Feedback control of particulate matter and nitrogen oxide emissions in diesel engines

Control Engineering Practice 21 (2013) 1809–1820

Contents lists available at ScienceDirect

Control Engineering Practice journal homepage: www.elsevier.com/locate/conengprac

Feedback control of particulate matter and nitrogen oxide emissions in diesel engines Fre´de´ric Tschanz n, Alois Amstutz, Christopher H. Onder, Lino Guzzella Institute for Dynamic Systems and Control, Department of Mechanical and Process Engineering, Swiss Federal Institute of Technology, ETH Zurich, Sonneggstrasse 3, 8092 Zurich, Switzerland

a r t i c l e i n f o

abstract

Article history: Received 11 January 2012 Accepted 21 September 2012 Available online 18 October 2012

The continuing reduction of the emissions of diesel engines has caused an increasing complexity of calibration and expensive aftertreatments of the exhaust gas. These issues have a potential for being relaxed if the NOx and particulate emissions are integrated into a feedback loop. For this purpose, a novel controller is developed. A model-based observer for the emissions is used to overcome the relatively slow dynamics of the available sensors. Furthermore, the controller includes self-calibration of the EGR. The experimental validation of the control structure shows that different emission strategies are feasible with just a minimal calibration effort. & 2012 Elsevier Ltd. All rights reserved.

Keywords: Diesel engines Model-based control Automotive emissions Engine control Engine management

1. Introduction In recent years, automotive diesel engines have experienced a significant technological development resulting in increased requirements on calibration and control (Guzzella & Amstutz, 1998). This development has been driven by the tightening of the legislative limits set on the pollutant emissions. To reduce these emissions, additional components have been implemented in the air and fuel paths. Examples are cooled exhaust gas recirculation (EGR), the variable nozzle turbocharger (VNT), swirl valves, and the common rail injection system. To date, in production-type engines, the management of these systems has been performed without any information on two main variables of interest, namely the amounts of nitrogen oxide ðNOx Þ and particulate matter (PM) emitted. In order to remain within the limits set by legislation despite this lack of information, the calibration complexity has increased significantly. This drawback is even accentuated due to drift effects due to external influences in the medium and long timescale. Examples of such influences are the production spread and the aging of components or variations of the ambient conditions and fuel quality. The drawbacks with respect to the missing feedback of the engine-out emissions can be summarized as follows:

 The feasible design range within the legislative emission limits is reduced to compensate for emission variations resulting

n

Corresponding author. Tel.: þ41 44 6322464; fax: þ41 44 6321139. E-mail addresses: [email protected] (F. Tschanz), [email protected] (A. Amstutz), [email protected] (C.H. Onder), [email protected] (L. Guzzella). 0967-0661/$ - see front matter & 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.conengprac.2012.09.014

 

from drift influences such as aging, production spread, or fuel quality variations. A substantial number of compensation functions need to be calibrated in order to limit the influences of measurable drift sources such as variations in the ambient conditions. Engine-out emission information for optimized aftertreatment operation is missing.

With aftertreatment devices such as selective catalytic reduction (SCR) catalysts or diesel particle filters, a significant reduction of the tail-pipe emissions has been achieved. Nevertheless, due to the limited conversion efficiency of a NOx reduction device (Johnson, 2009), the need for appropriate urea dosing, and the additional fuel required for filter regeneration (Salvat, Marez, & Belot, 2000), well-balanced and known engine-out emissions are still desirable. The contribution of this paper is a novel control architecture that includes feedback of the engine-out emissions of NOx and PM in order to address the issues of drift influences and calibration. In particular, the control structure allows the emissions to be set in a wide range with a significantly reduced calibration effort. As a result, the cumulative NOx and PM engine-out emissions on the new European driving cycle (NEDC) can be placed effectively and accurately within the legislative limits to enable the desired emission strategy. For the measurement of particulate matter, production-type sensors are not available as yet. However, production-type PM sensors are expected to become available in the near future (Steppan et al., 2011). For this study, an AVL Micro Soot Sensor was used for the PM measurements. Due to the layout and installation of this sensor, a significant delay and a filtering effect

1810

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

are present on the measured signal. The same effects are present on the measured signal of the NOx emissions, but the delay is substantially smaller. Particularly for the measured PM signal, the delay is a highly limiting factor for an efficient feedback control. To overcome the issues in connection with these dynamic characteristics of the emission sensors, a model-based observer for the instantaneous engine-out NOx and PM emissions has been developed. The paper is organized as follows: in Section 1.1, a short overview to preceding contributions in the field of emission control is presented. Section 2 describes the experimental setup. In Sections 3 and 4 the design of the model and of the observer and the development of the controller are explained and analyzed, respectively. Section 5 contains the experimental validation of the controller. The paper closes with the conclusions and an outlook. In the following the term ‘‘emissions’’ refers to the engine-out emissions.

Fig. 1. Sketch of the engine with the most important components.

1.1. Review First steps in the field of emission control were presented by Ruckert et al. (2005) with a model predictive control structure including NOx feedback for the air path of heavy duty diesel engines. Alfieri (2009) proposed a control structure to jointly control the NOx and the oxygen content of the exhaust. A model predictive approach for controlling the air path while limiting the NOx was proposed by Stewart and Borrelli (2008). A study with a model predictive approach, including the torque, the center of heat release, the maximum heat release rate, the NOx , and the PM (opacity) as controlled variables, was presented by Karlsson, Ekholm, Strandh, Johansson, and Tunestal (2010). A further study applying model predictive control was presented by Deng, Xue, Stobart, and Maass (2011). The work presented here contributes to the attempt to reduce the pollutant emissions with an innovative way of integrating both the PM and the NOx into a control structure that has its primary focus on these emissions.

2. Experimental facility Experiments have been carried out on a production-type 6-cylinder 3-liter diesel engine provided by Daimler AG. The calibration of the engine is intended for use with a particle filter, an oxidation catalyst and a SCR catalyst, neither of which are installed on the test bench. The back pressure increase resulting from these devices is simulated by throttling the exhaust flow. The relevant characteristics of the engine are listed in Table 1. For bypassing and data acquisition the engine control unit (ECU) is accessed with an ETAS ES910 rapid prototyping module. The standard sensors of the engine have been supplemented by several additional sensors. The most relevant of which are the NOx sensor (Continental, Uninox24V) and the PM sensor (AVL Micro Soot Sensor), which are located downstream of the turbine. Fig. 1 shows a sketch of the test bench architecture.

Table 1 Engine characteristics. Type Technical data Power Torque (limited) Features

OM 642, Euro 5 2987 cm3, six cylinders 160 kW @ 3800 rpm 400 Nm Cooled EGR, VNT, swirl valves, common rail injection

Fig. 2. Block diagram of the basic processes relevant for the engine-out NOx and PM emissions. Inputs: intake pressure pim intake temperature Wim , BG ratio xbg , rail pressure pinj , SOI, injection amount qinj , swirl valve position usv .

