, to study the effects of a failure in one

injector, the engine was run with one of the injectorsand a wider one related to the main injection. Basedon this last figure, it can be observed that, even when disconnected. A summary of the experiments is

presented in Table 2. To illustrate the results, onlythe detection of the injection process seems to bepossible, discrimination between several injections tests 3, 13, and 15 will be described; these correspond

to experimental points with pilot injection, with awill not be easily achievable if these injections arevery near one another. disconnected injector, and fault free but without pilot

injection respectively (from here on these casesIn the tests performed on the common-rail testbench, only one injector is working. Thus, the will be referred to as a, b, and c). All rail pressure

measurements were carried out with the productioninjection caused a significant drop in the rail pressurewhich is compensated for during the rest of the Bosch sensor.

JAUTO34 © IMechE 2006 Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from

352 F Payri, J M Lujan, C Guardiola, and G Rizzoni

sensor compared with the research sensor. Figure 6also depicts the results of the application of a low-pass ideal filter (cutting frequencies above 500 Hz).In this figure the injection events have been markedand a pressure drop associated with each injectionis easily distinguished. Additionally, it can be inferredthat the pressure drop is somewhat proportional tothe mass of fuel injected, since the pressure drop isconsiderably smaller in the case of the pilot injection.

In Fig. 7 the DFT of the signal is shown. Easilydistinguishable peaks are associated with the injectionevents (4 f

0) and the rail natural frequency, f

r. The

peaks related to the pumping process (3 f0) are

masked by the injection-related events, and the onlydistinguishable harmonic is the one that is commonto the pumping and the injection process (12 f

0).

High-frequency noise appears in the 10–20 kHzband; these peaks could be related to the sensorcharacteristics or transverse natural frequencies ofthe rail.

The spectrogram (Fig. 8) clearly shows the injectionevents in the fault-free tests (cases a and c). As in theprevious section, the injection causes a significantvariation in the signal contents in a wide low-frequency band. It should be noted that the railnatural pressure is also excited, causing an increasein the contents in the f

rand 2 f

rregion. In the case

of the faulty injector (case b), the discrimination ofFig. 5 Spectrogram of the rail pressure for the the pilot injections corresponding to the cylinders

common-rail test bench tests: (a) without pilot preceding and following the faulty one (in the firinginjection; (b) with pilot injection order) is not so clear; this effect will be analysed in

the section devoted to STFT analysis.It should be evident at this point that the railIn Fig. 6 the time evolution of the signal is

pressure is visibly modified by the injection event.represented for each one of the selected tests. In thisFor the moment, only subjective criteria based oncase, the noise is much more evident owing to the

reduced accuracy and resolution of the production the spectrogram observation have been considered.

Table 2 Operating conditions for the tests performed on the engine test bench

Rail pressure Pilot i. timing Main i. timing Pilot i. duration Main i. duration Engine speed#test [MPa] [°BTDC] [°BTDC] [ms] [ms] [rpm]

1 30 32.67 −2.3 365 625 12862 30 30 −4.99 365 625 12863 40 39.4 0.25 360 760 12864 44 42.20 1.9 340 870 12865 49 43.2 2.85 310 940 12866 46.5 32.9 −3 250 550 17507 50 42.5 −0.02 265 700 17508 64 39.8 1.6 265 860 17509 50 36 0.9 225 450 2500

10 60 37 −1.24 205 490 250011 67 38.67 −1.33 209 560 250012 83 43.9 1.75 200 690 250013(*) 41.5 36.5 −0.9 300 620 128614 62 44.3 4.5 168 465 350015 102 – 7.2 – 588 3500

*One of the injectors was disconnected in order to simulate a severe injection fault.

JAUTO34 © IMechE 2006Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from

353Injection diagnosis through common-rail pressure measurement

Fig. 7 Spectra of the pressure for the engine test benchtests: (a) pilot and main injection (test 3);(b) pilot and main injection with a faulty cylin-der (test 13); (c) main injection only (test 15)

The objective is now to design an algorithm for theisolation of the injection events from the rail pressuremeasurement. Such an algorithm could be integratedwith the OBD logic for the injection system diagnosis.Furthermore, if the failure is due to an extremelyshort duration of the injection current that does noteffectively open the injector nozzle, then a correctiveaction could be taken, progressively increasing theinjection time until a proper injection takes place. Inthis last case, the integration of the algorithm in thecontrol system may improve engine performance.

