5 Diagnosis of diesel engines

Transcription

1 5 Diagnosis of diesel engines In order to develo advanced fault-detection and diagnosis systems for diesel engines a modular structure is used, which is based on the available inut and outut signals from the ECU and comrises the actuators and engine comonents and their sensors, comare Tables 3.3. to and Fig Then, so called detection modules are defined which allow the fault detection of the interacting comonents by using their signal readings. This leads to the detection modules for the intake system, injection, combustion, and exhaust system, as shown in Fig The detection modules generate certain symtoms as deviations of features from the normal (fault-free) behavior. These symtoms are then the basis of the fault diagnosis for the overall engine, using fault-symtom relationshis for the comonents of the engine and their interrelations indicating the tye and location of the faults. An imrovement of engine diagnosis systems over the imlemented OBD functions is obtained by using the inherent roerties of measured signals and relations between the signals. Therefore, model-based methods are described in the following sections which use both, signal models and rocess models. These models must be able to exress the influence between faultless and faulty behavior. Because only relatively few measured signals are available in mass-roduction engines a grouing into the mentioned detection modules will be made. Several signals are eriodic because of the reetitive cylinder charging and combustion oerations. Hence, methods of harmonic signal analysis, like Fourier or Wavelet analysis, can be alied. The rocess models, relating inut and outut signals, use as a basis hysical models, which have to be simlified, because of realtime comutational demands. However, as hysically based models of the engine include nonlinear fluidic and thermodynamic rocesses many arameters have to be estimated exerimentally, because they are not known in advance and vary strongly. Then it may be more straightforward to use rocess models based on identification and arameter-estimation methods. If the structure of these models is based on simlified hysical models, this results in semi-hysical models, Töfer et al (2002). Because of the strongly nonlinear behavior of combustion engines, also neural-network aroaches may be used. Here, a ractical feasible tye is the use of local linear models, based on arameter-estimation methods and weighting with radial basis func- Sringer-Verlag GmbH Germany 207 R. Isermann, Combustion Engine Diagnosis, ATZ/MTZ- Fachbuch, DOI 0.007/ _5

2 34 5 Diagnosis of diesel engines inut signals actuators u th u sfa throttle late intake air flas comonents intake air system blow-by air comressor intercooler signals symtoms T detection modules a, diagnosis system ṁ air 2, T2 s af n eng faultdetection module intake air system S int, i electronic control unit u mv u cv u inj metering valve ressure control valve injectors common rail injection system common rail high ressure um re-feeding. and fuel filter combustion and mechanics cylinders crankshaft cr I mv I cv u inj cam n eng cam u inj faultdetection module injection system faultdetection module combustion S inj,i S com,i fault diagnosis fault symtom tables classification methods inference methods camshaft O 2 u egr u t egr valve turbine actuator exhaust gas system egr ath exhaust gas turbine exhaust gas aftertreatment O 2 egr,tegr 4, segr, st u inj 2, T2 ṁair, neng faultdetection module exhaust gas system S exh,i Fig Modular structure of a diesel-engine fault-diagnosis system with four detection modules. tions (e.g. LOLIMOT method). The rocedure described in the next sections was mainly develoed by Schwarte et al (2004), Kimmich (2004), Kimmich et al (2005), Clever (20), Eck et al (20), and Sidorow (204). The available sensor signals deend on the tye of the diesel engine. A first difference comes from the kind of fuel injection. Cam-driven distributor injection ums with axial or radial istons for all cylinders or unit injector ums for each cylinder ossess solenoid valves for determining ilot injection, main injection and ost injection. The electronic control is either integrated in the um device or art of the engine ECU and uses (internally) in addition to the begin of injection and the fuel