3. System analysis and modeling 3.1. Control-oriented analysis of the emission formation The appropriate choice of manipulated variables or actuators is crucial in order to obtain the desired effect on the emissions to be controlled without compromising other requirements. Fig. 2 shows a simplified block diagram with the processes relevant for the engine-out emissions in a conventional diesel engine. It illustrates the fact that the instantaneous emissions within the cylinder are essentially defined by the local temperature development and by the local availability of oxygen. The local temperature is a function of the combustion subsystem, which includes the ignition delay and the premixed and diffusion combustion. It is influenced mainly by the start of injection (SOI), the pressure and temperature of the gas, the local concentration of evaporated fuel cevap , and the local concentration of oxygen cox . The local concentration of oxygen is mainly influenced by the dilution of the cylinder mixture with burned gas1 (BG) and by the turbulence-influenced transport phenomena of the gas.

1 The stoichiometric burned-gas part of the recirculated exhaust serves as inert gas.

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

3.2. NOx control As described above, the NOx emissions are essentially defined by the local temperature trajectory and by the availability of oxygen, which are influenced by the intake conditions, the BG rate, the SOI and the turbulence. Helmantel, Somhorst, and Denbratt (2003) have shown that the influence of the turbulence on the PM oxidation outweighs its influence on the amounts of NOx emitted. Those results have been verified with experiments on the test bench. Turbulence generators thus are not suitable for NOx control when the PM is also considered. The intake pressure pim also is not suitable for NOx control since it is not controllable at low speed and load regions with the given engine configuration. The BG rate and the SOI thus are the remaining candidates for controlling the NOx emissions. Fig. 3 shows the impact of positive and negative variations of SOI and BG rate at various operating points (OP). As expected, the injection timing and the BG rate have a monotonous influence on the NOx . While the SOI influence on the PM is small, the BG rate clearly has a strong impact on the PM. Controlling the NOx with the SOI therefore appears to be a promising approach. However, the fact that the SOI also influences the thermal efficiency, as well as the exhaust gas temperature and the combustion noise has to be taken into account. Furthermore, the SOI is not influenced by drift. Therefore, the control of the NOx level with the SOI alone can lead to a deterioration of the overall engine performance in case of drift. 3.3. PM control Based on the considerations described in Section 3.1, the first-sight candidates as manipulated variables for the control of PM emissions are the BG rate, the boost pressure and turbulencegenerators such as the swirl valve,2 and the fuel injection. The fuel injection based turbulence is influenced by the pressure in the common rail and the characteristics of the injection, i.e. the number of injections and the amount of fuel that is injected in each event. Splitting up the injection into multiple events is of interest due to the facts that the injection-based turbulence dissipates rapidly and that the fuel distribution within the 2 The swirl valve is a butterfly valve that continuously closes one of the two intake channels of each cylinder. Due to the resulting asymmetric flow into the cylinder the swirl, i.e. the turbulence within the cylinder is increased.

102 mpm [mg/m3]

SOIvar.

102 NOx [ppm] mpm [mg/m3]

Note that the oxygen sink due to combustion is not explicitly illustrated in the block diagram. With the fuel-mixture preparation block, the processes of the distribution of the fuel droplets within the cylinder and their evaporation are represented. More detailed information about these processes and relationships can be found in Heywood (1988), for example. For the NOx emissions, the local peak temperature is most relevant. Secondarily, the local oxygen concentration plays a role in the actual reaction process with the nitrogen. For the PM emissions, the principles of formation and oxidation are complex. Details to these issues can be found in Bertola (2004) or in Tree and Svensson (2007), for example. The local concentration of oxygen cox and a sufficiently high temperature for the oxidation of the forming and growing particulates are of major importance. This means that rapid mixing, i.e., a high turbulence is required even after the combustion is completed. While for low levels of NOx , low oxygen concentrations and low temperatures are required, the conditions for obtaining low levels of PM are just the opposite, which illustrates the PM/NOx trade-off. In the following, the influence of each input on the emissions and its appropriateness for control are analyzed and discussed.

1811

102

EGRvar.

102 NOx [ppm] Fig. 3. Influence of 7 31 SOI and 7 3% BG-rate variations on the PM and NOx at various operating points.

cylinder is changed. Furthermore, pilot injections influence the portion of fuel that is burned in the PM-less premixed phase. Many of the potential manipulated variables, however, are not reasonably applicable for PM control. This is the case for the boost pressure due to its poor controllability in the range of low speed and load with its high PM emissions. The BG rate would provide a very powerful manipulated variable. However, it is in conflict with the NOx requirement. The rail pressure generally is already maximized, and further increase could lead to wall impingement resulting in a massive increase of the PM emissions and in oil dilution. Therefore, control of the PM with the rail pressure is inadvisable. The remaining candidates are the swirl valve and the characteristics of the injection, which are analyzed in more detail in the following.

3.3.1. Injection characteristics for PM control The injection parameters available are the number, the amount, and the timing of pilot injections as well as the timing and amount of an early post injection. Benajes, Molina, and Garcia (2001) and Schnorbus et al. (2008) have shown that turning-off pilot injections can result in a decrease in PM emissions. Experiments on the engine used for this work confirmed this behavior. However, whenever the amounts of the pilot injections are varied, a controller is required for the combustion in order to guarantee combustion stability under varying external conditions such as the fuel quality. The pilot injection characteristics thus are better suited for use in a combustion rather than in an emission control loop. An investigation of the potential of an early post injection for PM control yielded results similar to those reported by Benajes et al. (2001). However, the sensitivities of the PM emissions on the parameters of the post injection where found to be highly variable and thus no general rule could be found about which parameter of the post injection would cause what PM emission influence. Furthermore, an investigation of the post injection influences on the PM emissions for variations of the swirl valve position revealed even a disadvantageous PM influence of the post injection. This unexpected behavior is shown in Fig. 4, where the swirl valve position is varied with activated and deactivated post injection. The activation of the post injection is indicated in the top plot. As expected, it is followed by a PM reduction. As the swirl valve is closed stepwise from 60% to 100%, as shown in the middle plot, the PM emissions slightly drop further in the first part. However, when the swirl valve is closed from 80% to 90%,

1812

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

5 0

102

total main post

10

0

20

usv = 90%

40

60

100 usv [%]

usv = 10%

mpm [mg/m3]

qinj [mm3]

15

101

75 50

mpm [mg/m3]

0

20

40

60

30

102

20

NOx [ppm]

10 0

Fig. 6. Influence of the swirl valve at various operating points.