5 DETECTION ALGORITHM

Different approaches to the design of the detectionalgorithm will now be presented. The objective of thepresent work is not to present a definitive algorithmfor fuel injection diagnosis, but to provide a pre-liminary study for further work. It may be that noneof the basic algorithms proposed fits the designrequirements (in terms of robustness, detectionrate, and overdetection rate); in this case it may be

Fig. 6 Time evolution of the pressure for the enginepossible to use the integration of different criteria [24].test bench tests: (a) pilot and main injection

(test 3); (b) pilot and main injection with a faultycylinder (test 13); (c) main injection only (test 5.1 DFT analysis15). The original signal is represented above,

The analysis of the signal spectrum reveals problemsand the filtered signal below. Injections haverelating to the periodicity of the signal. When therebeen marked with arrows and faulty injections

with crosses is a lack of injection in one of the cylinders, the DFT

JAUTO34 © IMechE 2006 Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from

354 F Payri, J M Lujan, C Guardiola, and G Rizzoni

between cases a and b. However, this may not be aproper tool to determine the number of injectionsper cylinder or if a pilot injection has failed: thespectrum of case c does not differ qualitatively fromthat of case a.

Although the methods proposed in the nextsections show better performance in locatingindividual injections, the high accuracy and robust-ness of this method for detecting the total lack ofinjection in one cylinder should be borne in mind.In addition, if the DFT calculation is applied to asingle frequency, the method is more amenable toonline implementation.

5.2 STFT analysis

As shown earlier in this paper, local injection eventsare distinctly revealed by the spectrogram. Thetime representation of the spectrogram for a givenfrequency (which corresponds to a specified harmonicof the short-time Fourier transform with the windowmoving with time) could be used for diagnosis of theinjection. Also, in this case the computation couldbe reasonably fast, since there is no need to calculatethe whole spectrogram but just the STFT componentat the selected frequency (or frequency range).

The specific frequency must be selected in such away that the content of the signal at that frequencyis significant only when an injection is taking place.It is also important that the selected frequency makesit possible to distinguish several injections performedby the same injector within a short time. Figure 9shows an example of the method, where the STFTcontent of the signal at 683 Hz is shown. The figureillustrates that the method works properly except inthe case of pilot injections near the faulty cylinder(grey arrows in case b). This result is attributed tothe change in rail pressure caused by the total lackof one injection, as can be observed in Fig. 6(b); theemptying and filling process masks the effects of

Fig. 8 Spectrogram of the pressure for the engine test minor (pilot) injections. In case b, the pressure dropbench tests: (a) pilot and main injection (test 3); caused by the second and third pilot injections(b) pilot and main injection with a faulty cylin- (numbered in the firing order) is slower and smaller,der (test 13); (c) main injection only (test 15) thus hindering STFT-based detection. This effect was

also present in the spectrogram of Fig. 8(b).

can easily reveal it, as a peak appears in the cycle 5.3 Ideal filtering and further differentiation offrequency. However, isolation of local events is poor the signalowing to the nature of the DFT algorithm.

Finally, a low-pass ideal filter was performed using theReturning to Fig. 7 and analysing the low-frequencyDFT. The filtered signal was differentiated in a secondband, it can be seen that the total lack of injectionstep and its sign was changed (thus marking dropsin one of the cylinders can be determined from the

comparison of the peak at the firing frequency (4 f0) in the rail pressure as a positive peak in the resulting