3 5.3 Common-rail injection system Common-rail injection system High ressure injection systems lay a central role in any direct injection internal combustion engine. Many faults, malfunctions and failures of the engines origin in the highly stressed injection um, common-rail ressure control and injectors. Therefore, it is of considerable interest to detect and diagnose early inciient faults before they lead to intermittent and ermanent drastic faults and even failures of the injection followed by failures of the combustion rocess. In the frame of the OBD diagnosis the injection system is already suervised with regard to some emission-related faults. Hence, larger faults of the sensors, the rail ressure control deviations and wire connections are already checked and examined for lausibility. However, a detailed fault detection and diagnosis is usually not rovided. In the following a model-based fault-detection module for an electronically controlled common-rail injection system for diesel engines is described. It is the result of an in-deth case study, Clever and Isermann (2008), Clever (200), Clever and Isermann (200), Clever (20). The investigated injection system is the tye CPH from Bosch. A corresonding scheme is deicted in Fig Measurements are made on a test bench with a four-cylinder diesel engine, Oel/GM,.9l, 0 kw, 35 Nm, as deicted in Fig. 5.4.a). The dynamic models of the comonents of a high ressure common-rail system were already established in Isermann (204), Sect , resulting in the signal flow chart, Fig , for the common-rail ressure. camshaft driven high ressure um n eng metering valve including coil current measurement I mv ressure control valve including coil current measurement I cv fuel filter including fuel temerature measurement T f common rail ressure sensor cr damer re-feeding fuel um injectors u inj tank Fig Scheme of a common-rail injection system.

4 62 5 Diagnosis of diesel engines l A mv n eng high ressure um and metering valve V h A cv ressure control valve V cv + - E V cr cr cr cr u inj injectors V inj Fig Signal flow chart for the common-rail ressure, Isermann (204) Analysis of the rail ressure signal The common-rail ressure consists of a mean value according to the reference variable of the ressure control loo and eriodic oscillations through the discontinuous delivery of the radial iston um and eriodic oening and closing of the injectors. The mean value as well as the oscillation are subject to changes if faults in the injection system arise. Before corresonding fault-detection features are stated, first the signal characteristics of the oscillations are analyzed. Because these oscillations are comosed of eriodic signals with different frequencies, they are modeled by a Fourier series with a ν, b ν ω 0 = 2πf 0 f 0 y(t) = a a ν cos(νω 0 t) + b ν sin(νω 0 t) (5.3.) ν= Fourier coefficients characteristic angular frequency fundamental frequency. ν= The Fourier coefficients are defined as a ν (νω 0 ) = 2 T y(t) cos (νω 0 t)dt, (5.3.2) T 0 b ν (νω 0 ) = 2 T y(t) sin (νω 0 t)dt. (5.3.3) T 0 The time eriod T needs to be a multile of the fundamental eriod T = r, r =, 2,... (5.3.4) f 0

5 5.3 Common-rail injection system 63 Given the Fourier coefficients, the amlitude of the oscillation for a characteristic angular frequency can be calculated by A y (νω 0 ) = a 2 ν + b 2 ν. (5.3.5) In order to get the coefficients from samled data, (5.3.) has to be discretized: a ν (νω 0 ) 2 L L k=0 b ν (νω 0 ) 2 L L k=0 y(kt 0 ) cos (νω 0 kt 0 ), (5.3.6) y(kt 0 ) sin (νω 0 kt 0 ), (5.3.7) L = T T 0, (5.3.8) where T 0 is the samling time and N the signal length. Figure shows the amlitude sectra of a common-rail ressure sensor signal while the engine was in a stationary oeration oint in overrun (no fuel injections) and the common-rail ressure was in oen loo. The values of the main variables for the oeration condition are also shown. The shown measurement is mainly characterized by an oscillation with the angle eriod main um cycle: τ ist = 80 CS. (5.3.9) This oscillation is forced by the high ressure um. Because the radial iston um consists of three istons, and is driven by the camshaft belt at a ratio of 2:3 relating to the crankshaft, each 80 CS one um element delivers fuel leading to the main um cycle. Usually, the three um elements deliver the same amount of fuel in a stationary oeration oint. If the delivery quantity of at least one element differs, an additional oscillation aears (see Fig ). Since each of the three um elements delivers fuel once in 540 CS, the cycle angle of this oscillation is second um cycle: τ h = 540 CS. (5.3.0) If fuel is injected, the considered four-cylinder engines oens the injectors each 80 CS, yielding the main injection cycle: τ inj = 80 CS (5.3.) which has the same cycle angle as the main um cycle, see Fig If the volume flow to at least one injector differs due to different recycle flows or injection quantities, the cycle angle of additional oscillations becomes second injection cycle: τ bank = 720 CS. (5.3.2) This second injection cycle now differs from the second um cycle.