0

20

40

60

t [s] Fig. 4. Activating the post injection (top plot) results in an initial improvement of the PM emissions (bottom plot). As the swirl valve (middle plot) is closed, this improvement is reduced. When the post injection is turned-off, the PM emissions drop significantly.

mpm [mg/m3]

motivation for integrating the swirl valve into an emission control loop because simply closing the swirl valve could lead to increased NOx emissions. Accordingly, for a good performance in the PM/NOx trade-off, an appropriately controlled swirl valve position is required. 3.4. Emission models and observer

80

1900 [rpm]; 140 [Nm] 1000 [rpm]; 50 [Nm] 1300 [rpm]; 280 [Nm]

60 40 20 0

103

0

20

40

60

80

100

usv [%] Fig. 5. PM emissions in function of the swirl valve position.

the PM emissions rise as indicated in the bottom plot. The deactivation of the post injection at 55 s then causes a significant decrease of the PM emissions, which is also indicated in the top and bottom plots. Hence the sensitivity of the PM emissions on the post injection has changed signs during the swirl valve sweep. The utilization of the post injection for PM reduction in the case of a closed swirl valve thus might be completely counterproductive. Further investigations on the reasons for this phenomenon are continuing. However, based on the findings described above, the early post injection does not seem to be suitable as a manipulated variable in a PM control loop.

3.3.2. Swirl valve for PM control Fig. 5 shows the impact of the swirl valve position on the PM emissions in function of its closing ratio. Clearly the swirl valve has a massive impact on the PM emissions, particularly for closing ratios above 50%. This influence is present in the whole operating range, as shown in Fig. 6. Furthermore, the sensitivity of the PM emissions on the swirl valve is monotonic for all OP, which is a desirable feature for control. As a result of these considerations, the swirl valve with its desirable sensitivity characteristics and its high PM impact clearly is the best choice as a manipulated variable for PM control. However, for OP with low PM and high NOx emissions, an increasing influence on the NOx emissions is observed. This influence on both the NOx and PM emissions is an additional

As mentioned above, the sensors for the NOx and PM emissions have a relatively slow response time with respect to the process to be controlled. In order to increase the feasible bandwidth of the emission control loops, a model-based observer for the emissions of NOx and PM has been developed. The use of such model-based observers for control is common practice in the field of automotive control (Arsie, Pianese, Rizzo, & Cioffi, 2003; Powell, Fekete, & Chang, 1998). For the present work, the observer is based on control oriented models of the engine-out emissions and of the sensor dynamics. The model for the NOx emissions has been slightly adapted from the one described by Schilling, Amstutz, and Guzzella (2008). The PM model is an extension of the model presented in previous work (Tschanz, Amstutz, Onder, & Guzzella, 2010). In the following, a short explanation of the models for the sensor dynamics and for the engine-out emissions is given. The dynamic characteristics of the sensor signals can be modeled as first-order lag elements with a time delay resulting from the transportation of the gas to the sensor (Mrosek, Sequenz, & Isermann, 2011; Schilling et al., 2008; Tschanz et al., 2010). The signal zs,i of the emission sensor i, thus can be modeled with the following differential equation, where i corresponds to the emission species NOx or PM  d 1  z ðtÞ ¼   z ðtÞzi ðtDt s,i Þ : dt s,i ts,i s,i

ð1Þ

The variable zi refers to the corresponding engine-out emission which reaches the sensor after a transport delay of Dt s,i . The time constant ts,i of the first-order lag takes into account the response characteristics of the sensor, and the mixing of the gas on its way to the sensor. For the NOx sensor, a time constant of ts,nox ¼ 1:83 s and a transport delay of Dt s,nox ¼ 0:3 s have been identified. The values for the PM sensor are ts,pm ¼ 0:34 s and Dt s,pm ¼ 1:25 s. The model for the engine-out emission zi is based on the OP-dependent steady-state value z i and its relative deviation dzi . In stationary conditions, it is assumed that the engine state is defined by its OP, i.e. by the engine speed ne and by the amount of injection qinj . A first approximation of the emissions can thus be obtained by interpolating on the steady-state maps z i ðne ,qinj Þ for the NOx and PM emissions, respectively. In case of a deviation of

qinj [mm3]

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

nox, pim [-]; vpim, n = 10000 [Pa] 40 -0.02

40

35

35

35

0.08

30 25

0.08

25

0.075

0.075

0.065

15 1500

2000

0

30

20 0.065

2500

20

0.1

15

0.2

3000

1500

35

25

-0.02

-0.1

20

20 0.02

-0.2

2000 2500 ne [rpm]

3000

25

-0.2

20 -0.1

1500

0.04

0

15 2000

2500

1500

3000

2000

2000 2500 ne [rpm]

2 1

1.5

35

-0.3

30 25

2

20

1.5 1

15 2000 2500 ne [rpm]

3000

-0.1

-0.1

-0.2

2

1500

3000

pm, usv [-]

3

15 3000

2500

40

30

-0.4

15

1500

0.06

20

-0.6

35

25

0

25

pm, xbg [-] 40

-0.2

30

0.02

30

-0.7

pm, pim [-]

-0.03

-0.8

1500

-0.6 -0.03

-0.9

15 3000

35

30

15

2500

40

0.02 0

0

35

20 -0.02

nox, usv [-]; vsv, n = 10 [%] 40

-0.8

25

0 0.05

2000

-0.9

30

-0.02

pm, soi [-] 40 qinj [mm3]

nox, xbg [-]; vxbg, n = 0.1 [-]

nox, soi [-]; vsoi, n = 1 [°BTDC] 40 0.075

-0.05

1813

-0.2

-0.3 -0.6 -1

1500

-0.1

2000 2500 ne [rpm]

3000

Fig. 7. Maps of the OP-dependent sensitivities yi,j for the NOx (top) and PM model (bottom). The normalization value of each input channel is specified above the top graphs.

the engine from its stationary state at a given OP due to transient effects, the relative deviation of the emissions dzi can be modeled as a function of the OP and of the vector of model inputs dvi . This leads to the following basic structure of the models for the NOx or PM emissions i, respectively: h i zi ¼ z i ðne ,qinj Þ 1þ dzi ðne ,qinj , dvi Þ : ð2Þ The relative deviation of the emissions dzi is obtained with the vector of OP-dependent sensitivities hi of the emissions on deviations in the model input vector dvi . Additionally, the model is extended by a variable xi which contains all influences that are not reproduced by the relative deviation model dzi . This variable xi is integrated into the model in the form of an integrator that is driven by white noise n_ i . The resulting model for the engine-out emissions can be written as follows: d x ¼ n_ i dt i