JAUTO34 © IMechE 2006Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from

355Injection diagnosis through common-rail pressure measurement

Fig. 9 STFT analysis of the pressure for the engine testbench tests: (a) pilot and main injection (test 3); Fig. 10 Ideal filtering and further derivation of the(b) pilot and main injection with a faulty cylin- pressure for the engine test bench tests:der (test 13); (c) main injection only (test 15) (a) pilot and main injection (test 3); (b) pilot

and main injection with a faulty cylinder(test 13); (c) main injection only (test 15)

signal). Finally, the proposed diagnosis algorithmopens a time window each time an injection is per-formed and checks if a threshold is reached in thefiltered and differentiated signal. The time window 6 SUMMARY AND CONCLUSIONSshould consider the needle opening delay, whichcould exceed a few hundreds of microseconds A diagnosis principle for the injection process in

common-rail diesel engines, based on the analysis ofdepending on the injector model and the operatingconditions. slight variations in the rail pressure signal, has been

proposed in this paper.A convenient selection of the low-pass filter limitsand threshold permits the detection of the different From measurement of the instantaneous variation

in the common-rail pressure it is possible to detectinjections. Figure 10 depicts a case in which the cut-off frequency was 450 Hz for cases a and b and when an injection has taken place. The effect of the

injections overlaps that of the pumping process.500 Hz for case c. It can be seen that it is possibleproperly to detect the pilot injection. In the same However, with proper signal processing, it is possible

to discriminate between effects due to the injectionway the lack of injection in one of the injectors iseasily detected. It remains to be determined where and to pump pressure fluctuations.

It is possible also to discriminate between differentthe optimum threshold is, and whether it is constantfor different engine operating conditions. Note that, injections in the same cylinder even when they are

separated only by a few crankshaft degrees. This pointfor the case of the main injections, this method hasa lower signal-to-noise ratio than the STFT-based could be crucial for advanced injection techniques

(pilot injection, post-injection, HCCI engines, splitmethod; however, the detection of the pilot injectionis improved in comparison with the previous method. injections, etc.).

JAUTO34 © IMechE 2006 Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from

356 F Payri, J M Lujan, C Guardiola, and G Rizzoni

5 Zhang, L. A study of pilot injection in a DI dieselThere is a relationship between the pressure dropengine. SAE paper 1999-01-3493, 1999.in the rail and the quantity of fuel injected. Although

6 Benajes, J., Molina, S., and Garcıa, J. M. Influencethis topic has not been studied specifically, it doesof pre- and post-injection on the performance and

not seem impossible to design fuel mass flowrate pollutant emissions in a HD Diesel engine. SAEobservers from the rail pressure variation. However, paper 2001-01-0526, 2001.it must be noted that the injector is fitted with a 7 Glenn, R. and Foster, D. E. The effect of split

injection on soot and NOx production in an engine-return line to the low-pressure system; thus, the railfed combustion chamber. SAE paper 932655, 1993.pressure drop will be affected by the returned fuel

8 Beatrice, C., Belardini, P., Bertoli, C., Lisbona, M. G.,quantity too, and consequently a correction shouldand Rossi Sebastiano, G. M. Diesel combustionbe used. This issue should also be considered in thecontrol in common rail engines by new injection

diagnostic process, as minor defects in the injected strategies. Int. J. Engine Res., 2002, 3(1).fuel quantity can be hidden by the returned fuel 9 Schmidt, M., Kimmich, F., Straky, H., andquantity effect on the rail pressure. Isermann, R. Combustion supervision by evaluating

the crankshaft speed and acceleration. SAE paperThe joint use of ideal (non-causal) filtering and2000-01-0558. 2000.differentiation of the rail pressure signal seems to be

10 Ganser, M. A. Common rail injectors for 2000 barthe best method for the detection of the injectionand beyond. SAE paper 2000-01-0706, 2000.events from the rail pressure measurement. Further

11 Han, J. S., Wang, T. C., Xie, X. B., Harrington, D. L.,work must be done to optimize the band-pass Pinson, J., and Miles, P. Dynamics of multiple-filter and to select the threshold. Also, the extreme injection fuel sprays in a small-bore HSDI Dieselcases must be explored (least separation between engine. SAE paper 2000-01-1256, 2000.