6 64 5 Diagnosis of diesel engines Acr [bar ] common rail ressure [bar ] second um cycle amlitude sectrum frequenzy [Hz] harmonic of second um cycle (540 ) (270 ) (80 ) angular frequency (eriodic angle [ CS]) [ C S [ main um cycle detail of the common rail ressure signal used for the calculation duration [ C S ] oeration oint common rail ressure controlled in oen loo essential variables n eng V inj I cv I mv values 800 rm 0 mm 3 cyl.05a 0.87A Fig Measured common-rail ressure signal and its amlitude sectrum in overrun. The basic oscillation is forced by the three um elements of the camshaft belt driven radial iston ressure um. If the delivery quantity of at least one of the three um elements is different from the others and at the same time the volume flow to at least one of the four injectors differs from the others, all of the above discussed oscillations are resent in the signal but have different amlitudes. This leads to several combinations of different fuel delivery quantities on the one hand and different volume flows to the injectors on the other. These combinations reeat after the eriod suerimosed cycle: τ su = τ h,bank = CS = 260 CS. (5.3.3) Because the amlitudes of the common-rail ressure oscillations forced by the high ressure um and the injections are small comared to the mean common-rail ressure, it can be assumed that the volume flows through the ressure control valve and through the leakages are constant for a stationary oerating oint:

7 5.3 Common-rail injection system 65 Acr [bar ] second um cycle amlitude sectrum frequenzy [Hz] harmonic of second um cycle main um cycle common rail ressure [bar ] (540 ) (270 ) (80 ) angular frequency C S (eriodic angle [ CS]) [ ] detail of the common rail ressure signal used for the calculation duration [ C S] oeration oint common rail ressure controlled in oen loo essential variables n eng V inj I cv I mv values 800 rm 0 mm 3 cyl.5a 0.72A Fig Measured common-rail ressure signal and its amlitude sectrum in overrun. One um iston delivers less fuel, leading to a second um cycle with eriod 540 CS. V cv + V leak constant for a stationary oeration oint. (5.3.4) Thus, these volume flows do not cause additional oscillations of the common-rail ressure. Figure illustrates the discussed oscillations with the cycle angles τ inj, τ bank, τ ist, τ h and τ su. The amlitude of the main um oscillation at 80 CS deends on the delivered fuel and angular seed. The delivered fuel to the high ressure um from the low ressure um is feedforward controlled by the metering valve. This valve determines the fuel flow rate to the um in order to save uming energy and unnecessary heating-u of the fuel, Robert Bosch GmbH (2005). Therefore, the high ressure um has some non-delivery rotation angles 0 χ 80 CS. As shown in Clever (20), the amlitudes of the common-rail ressure change in deendence on the engine seed and metering valve osition resectively valve current