ð3Þ h

i

zi ¼ z i ðne ,qinj Þ  1þ hi ðne ,qinj Þ  dvi þ xi : The vector of the model inputs dvi is composed of the deviation of each input channel dvk,i with respect to the corresponding OPdependent steady-state or reference value v k,i and a normalization value vk,n as follows:

dvk,i ¼

vk,i v k,i ðne ,qinj Þ : vk,n

ð4Þ

For the present work, the vector of the model inputs has been adapted from Schilling et al. (2008) and Tschanz et al. (2010) to take into account only the most relevant inputs. The reduction of the number of input channels is justified by the inclusion of the state xi , which takes into account unmodeled influences such as the engine temperature. The input channels for both the NOx and the PM model are the BG rate in the intake, the boost pressure, the SOI, and the swirl valve position vi ¼ ½xbg pim usoi usv . The BG rate in the intake manifold xbg is defined as the mass fraction of burned gas to the total mass of gas in the intake manifold. While

it cannot be measured directly, it can be estimated on the basis of _ air , the EGR mass flow m _ egr , the measured fresh air mass flow m and the measured air-to-fuel ratio l in the exhaust manifold, which is measured by the lambda sensor (Guzzella & Onder, 2010). A model-based estimate for the EGR mass flow is provided by the ECU. Measurements of stepwise changes in the SOI and in the swirl valve have been used for the identification of the sensor dynamics of the NOx and PM sensors, respectively. Due to the fact that a stepwise change in one of the actuators quasi-statically influences the engine-out emissions, all delay and filtering effects in the exhaust pipe and in the sensor can be identified with such measurements. The maps for the steady-state emissions z i ðne ,qinj Þ and for the steady-state model input levels v k,i have been created with stationary measurements. For the identification of the OP-dependent sensitivities hi , stationary measurements with deviated model inputs at each OP have been used. Fig. 7 shows the OP-dependent sensitivities of the NOx and PM models on the normalized input channels respectively. For example a deviation of the normalized SOI by dvsoi ¼ 1, which corresponds to a 11 shift towards early, results in a NOx increase of approx. 7.7% at an OP of 1500 rpm and 25 mm3. 3.4.1. Observer design The structurally identical models for the PM and NOx sensors and emissions summarized in (1) and (3) can be rewritten in state-space form. An additional first-order lag is integrated to approximate the discrete behavior of the engine cycles as a continuous process. This is required to avoid any direct feedthrough on the controlled variables, as will be clarified later. Furthermore, the transport delay is moved from the sensor input to its output. The system for the emission species i thus can be reformulated as follows: d x ðtÞ ¼ Ai xi ðtÞ þ Bui ðtÞ dt i yi ðtÞ ¼ Cxi ðtDt s,i Þ,

ð5Þ

1814

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

where yi ¼ zs,i is the output of the system and xi ¼ ½zs,i zi xi  represents the state with zi as the state variable for the continuous approximation of the engine-out emission species i. With the time constant for the engine cycles tcyc , the system matrices are obtained as follows: 2 3 2 3 1 1 0 7 0 0 6 7 6 ts,i ts,i 6 7 6 7 6 7 6 7 1 1 1 7, B ¼ 6 Ai ¼ 6 ð6Þ 7 0  6t 7 6 0 7 4 cyc 5 4 tcyc tcyc 5 0

0

0

0

1

C ¼ ½1 0 0: The time constant of the engine cycle is defined such that within one engine cycle (two rotations) the first-order lag reaches 95% of its steady-state value, which corresponds to three times the time constant of the first-order lag

tcyc ¼

21 : ne 3

ð7Þ

Based on the engine-out emission model summarized in (3), the input vector ui of the state-space system (5) is obtained as follows: " # z i ½1þ hi dvi  ui ¼ : ð8Þ n_ i The OP dependency of z i and hi has been omitted for better readability. A Kalman filter with the gain Li has been designed for the NOx and PM systems (5) such that a model-based estimate for the engine-out emissions z^ i is provided for control. In Fig. 8, a block diagram of the filter for the emission species i is shown. The estimated sensor signal z^ s,i has to be delayed with the respective transport delay to yield the appropriate residual.

NOx [ppm]

Fig. 8. Block diagram of the Kalman filter with the estimated engine-out emission species z^ i that is observed with the measured signal zi,s and the output of the basic engine-out emission model ui,ð1Þ .

275 200

As a result of the model design with the state variable for unmodeled influences xi , the estimate z^ i converges to the measured value zs,i . Due to the fact that the estimated engine-out emissions z^ i do not include the sensor dynamics, they allow a higher bandwidth of the controlled system. The Kalman filters have been designed at a reference speed of 1500 rpm. The rate of convergence of the filters is obtained with the assumed covariances of the process and measurement noises Q i and Ri (Kwakernaak, 1972). For the process noise covariances Q i , the identity matrix is used, while for the measurement noise covariances Ri , values of Rpm ¼ 3  104 and Rnox ¼ 5  103 have been chosen for the PM and NOx filter, respectively. The observer for the engine-out emissions in the following designates the two Kalman filters for the NOx and PM emissions. Fig. 9 shows the performance of the observer for load steps at 1700 rpm. Clearly, the estimated engine-out emission signal z^ i provides information about the emissions to be expected before the sensor value shows a reaction.

4. Controller design In this section, the new controller for the emissions of NOx and PM is presented. Compared to conventional diesel engine controllers, the advantages of this controller for the engine-out emissions are that it provides an innovative way of relaxing the following issues in automotive diesel engine management: 1. The requirement of keeping within the emission limits set by legislation with minimal compromise on other requirements such as low levels of consumption and system cost. 2. The higher complexity of calibration in order to limit drift influences on the emissions, resulting from controlling merely representative variables instead of the signals of interest, i.e. the emissions. Both issues are based on the fact that due to the lack of emission feedback, additional measures need to be taken to ensure compliance with the legislative limits and to minimize undetected drift. Measures that compromise other requirements could be an engine calibration with widened margins from the emission design point to the legislative limits and a suboptimal configuration of the aftertreatment devices with respect to the engine potential, for example. The higher complexity of calibration is a result of controlling merely representative variables instead of the signals of interest, namely the emissions. To compensate for external influences on the emissions due to ambient conditions or fuel quality, for instance, a number of additional functions need to be calibrated. Taking into account these considerations and those described in Sections 1 and 3, the following key attributes of the control structure have been determined:

125

mpm [mg/m3]

0

10

50

20

30

meas. est. ˆi obs. zˆ s, i

30



10 0

20

10

 The observer-based engine-out emissions are used as con-

30

trolled variables. The manipulated variables are the SOI and the position of the swirl valve usv for the values NOx and PM, respectively. These choices are motivated by the system analysis presented in Section 3. Based on the statements in Section 3.2, an additional adaption loop for the burned gas reference is integrated to avoid any long-term deviation of the SOI and to automatically calibrate the BG reference map.

t [s] Fig. 9. Contrasted with the measured NOx and PM emissions, the estimated engine-out emission signal z^ i evidently provides predictive information. The estimated sensor signal z^ s,i (thin line) is also shown.