12 Directive 98/69/EC of the European Parliamentdetectable injections, smallest fuel quantity detect-and of the Council of 13 October 1998 relating toable, etc.). The STFT method seems most robust formeasures to be taken against air pollution bythe detection of the main injection. DFT is effectiveemissions from motor vehicles and amendingin detecting total lack of injection in one cylinder. Council Directive 70/220/EEC. L 350/1 Official

The fusion of multiple criteria could be studied for Journal of the European Communities, 1998.improving overall detection efficiency. 13 System requirements for 2004 and subsequent

model-year passenger cars, light-duty trucks, andmedium-duty vehicles and engines (OBD II), Title13, California Code Regulations, Section 1968.2,ACKNOWLEDGEMENTSMalfunction and Diagnostic, 2003.

14 Freestone, J. W. and Jenkins, E. G. The diagnosisThis work was partially funded by a PPI-00-02- of cylinder power faults in a diesel engine by fly-1808 UPV grant. The authors are grateful for the wheel speed measurement. Proc. IMechE, Part D:collaboration of the CMT-Motores Termicos staff, J. Automobile Engineering, 1986, 200(D1), 37–43.with special thanks to Antonio Peris for carrying out 15 Williams, J. An overview of misfiring cylinder engine

diagnostic techniques based on crankshaft angularthe experimental measurements and to Enriquevelocity measurements. SAE paper 960039, 1996.Moreno for his work on analysing the data.

16 Kim, Y. W., Rizzoni, G., Wang, Y. Y., and Samimy, B.Analysis and processing of shaft angular velocitysignals in rotating machinery for diagnostic appli-

REFERENCES cations. IEEE International Conference on Acoustics,Speech and Signal Processing, ICASSP 1995, Detroit,

1 Hong, H. and Torab, B. Solenoid-operated diesel Michigan, May 1995.injectors for low-viscosity liquid and gaseous fuels. 17 Azzoni, P., Cantoni, G., Minelli, G., Moro, D.,Proc. IMechE, Part D: J. Automobile Engineering, Rizzoni, G., Ceccarani, M., and Mazzetti, S. Measure-2001, 215. ment of engine misfire in the Lamborghini 533 V-12

2 Lee, H.-K., Russell, M. F., and Bae, C. S. engine using crankshaft speed fluctuations. SAEMathematical model of diesel injection equipment technical paper 950837, 1995.incorporating non-linear fuel injection. Proc. IMechE, 18 Lee, D. and Rizzoni, G. Detection of partial misfirePart D: J. Automobile Engineering, 2002, 216(D3). in IC engines using a measurement of crankshaft

3 Baik, S., Blanchard, J., and Corradini, M. Develop- angular velocity. SAE technical paper 951070, 1995.ment of micro-diesel injector nozzles via MEMS 19 Connolly, F. and Rizzoni, G. Real time estimation oftechnology and effects on spray characteristics. SAE engine torque for the detection of engine misfires.paper 2001-01-0528, 2001. Trans. ASME, J. Dynamic Syst. Measurement and

4 Durnhold, M., Endres, H., and Frisse, P. Pre- Control, December 1994, 116(4).injection – a measure to optimise the emission 20 Stumpp, G. and Ricco, M. Common rail – anbehavior of DI-diesel engine. SAE paper 940674, attractive fuel injection system for passenger car DI

diesel engines. SAE paper 960870, 1996.1994.

JAUTO34 © IMechE 2006Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from

357Injection diagnosis through common-rail pressure measurement

21 Cohen, L. Time–frequency analysis, 1995 (Prentice- and 373 K and pressures up to 100 MPa. 14thSymposium on Thermophysical Properties, Boulder,Hall).

22 Proakis, J. G. and Manolakis, D. G. Advanced Colorado, 2000.24 Liu, S., Gu, F., and Ball, A. The on-line detection ofdigital signal processing, 2nd Edition, Vol. 18, 1992

(Macmillan). engine misfire at low speed using multiple featurefusion with fuzzy pattern recognition. Proc. IMechE,23 Ball, S. J. and Trusler, J. P. M. Speed of sound and

derived thermodynamic properties of n-hexane Part D: J. Automobile Engineering, 2002, 216,391–402.and n-hexadecane at temperatures between 298 K

JAUTO34 © IMechE 2006 Proc. IMechE Vol. 220 Part D: J. Automobile Engineering

by guest on September 6, 2012pid.sagepub.comDownloaded from