8 66 5 Diagnosis of diesel engines Acr [bar ] common rail ressure [bar ] amlitude sectrum frequenzy [Hz] second injection cycle main injection cycle and main um cycle (720 ) (80 ) angular frequency C S (eriodic angle [ CS]) [ ] detail of the common rail ressure signal used for the calculation duration [ CS] oeration oint common rail ressure controlled in oen loo essential variables n eng V inj I cv I mv values 300 rm mm 3 cyl.05a 0.80A Fig Measured common-rail-ressure signal and its amlitude sectrum for active injectors. One injector injects more fuel than the others, leading to a second injection cycle with eriod 540 CS. 4 bar A cr (80 CS) 5 bar for overrun and can be reresented as normal values in a look-u table. The measured oscillations of the common-rail ressure finally are a suerosition of the individual induced oscillations by the high ressure um and the injectors. The suerosition leads to following measurable eriods: first sum eriod second um eriod second injector eriod second sum eriod τ ist,inj = 80 CS τ h = 540 CS τ bank = 720 CS τ h,bank = 260 CS (5.3.5)

9 5.3 Common-rail injection system 67 oscillation forced by injectors high ressure um inj bank su 2340 CS ist 360 h su 2340 CS Fig Illustration of the oscillations forced by the volume flows through the high ressure um and the injectors. (Each number indicates the injector which is currently active and the um element which currently delivers fuel, resectively. In order to illustrate the oscillations caused by differences of the volume flows to the injectors on the one hand and differences of the delivery quantities of the three um elements on the other, different amlitudes of the forced oscillations are assumed. For the injector flows it is assumed that the volume flows to the injectors and 4 are smaller than the flows to the injectors 2 and 3. For the delivery quantities of the um elements it is assumed that the delivery quantity of um element 2 is smaller than the delivery quantities of the other um elements). where it holds for the resulting frequencies with f = n eng τ/360. For n eng = 800rm = 30rs one obtains first sum frequency second um frequency second injector frequency second sum frequency f ist,inj = 60Hz f h = 20Hz f bank = 5Hz f h,bank = 5Hz (5.3.6) Model-based fault diagnosis A failure mode and effect analysis (FMEA) with faults of the fuel filter, metering valve, um mechanics, ressure control valve, injector including seals and ossible failure effects on the engine is reorted in Clever (20). Based on this analysis, artificially introduced faults have been selected for exeriments. The fault detection is based on outut residuals between measured or calculated quantities and their normal values, see Fig a) Mean common-rail ressure Figure shows the used signal flow diagram of the ressure build-u in the common-rail. In order to remove the oscillations of the common-rail signal, it is low-ass filtered. Based on hysical modeling leading to the model structure, a common-rail ressure model can be identified from measured data. But not all of the inuts in Fig are available as measurements. For this reason the model structure, deicted in

10 68 5 Diagnosis of diesel engines cr, I mv, common rail I cv u inj, n, eng ressure mean value r inj, cr smooth fuel delivery r inj,2 cr cr, I mv, n eng smooth injection fuel delivery r inj,3 r inj,4 Fig Measured inut variables and residuals for the model-based fault detection of the common-rail injection system. Fig , is used. It is assumed that the ressure before the metering valve is constant. Thus, it is not used as an inut for the model anymore. Further it is assumed that the oening cross section of the metering valve and the ressure control valve are aroximately roortional to the measured current through the electromagnetic actuators. Thus, the measured currents through the two valves are used as model inuts. The hysteresis of the osition of the ressure control valve is taken into account by offset values for the maximum hysteresis width. I mv n eng I cv u inj f cr cr E cr (400bar,60 C) Ecr( cr, Tcr) cr,corr Fig Signal flow scheme for the common-rail ressure with used measured inut signals. By means of a model for the bulk modulus, the measured common-rail ressure is related to a bulk modulus at a articular ressure (e.g. 400 bar) and at a certain temerature (e.g. 60 C). Also the fuel temerature is calculated based on a model. The data-based modeling and identification of the common-rail ressure is conducted using the net model LOLIMOT with the inuts: I mv I cv n eng u inj suly current of the metering valve suly current of the ressure control valve engine seed desired injection quantity On basis of the estimated model following arity equation for the mean commonrail ressure can be formulated:

11 5.4 Turbochargers with wastegate and variable geometry 73 Table Fault-symtom table for the common-rail system. 0, +, : symtom is zero, ositive,, : yes, no. S inj, S inj,2 S inj,3 S inj,4 S inj,5 S inj,6 isolable F inj, low delivery quantity of one um iston F inj,2 reduced injection quantity of one injector F inj,3 ressure loss in front of high ressure um (e.g. a lugged fuel filter) F inj,4 ressure in front of high ressure um too high (e.g. a faulty metering valve) F inj,5 oening of the ressure control valve is too large F inj,6 oening of the ressure control valve is too small F inj,7 ressure sensor signal is too high F inj,8 ressure sensor signal is too low are not necessary for the diagnosis. However, since the symtom S inj,4 is calculated in a wider oeration range, it may be advantageous to use it in the diagnosis rocedure in order to be able to diagnose the faults F inj,5 and F inj,6 earlier. Table shows that all faults can be distinguished from each other if the fault detection results are combined in a single fault-detection system. The fault-detection method for the common-rail system was described for a fourcylinder engine with a three-iston high ressure um. However, the method can also be adated to engines with three and five to eight cylinders and high ressure ums with one or two istons, Clever (20). As the highest frequency resectively the shortest eriod is 80 CS, see (5.3.5), the samling eriod should be 90 CS according to Shannon s samling theorem. Exeriments with a samling eriod of 60 CS have shown the same results as in Fig , Clever (20). The described fault-diagnosis method was filed to atent, Clever and Isermann (200). 5.4 Turbochargers with wastegate and variable geometry As all modern diesel engines and increasingly also gasoline engines are equied with charging units which are mechanically driven suerchargers or exhaust-gas turbochargers, the monitoring and fault diagnosis of these highly stressed comonents is mandatory. Usually, the charging ressure 2i and charging air temerature T 2i is measured. Together with the air mass flow rate ṁ air this allows to determine the

12 74 5 Diagnosis of diesel engines ower of the comressor with efficiency mas. These mas can be used for fault detection. In the case of turbochargers, the suervision of the turbine ower additionally requires the measurements of the exhaust ressure 3 and exhaust temerature T 3 ustream the turbine, in the exhaust manifold, and several efficiency mas. For modeling turbochargers the steady-state mas delivered from manufacturers can be used. However, they are frequently not recise in the required engine oeration range and do not consider the ulsating exhaust gas flow. Therefore more comrehensive and dynamic models of turbochargers will be considered. As a basis for model-based fault detection the nonlinear dynamic models of variable-geometry turbine (VGT) chargers and wastegate (WG) chargers will be considered in the next section. The models are derived for the diesel engine shown in Fig. 5.4.a) with the variables shown in Fig. 5.4.b) Models of VGT turbochargers Dynamic models of turbochargers can be derived based on thermodynamic changes of state or based on fluid dynamic aroaches by using Euler s equation for turbomachinery. Both modeling ways are described in Isermann (204), Zahn (202) and Sidorow (204). For fault detection thermodynamic models for the turbine and comressor ower will now be referred, because they need fewer arameters to be identified, Sidorow (204). It is assumed that the engine is in normal warm state. The dynamic behavior at one oerating oint [u inj, n eng ] results from the balance equation for the angular momentum dω tc (t) J tc = M t (t) M c (t) M f (t) (5.4.) dt with torques M t for the turbine, M c for the comressor and M f for the friction, see Fig Turning the equation to owers with P = Mω leads to J tc ω tc (t) dω tc(t) = P t (t) P c (t) P f (t). (5.4.2) dt The friction torque is assumed to be dominated by viscous friction M f = c f ω tc and therefore P f (t) = c f (T oil ) ω 2 tc. (5.4.3) The turbine ower follows according to P t = ṁ t c t T 3 η t,is ( ( 4 3 ) κ ) κ (5.4.4) with the mass flow determined by a throttle equation 2 ṁ t = A t,eff 3 RT3 ψ( 4 ) (5.4.5) 3 ψ( ( ) [ 4 4 κ ) = c ( ) κ ] t 4 κ. (5.4.6) 3 R 3 3