The PM loop consists of a PI controller on the estimated engineout PM emissions with the swirl valve position as manipulated variable. The NOx controller also contains a proportional (P) and

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

1815

Fig. 10. Block diagram of the entire control loop including the engine, the controllers, the sensors, and the observer for the engine-out emissions. The proposed control structure consists of the PI controller for the PM and of the NOx controller with the P element on the SOI and the learning feedforward BG adaption drbg as an I element. The air-path controller is not part of the proposed control structure. Actuator saturations and anti-windup are not shown.

40 0.95

1

1.1 1.15

0.95

1

1.15

35 qinj [mm3]

an integrator (I) element. Their respective outputs, however, are not fed into the plant via one single manipulated variable but via the SOI and via the reference value for the BG rate r bg of the air-path controller. The reason for including the BG rate in the NOx loop is the fact that one of the main purposes for the recirculation of exhaust gas is the reduction of the NOx emissions (Amstutz & del Re, 1995). Therefore, the exact amount of recirculated exhaust gas is of importance mainly with respect to the resulting NOx emissions, and it makes sense to adapt the reference value of the BG rate in case the NOx emissions deviate from their reference value. However, due to the fact that the BG rate can be estimated on a significantly higher bandwidth than the NOx emissions, direct EGR valve actuation based on the NOx without a cascaded EGR control is not recommended. As shown in Section 3.2, the SOI provides an important influence on the NOx emissions. By primarily controlling the NOx with the SOI, it is therefore possible to compensate for deviations of the BG rate in transients, for example. In accordance with these considerations, the P element of the NOx controller is connected to the SOI in order to provide fast NOx control. To obtain a similar absolute NOx impact over the operating range with the P-part of the controller, the NOx error variable enox for the P-part of the NOx controller is normalized with the operating point dependent steady-state level z nox ðne ,qinj Þ. This is justified with the fact that the variability of the sensitivity ynox,soi from SOI to NOx is relatively small over the operating range. The I element is used to adapt the BG reference r bg of the air path controller such that the BG rate matches with the NOx reference for the current OP and conditions. To avoid any excessive adaption of the BG reference in transients when the NOx deviation temporarily might be high, the input to the I element of the controller is limited by a saturation. To provide automatic calibration, learning feedforward control is implemented on the BG adaption. This concept is explained in Section 4.1. The validity of the presented control structure for the combined control of the PM and NOx emissions is limited to the engine operating range where the sensitivities of the PM and NOx emissions on the manipulated variables allow for combined control with the chosen manipulated variables. This is discussed in Section 4.2. To avoid any performance deterioration in case of actuator saturation, all integrators are implemented with anti resetwindup. In particular, anti reset-windup is used to recover the BG rate reference in case of saturation of the PM controller.

30 1.05 25 1

15 1000

0.85

0.95 1500

1

0.95

20

2000 ne [rpm]

2500

0.9 3000

Fig. 11. Value of the first diagonal element of the relative gain array. The level close to one justifies the use of SISO loops.

A block diagram of the entire control structure except for the antiwindup, the saturations and the normalization is shown in Fig. 10. The permissibility of controlling the NOx and PM emissions in two SISO loops has been analyzed with the relative gain array (RGA), which represents the cross-coupling between the channels. Regarding the NOx loop, this analysis has been conducted only for the SOI manipulated variable. This is justified with the intention to adapt the BG rate reference in a slow manner. Therefore, the coupling of the BG rate to the PM is an issue only on frequencies significantly lower than the bandwidth of the PM control loop. This will be shown in Section 4.3. Fig. 11 shows the first diagonal element of the RGA that has been obtained with the 2  2 nominal plant from SOI and the swirl valve to the NOx and PM engine-out emissions. The value of the first diagonal element is close to one in the OP range considered. Therefore, since the cross-coupling within the plant for the considered input channels is small, the controller design with two SISO loops is justified. Note that the RGA-elements are frequency independent due to the fact that the considered dynamics, i.e. the continuous approximation of the engine cycles are equal on both channels.

4.1. Learning feedforward control To improve the performance of the controller and to provide automatic calibration, learning feedforward control (Vogt, Muller, & Isermann, 2004), which corresponds to automatic OP-dependent

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

calibration, has been implemented for the adaption of the BG rate reference, i.e. the I element of the NOx controller. In the following, a short explanation of the learning feedforward control implementation is given. For a detailed description, the reader is referred to Vogt et al. (2004). The starting point for learning feedforward control on a map M is the vector w that contains the values in each grid point in M. Here, wbg contains the adaptive offset of the BG rate reference in each grid point of the operating range. The value of dr bg ðne ,qinj Þ for some momentarily active OP is then obtained with the vector of interpolation weights Uðne ,qinj Þ for the current OP.

dr bg ¼ wTbg  Uðne ,qinj Þ:

ð9Þ

The elements fk of U are unequal to zero solely on the four closest nodes around the OP. Furthermore, the following equation holds: n X

fk  1, fk A ½0,1 8k:

ð10Þ

k¼1

The online modification of the vector wbg , i.e. the integration of the deviation enox in the I element of the NOx controller, is achieved with the same vector of interpolation weights U. d w ¼ Uðne ,qinj Þ  enox : dt bg

ð11Þ

4.2. Operating range of the control structure The control structure for the combined control of the PM and NOx emissions is intended for operating regions with a sufficiently high sensitivity of the PM on the swirl valve. Such conditions are found in the lower speed and lower load range of the given engine. Under conditions of a very low PM level and high NOx emissions that result for example from a very low BG-rate, the appropriateness of PM feedback control is arguable because the sensitivity of the PM on the swirl valve is low. In such regions, NOx control clearly is of highest priority while the PM can be expected to remain in a narrow band without feedback control. Fig. 12 indicates the nominal limit of operating range of the control structure for the given engine and calibration. It has been obtained with the sensitivities of the emissions on the manipulated variables. The levels in Fig. 12 correspond to the normalized DC gain from the swirl valve to the PM emissions with the NOx control loop closed. They show the sensitivity of the PM on the swirl valve plus the sensitivity of the PM on the BG-rate adaption which is required to compensate for the NOx influence of the swirl valve. As highlighted in the figure, for high speeds and higher loads, the DC-gain changes sign. This sign change is thus a result 40

0

of cross-couplings which are present in the plant on low frequencies due to slow BG-rate reference adaption of the NOx control loop, in combination with a relatively low sensitivity of the PM on the swirl valve. In the region with the reversed sign, the control structure cannot be used.

4.3. Controller tuning The following considerations have been made when tuning the controllers: 1. The cumulative PM emissions on a certain driving cycle are influenced to a high extent by transient PM emission peaks (Giakoumis & Alafouzos, 2010). Therefore, fast PM control is desirable. 2. The cumulative NOx emissions are less influenced by transients. Therefore, a lower bandwidth is acceptable on this controlled variable. 3. The bandwidth on the BG adaption loop must be well below the bandwidth of the PM control loop due to its strong crosscoupling with the PM emissions. The magnitudes of the loop gain 9Li ðjoÞ9, the sensitivity 9Si ðjoÞ9, and the complementary sensitivity 9T i ðjoÞ9 of the PM and NOx loops tuned according to the above considerations are shown in the top and bottom plots of Fig. 13, respectively. The bottom plot additionally shows the loop gain 9Lbg 9 for the BG-rate adaption channel of the NOx controller, i.e. the I element with deactivation of the P element. Clearly, the PM and NOx loop crossover frequencies, which are indicated by diamond markers, are separated by more than two octaves. Accordingly, the PM loop is significantly faster, which is illustrated by the higher bandwidth on the complementary sensitivity that is indicated by triangle markers. The disturbance rejection bandwidth, indicated by circle markers, is similarly high for both loops, which is a result of the first-order lag approximation of the engine process. When considering the NOx loop with the deactivated P element, i.e. control with the BG-rate adaption only, the crossover frequency on Lbg is further reduced. The crossover frequency on this channel is significantly lower than the bandwidth of the PM control loop and than a common bandwidth of EGR controllers (Alfieri, Amstutz, & Guzzella, 2009).

20 |Lpm| [dB]

1816

0

10−1

20 -17

-0.5 -4

-2 -4

-7

100

101

102

20

-1

|Lnox| [dB]

qinj [mm3]

25

0 -3 -10

35 30

L S T

10

-0.5

-12

Lbg

10 0 -3 -10 10−1

15 1000

1500

2000 ne [rpm]

2500

3000

Fig. 12. Normalized DC gain of the plant from the swirl valve to the PM emissions with a closed NOx control loop. The sign change of the DC gain is highlighted.

100

101

102

 [rad/s] Fig. 13. Magnitudes of the loop gain L, the sensitivity S, and the complementary sensitivity T of the PM and NOx loops. In the bottom plot, the loop gain Lbg is shown, which is obtained by considering solely the BG-rate adaption channel within the NOx controller.

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

1817

20

50

140 120

0

5

10 t [s]

15

30 20

mpm [mg/m3]

meas. ref. obs.ˆi

350

40

NOx [ppm]

160

mpm [mg/m3]

NOx [ppm]

180

300 250

10 5

200 20

10

0

5

10 t [s]

15

0

20

15

45

30 t [s]

0 0

60

15

30 t [s]

45

60

15

30 t [s]

45

60

4

xbg [-]

0

0.25

80

5

10 t [s]

15

20

xbg

0.2

rbg, corr

0.15

70 0

rbg, base

usoi [°BTDC]

90 usv [%]

SOI [°BTDC]

2

-2

mod. w/o ctr. ref. meas.

15

0.1 0

5

10 t [s]

15

0

20

Fig. 14. The top plots show the measured emissions, their reference value, and the observed engine-out emissions for stepwise changes in the PM reference. The bottom plots show the SOI and the swirl valve position, respectively.

Therefore, the two loops are also decoupled when taking into consideration the BG rate adaption channel from the NOx controller.

15

30 t [s]

45

2 0 -2 -4

60

Fig. 15. The top plots show the measured emissions, their reference values, and the trajectory of modeled engine-out emissions that would be obtained without control action on the SOI. The bottom left plot shows the unadapted BG rate reference r bg,base , the estimated BG rate xbg , and its reference trajectory which is adapted by the NOx controller. The bottom right plot shows the SOI control action dusoi .

25 mpm [mg/m3]

400 250 100

0

40

20

20

meas.

15

ref.

5 0

60

obs. ˆi

10

0

20

8

100

4

75

0

40

60

40

60

50

0

40

20

0

60

0

20

t [s]

t [s] 60

qinj [mm3]

0.3 0.2 0.1 0

60

25

-4 -8

40 t [s]

usv [%]

SOI [°BTDC]

t [s]

xbg [-]

The performance of the proposed control structure has been tested and validated with a number of experiments and with the NEDC driving cycle. Figs. 14 and 15 show the performance of the controller for steps in the PM reference value and for a BG deviation. The advantage of the observer-based controlled variables is clearly visible in Fig. 14. The estimated engine-out PM emissions z^ pm shown in the top right plot are set to the new reference value immediately, while the measured value reaches the new setpoint later due to the transport delay and filter effects that are inherent to the sensor. Fig. 15 shows the performance of the NOx control loop. A stepwise change in the BG rate at 2000 rpm and 170 Nm is shown in the bottom left plot. A deviation of the BG rate could appear in transients or, on a long timescale, due to aging for example. In a conventionally controlled diesel engine these cases could lead to significant deviation of the emissions, as illustrated by the modeled NOx emissions with deactivated emission control in the top left plot. With the proposed control architecture, the emission deviation resulting from this BG rate deviation is minimized as shown in the top left measured NOx emissions. Additionally, the SOI deviation for NOx control is eliminated by the automatic recovery of the BG reference to a value where the NOx emissions are met without SOI control action. In the case of drift, a deviation of the BG rate might not be apparent on the estimated engine-out emissions z^ nox . Accordingly, the compensation in such a case could be limited by the slower convergence speed of the observer. However, this would still be significantly faster than the rate of change of most conceivable drift sources. An experiment with load steps from approx. 200–350 Nm at 2200 rpm is shown in Fig. 16. The bottom right plot shows the amount of injection. The top left plot shows the NOx emissions which correspond well to the reference trajectory. The sensitivity of the PM emissions on the swirl valve is low, as can be seen in the top and middle right plots. The high load OP from 10 s to 37 s shown in Fig. 16 is in a OP region in the proximity to the limits of the engine and control-structure operating range. In accordance with the considerations in Section 4.2, at this OP, PM control makes limited sense. For the experiments on the NEDC, the engine governor is actuated such that the brake torque follows the reference load trajectory

NOx [ppm]

550

5. Results and discussion

0

0

40

20 t [s]

60

50 40 30 20

0

20 t [s]

Fig. 16. The top plots show the measured emissions, their reference value, and the observed engine-out emissions for load steps at 2200 rpm. The middle plots show the SOI and the swirl valve position, respectively. The bottom plots show the BGrate and the amount of injection.

of the cycle. The engine speed is controlled synchronously by the brake to follow the corresponding speed trajectory. The reference trajectories are based on measurement data of an NEDC test conducted with a vehicle equipped with an engine of the same type as the one that has been used for this work. The starting point for the emission reference maps in this work is based on the steady-state emission data at each node of the OP map and have been obtained with the conventionally calibrated and controlled engine. On the vehicle, aftertreatment devices are included that were not available for this project. Therefore, the results obtained here cannot be brought into direct context with legislative limits that the entire vehicle must comply with. The point of this work, however, is to provide a novel control structure with feedback of the engine-out emissions, such that they can be set to a certain reference value