13 5.4 Turbochargers with wastegate and variable geometry 75 air filter HFM T a T T 4 T 5 a 4 5 m n HFM tc DPF exhaust throttle T 2c 2c intercooler T 2ic 2ic m 2 T 2i 2i HP-EGR valve HP-EGR cooler egr m HP-EGR T 3 3 throttle m eng b) VSA actuator cylinders Fig a Scheme of the investigated diesel engine with common-rail direct injection and VGT turbocharger, Oel/GM,.9l, 0 kw, 35 Nm. Actuator and sensor signals as used on the IAT test bench. b Used variables for the diagnosis of turbochargers.

14 76 5 Diagnosis of diesel engines 3 T 3 * A t,eff 3 RT * 3 turbine 3 4 T * m t c t,is T * 3 P t M t tc M f c f m m c J tc m c c T c,is * 2 P c tc M c m t rotor tc comressor ma 2 a ( t exh m ) exh 4 exhaust ie comressor Fig Signal-flow chart for the thermodynamic model of a turbocharger (without internal heat transfer). tc t

15 5.4 Turbochargers with wastegate and variable geometry 77 The effective cross section area A t,eff (s t, 4 ) is a function of the actuator osition s t and the ressure 4 after the turbine. T 3 is the gas temerature before the turbine reduced by the heat transfer loss to the comressor and the environment. The isentroic efficiency of the turbine is defined as η t,is = T 4 T 3 ( ) κ 4 κ 3. (5.4.7) It is difficult to determine this efficiency for low turbocharger mass flows and seeds, because the heat transfer to the comressor becomes effective in these cases. Therefore it is determined exerimentally by a ma η t,is (s t, u ref ), see Fig , where u ref is the turbine blade seed ratio, Guzzella and Onder (200), u ref = u c u,max = d t3 π ( ( ) 2c T ) ω tc, κ κ which is roortional to the turbocharger angular seed ω tc, see also Isermann (204). 0.8 t,is s t [-] u ref [-] Fig Turbine isentroic efficiency ma, aroximated with a LOLIMOT net model, Sidorow et al (20). Hence, following variables have to be measured to calculate the turbine ower P t (u inj, n eng ) z T t = [ 3, 4, T 3, s t ]. (5.4.8) 4 can be determined by the measured ressure dro of the articulate filter 4 = + f (ṁ t ). (5.4.9)

16 78 5 Diagnosis of diesel engines For larger mass flows, T3 = T 3 can be assumed. The comressor ower is ( (2 ) P c = ṁ c c c T η c,is ) κ κ (5.4.0) where ṁ c is taken from a comressor mass flow ma ṁ c (n tc, 2 / ), see Fig , or is directly measured with ṁ c = ṁ air. 0.4 m c,corr [m 3 /h] n tc,cor [/s] 2c / Fig Comressor mass flow ma..8 2 T is the inlet temerature, taking into account the heat transfer from the turbine T = T + k 3A c c c ṁ c (T 3 T ) (5.4.) with the heat transfer coefficient k 3 and the effective surface area A c. The isentroic efficiency of the comressor is defined as η c,is = ( ) κ 2 κ T 2 T (5.4.2) and it is exerimentally determined in form of a ma η c,is (ṁ c, n tc ), see Fig The required measurements needed to calculate the comressor ower P c (u inj, n eng ) are z T c = [ ṁ air, T,, 2 ]. (5.4.3) To make the turbocharger characteristics indeendent of the changing environmental conditions, the variables are usually stated as referenced or corrected quantities using fluid mechanical laws. The reference conditions for the comressor inlet are T ref = 93K, ref = 0.98bar and for the turbine inlet they are T 3ref = 873K. This leads to the corrected rotational seed

17 htt://