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

NOx [ppm]

400 300 200 100 0 800

mpm [mg/m3]

while the calibration complexity to follow this emission reference is significantly reduced compared to conventional engine control. The potential and advantage of the proposed control structure is illustrated with the results described below. Fig. 17 shows a section of the urban part of the cycle. The top two plots show the measured emissions of NOx and PM with their reference values and the observer output z^ i . In plots three and four, the trajectories of the manipulated variables SOI and swirl valve are shown. The bottom pair of plots shows the trajectories of the engine speed and load, respectively. Clearly, the observer-based feedback of the PM emissions z^ pm follows the reference trajectory, except when the swirl valve is in saturation. Also for the NOx value the observed emissions z^ nox are set to the reference value by the SOI and with the appropriately auto-calibrated BG rate reference map. For the NOx emissions it has to be kept in mind that the actuation of the SOI is based on a P controller. Accordingly, some offset to the reference has to be expected in engine transients. The measured emissions (grey lines) satisfactorily correspond to the reference values when considering the slow dynamics of the NOx sensor and the significant time delay of the PM sensor. The overland part of the NEDC is shown in Fig. 18. For this higher load part of the cycle as well, the control performance on the emissions shows good results.

900

1000 t [s]

1100

100 meas. ref. obs. ˆi

75 50 25 0 800

SOI [°BTDC]

1818

900

1000 t [s]

1100

900

1000 t [s]

1100

900

1000 t [s]

1100

10 5 0 -5 -10 800

150

usv [%]

NOx [ppm]

100

100 50 0

715

705

50 25 0 800

735

725

75

225

meas. ref. obs. ˆi

40 20 0

715

705

725

1600 1200 800 800

735

Te [Nm]

2000 ne [rpm]

mpm [mg/m3]

t [s]

950 t [s]

1100

150 75 0 800

950 t [s]

1100

SOI [°BTDC]

t [s] Fig. 18. Overland part of the NEDC. The measured emissions are set to the reference value except in the case of actuator saturation.

8 0 -8 705

715

725

735

725

735

t [s] usv [%]

100 75 50 25 0 705

715 t [s] 180 Te [Nm]

ne [rpm]

1600 1200 800

705

720 t [s]

735

120 60 0

705

720 t [s]

735

Fig. 17. Section of the urban part of the NEDC. The observed engine-out emissions, and the delayed, filtered, measured emissions follow their respective reference trajectories. The trajectories of the manipulated variables SOI and swirl valve as well as the OP are shown in the lowest four plots.

The entire potential and performance of the proposed control structure are shown in Figs. 19 and 20, where the results of experiments on the entire NEDC are shown. Fig. 19 shows the emission mass flows with the emission controller in comparison to the conventionally controlled engine. To obtain the results shown in Fig. 19, the reference values of the NOx and PM emissions simply have been multiplied with a certain factor, i.e. the whole reference map has been scaled. For the PM emissions, this scaling factor corresponded to a PM reduction of 62%, while for the NOx reference an increase of 33% had to be allowed for feasibility. In the lower two plots, the normalized cumulative emissions are shown. The normalization was effected with the final value of the corresponding cumulative emission trace that has been obtained with the conventionally controlled engine. Clearly, the cumulative PM emission could be reduced significantly with this simple adaption of the reference values. It has to be noted that after the change in the emission references, the critical parts of the cycle had to be driven a few times to allow auto-calibration of the BG reference to take place. Fig. 20 shows the cumulative emissions over the entire NEDC for various emission references. For these experiments as well, the calibration work consisted solely of some simple adaption of the reference maps for the PM and NOx emissions. Clearly, for all measurements the deviation between the cumulated measured emissions and the corresponding cumulated reference values are

m ? nox [mg/s]

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

values shown in the results section and in particular in Fig. 20. With an extended design range on the emissions, a potential for improvement on other issues is provided.

20 10 0

0

200

400

600 t [s]

800

1000

600 t [s]

800

1000

2. The complexity of calibration is reduced because with the proposed control structure, OP-dependent calibration of the EGR reference and swirl valve position is provided automatically. Furthermore, the calibration of compensation functions to account for emission influences resulting from ambient condition variations, for instance, can be omitted.

m ? pm [mg/s]

4

0

200

400

130 100 70 40 10

mpm [%]

mnox [%]

Emission control ECU

2 0

0

1100

550 t [s]

130 100 70 40 10 0

1100

550 t [s]

Fig. 19. The proposed controller compared to the results obtained with standard ECU control. In the two bottom plots, the normalized cumulative emissions are shown.

0.03

PM [g/km]

1819

Euro4

0.02

 ref.  meas.

0.01 Euro6 0

0

Euro5 0.1

0.2 NOx [g/km]

0.3

Fig. 20. The cumulative measured and reference engine-out emissions over the NEDC.

small, which means that the control performance is good also over a wide range of emission references. Additionally, Fig. 20 shows that with effortless calibration work, analogously to the procedure described above, significantly differing emission strategies are feasible with the proposed control architecture.

6. Conclusions and outlook In this work it is shown that the integration of feedback of the engine-out emissions of PM and NOx into an adequate control structure relaxes issues that are present in automotive diesel engine applications. In particular, the proposed control structure provides improvements on the following issues: 1. The feasible design range within the legislative emission limits can be extended with the proposed control structure. This is due to the fact that emission variations resulting from drift and transients are reduced with the emission controller. This has been demonstrated with the good compliance of the engine-out PM and NOx emissions with their respective reference

The benefit on both issues is a results of the advantage of directly controlling the variables of interest, i.e. the pollutant emissions, instead of representative variables for the emissions. Therefore, drift influences such as production spread, aging and environmental condition variation as well as transient influences on the emissions can be reduced. With a reduced spread of engine-out emissions, a potential for improving the coordination between the engine and the configuration and operation of aftertreatment devices such as a SCR catalyst or a particle filter is offered. This is clearly illustrated with the significantly differing engine-out emission strategies shown in Fig. 20. Admittedly, the conditions of these test bench experiments do not entirely meet the requirements for a certification cycle. Nevertheless, they clearly illustrate the potential of improvement on the aftertreatment cost3 in order to meet Euro 6. Further steps in the field of engine-out emission control for diesel engines are the optimized determination of reference values for the respective emission species and the integration of control structures for the characteristics of the combustion. The engine-out emission reference values have to be determined under consideration of requirements on the entire system such as the configuration and efficiency of aftertreatment devices and depending on the legislative emission limits that have to be respected with the application. An example including costoptimal determination of the engine-out emissions has been provided in the study of Cloudt and Willems (2011). The integration of control structures for the characteristics of the combustion would provide the potential to control the processes at the source of emission generation. Therefore, such a controller would be useful on a lower level within an emission control loop. Additionally, this would further reduce the complexity of calibration on the many degrees of freedom of a modern injection system.

Acknowledgments The authors gratefully acknowledge the financial support by the Forschungsvereinigung Verbrennungskraftmaschinen e.V. (FVV) and the Swiss Federal Office for the Environment (BAFU). Furthermore, we thank Dr Josef Steuer, Mr Christian Dengler and Mr Johan Eldh of Daimler AG for their support on engine issues and the FVV research group for the helpful discussions and advice.

Appendix A. Abbreviations BG ECU EGR NEDC OP PM

burned gas electronic control unit exhaust gas recirculation new European driving cycle operating point particulate matter

3 Aftertreatment cost is related to the costs of the devices (-size) and of their operation due to consumption of urea and fuel for filter regeneration.

1820

RGA SCR SOI VNT

F. Tschanz et al. / Control Engineering Practice 21 (2013) 1809–1820

relative gain array selective catalytic reduction start of injection variable nozzle turbocharger

Appendix B. Mathematical notation ci mpm ne qinj p Te usv w xbg zs,i Dt dv

U W y

t x zi

local concentration of i (–) concentration of PM in the exhaust (mg/m3) engine speed (rpm) injected fuel amount per cylinder (mm3) pressure (Pa) engine brake torque (Nm) swirl valve position (%) vector of grid points in some map intake burned-gas ratio (–) observed/modeled sensor signal i transport delay (s) input vector to emission models (–) vector of interpolation weights (–) temperature (K) sensitivities of the emission models (–) time constant (s) state for unmodeled emission influences observed/modeled raw emission species i

References ¨ Alfieri, E. (2009). Emissions-controlled diesel engine. Ph.D. Thesis ETH Zurich. Nr. 18214. Alfieri, E., Amstutz, A., & Guzzella, L. (2009). Gain-scheduled model-based feedback control of the air/fuel ratio in diesel engines. Control Engineering Practice, 17, 1417–1425 (Special Section: The 2007 IFAC Symposium on Advances in Automotive Control). Amstutz, A., & del Re, L. (1995). EGO sensor based robust output control of EGR in diesel engines. IEEE Transactions on Control Systems Technology, 3, 39–48. Arsie, I., Pianese, C., Rizzo, G., & Cioffi, V. (2003). An adaptive estimator of fuel film dynamics in the intake port of a spark ignition engine. Control Engineering Practice, 11, 303–309. Benajes, J., Molina, S., & Garcia, J. M. (2001). Influence of pre- and post-injection on the performance and pollutant emissions in a HD diesel engine. In SAE world congress, 2001-01-0526. Bertola, A. G. (2004). Technologies for lowest NOx and particulate emissions in ¨ DI-diesel engine combustion. Ph.D. Thesis ETH Zurich. Nr. 15373.

Cloudt, R., & Willems, F. (2011). Integrated emission management strategy for cost-optimal engine-aftertreatment operation. SAE International Journal of Engines, 4, 1784–1797. Deng, J., Xue, Y., Stobart, R., & Maass, B. (2011). Fuel path control of a diesel engine at the operating point of the low load and medium speed. In Chinese control and decision conference (CCDC) Chinese (pp. 747 –751). Giakoumis, E., & Alafouzos, A. (2010). Study of diesel engine performance and emissions during a transient cycle applying an engine mapping-based methodology. Applied Energy, 87, 1358–1365. Guzzella, L., & Amstutz, A. (1998). Control of diesel engines. IEEE Control Systems Magazine, 18, 53–71. Guzzella, L., & Onder, C. H. (2010). Introduction to modeling and control of internal combustion engine systems (2nd ed.). Berlin: Springer-Verlag. Helmantel, A., Somhorst, J., & Denbratt, I. (2003). Visualization of the effects of post injection and swirl on the combustion process of a passenger car common rail DI diesel engine. In ASME conference proceedings36789 (pp. 631–640). ASME. Heywood, J. B. (1988). Internal combustion engine fundamentals. McGraw-Hill International Editions. Johnson, T. (2009). Review of diesel emissions and control. International Journal of Engine Research, 10, 275–285. Karlsson, M., Ekholm, K., Strandh, P., Johansson, R., & Tunestal, P. (2010). Multipleinput multiple-output model predictive control of a diesel engine. In Advances in automotive control. Munich: IFAC. Kwakernaak, H. (1972). Linear optimal control systems. New York: John Wiley & Sons Inc. Mrosek, M., Sequenz, H., & Isermann, R. (2011). Identification of emission measurement dynamics for diesel engines. In 18th IFAC world congress. Milano, Italy. Powell, J., Fekete, N., & Chang, C.-F. (1998). Observer-based air fuel ratio control. IEEE Control Systems Magazine, 18, 72–83. Ruckert, J., Richert, F., Abel, D., Herrmann, O., Pfeifer, A., Schlober, A., & Pischinger, S. (2005). A model based attempt to control boost pressure and EGR-rate for a heavy duty engine. AT: Automatisierungstechnik, 53, 79–86 German. Salvat, O., Marez, P., & Belot, G. (2000). Passenger car serial application of a particulate filter system on a common-rail, direct-injection diesel engine. In SAE world congress. 2000-01-0473. Schilling, A., Amstutz, A., & Guzzella, L. (2008). Model-based detection and isolation of faults due to ageing in the air and fuel paths of common-rail direct injection diesel engines equipped with a lambda and a nitrogen oxides sensor. Proceedings of the Institution of Mechanical Engineers Part D: Journal of Automobile Engineering, 222, 101–117. ¨ Schnorbus, T., Pischinger, S., Korfer, T., Lamping, M., Tomazic, D., & Tatur, M. (2008). Diesel combustion control with closed-loop control of the injection strategy. In SAE world congress. 2008-01-0651. Steppan, J., Henderson, B., Johnson, K., Khan, M. Y., Diller, T., Hall, M., Lourdhusamy, A., Allmendinger, K., & Matthews, R. (2011). Comparison of an on-board, realtime electronic pm sensor with laboratory instruments using a 2009 heavy-duty diesel vehicle. In SAE world congress. 2011-01-0627. Stewart, G., & Borrelli, F. (2008). A model predictive control framework for industrial turbodiesel engine control. In 47th IEEE conference on decision and control. (pp. 5704 –5711). Cancun. Tree, D. R., & Svensson, K. I. (2007). Soot processes in compression ignition engines. Progress in Energy and Combustion Science, 33, 272–309. Tschanz, F., Amstutz, A., Onder, C., & Guzzella, L. (2010). A real-time soot model for emission control of a diesel engine. In Advances in automotive control, AAC 2010 Munich. IFAC. Vogt, M., Muller, N., & Isermann, R. (2004). On-line adaptation of grid-based look-up tables using a fast linear regression technique. Journal of Dynamic Systems, Measurement, and Control, 126, 732–739.