DEVELOPMENT OF A VIRTUAL SOOT SENSOR IN DIESEL ENGINES

RESEARCH ENGINE MANAGEMENT AUTHORS DEVELOPMENT OF A VIRTUAL SOOT SENSOR IN DIESEL ENGINES DR. CHRISTOPHE BARRO is Research Assistant at the Aerothermochemistry and Combustion Systems Laboratory (LAV) at ETH Zurich (Switzerland). PETER OBRECHT is Head of the Experimental Group of Combustion Engine Research at the Aerothermochemistry and Combustion Systems Laboratory (LAV) at ETH Zurich (Switzerland). The increased number of actuators provides a high potential for optimisation concerning fuel economy and exhaust gas emissions. At the same time, the effort required to achieve optimum calibration has increased as well. The integration of the exhaust emission values into a closed loop control system would facilitate the process. Since a fast physical sensor for soot measurement is not yet commercially available, a new virtual soot sensor has been developed within the framework of a FVV project at ETH Zurich. This virtual soot sensor uses as inputs combustion characteristics calculated from the cylinder pressure, as well as actuator positions from the engine control unit (ECU). The calculation of exhaust soot using the virtual soot sensor is also performed online on the ECU. THE SEAL OF APPROVAL FOR SCIENTIFIC ARTICLES IN MTZ. PEER REVIEW RECEIVED 2013-12-12 REVIEWED 2014-01-30 ACCEPTED 2014-03-03 REVIEWED BY EXPERTS FROM RESEARCH AND INDUSTRY. DIPL.-ING. PETER LAUER is Research Engineer in the Department Advanced Engineering Exhaust Gas Aftertreatment at MAN Diesel & Turbo in Augsburg (Germany). PROF. DR. KONSTANTINOS BOULOUCHOS is Head of the Aerothermochemistry and Combustion Systems Laboratory (LAV) at ETH Zurich (Switzerland). 42

1 MOTIVATION 2 EXPERIMENTAL FACILITY 3 VIRTUAL SOOT SENSOR 4 RESULTS WITH THE IMPLEMENTED VSS 5 CONCLUSIONS 1 MOTIVATION In response to increasingly strict emission legislation, diesel engines have experienced significant technological development, which has led to increased complexity of the engine components and actuators. In order to decrease engine-out emissions, additional actuators have been implemented in the air and the fuel paths. As a consequence, the requirements for engine actuator control and parameter calibration have increased. Integrating the exhaust gas emissions into a feedback control loop would help to reduce the calibration effort in order to fulfil the emission legislation. An emission feedback control loop requires a continuous monitoring of the raw exhaust emissions. For closed loop control of NO x emissions, a fast physical sensor in combination with an observer is can be used [1]. For soot on the other hand, such a physical sensor is not yet commercially available. Therefore, a model based approach, which predicts the particulate matter (PM) raw emissions in a cycle resolved manner is required. This article outlines the development and evaluation of a virtual soot sensor (VSS) and its integration in the emission feedback control loop [1]. The article is based on the results of the FVVproject [2], which has been performed during the collaboration between the Institute for Dynamic Systems and Control (PM and NO x feedback controller) and the Aerothermochemistry and Combustion Systems Laboratory (virtual soot sensor). 2 EXPERIMENTAL FACILITY Experiments have been carried out on a production-type, six-cylinder, 3-l diesel engine (OM642) provided by Daimler, ❶. The engine employs a cooled exhaust gas recirculation (EGR), a variable geometry turbine (VGT), an intake port shut-off system (IPSO) and a common rail injection system. For bypassing the standard ECU functions and data acquisition, an Etas ES910 rapid prototyping module is used. One of the engine cylinders is fitted with an Optical Light Probe (OLP, prototype from Kistler). This sensor allows crank-angle resolved measurements of the in-cylinder soot mass, which is expressed by a so called kl factor (evaluated using three-colourpyrometry). Furthermore, the OM642 engine has been fitted with a NO x sensor from Continental and an AVL Micro Soot Sensor (PASS = photo acoustic soot sensor). For combustion analysis purposes, the engines are fitted with a pressure measurement system from Kistler. 2.1 ONLINE COMBUSTION ANALYSIS DEVICE For the real-time estimation of the combustion characteristics, an embedded device has been developed in-house, which samples and processes cylinder pressure data. A more detailed description of the device can be found in [3]. 3 VIRTUAL SOOT SENSOR The goal of the VSS is to predict the raw exhaust soot concentration. In order to be able to predict accurate soot con centrations outside its calibrated range, the VSS reflects the phenomenology of the soot formation and oxidation process, as ❶ Sketch of the OM642 engine with the most relevant components 06I2014 Volume 75 43

RESEARCH ENGINE MANAGEMENT ❷ Model concept including approx. injection rate and heat release rate and representative soot evolution (kl-evolution) and an exemplary model trace at engine operation conditions of 5 bar BMEP and 1300 rpm (schematic) is the case in similar models developed in the past, such as those presented in [4, 5, 6]. The calculation time of the VSS must be short enough to enable its integration in a feedback control loop. 3.1 MODEL The Soot formation and oxidation can be simply described in three consecutive processes phases; a soot formation-dominated phase, an equilibrium phase and an oxidation-dominated phase. These three phases have been experimentally evaluated using the in-cylinder three-colour-pyrometry measurement. The obtained kl-factor reflects the soot mass evolution. Using the in-cylinder pressure analysis, which is processed in real time, characteristic points of the heat release rate are used as a model input. The points used by the model as inputs are listed below, ❷: : start of combustion : maximum heat release rate : characteristic time at 37 % (= circa 1/e) of the maximum heat release rate in the decreasing flank : timing, corresponding to 10 % of the maximum heat release rate in the decreasing flank. These parameters define the characteristic heat release rate time scales (Δφ 1 Δφ 3 ) and are the needed for the calculation of the soot model time scales (Δφ 4 Δφ 6 ). The model equations are formulated as below, Eq. 1-3: Formation: EQ. 1 m Soot, Form = b 1 m fuel, diff b 2 The soot formation becomes proportional to the fuel mass consumed in the diffusion flame. 44 Equilibrium phase: EQ. 2 m Soot, Equilibrium = m Soot, Form ( 1+0 ( Δφ 5 Δφ φ 2 ref ) ) { The modelled soot mass does not change during the equilibrium phase (therefore, the factor 0). Its duration is described by Δφ 5. Oxidation: EQ. 3 m Soot, End = Δφ m Soot, Eq ( 0,01+exp ( B b 12 Δφ 3 φ ref )) 6 B = ( T ox T ref ) b3 (1+b4 EGR stoic ) b5 ( b 6 λ 2 ) b7 ( p rail b8 p rail,ref ) ( 5 ( rpm sin (IPSO π 2 ) ) b9 { m fuel,ref ) b11 rpm ref ) b10 ( m fuel The soot oxidation phase is represented by an exponential function. Its argument reflects on the one hand the oxidation velocity as a function of turbulence, reaction kinetics and the engine load. On the other hand, it reflects the oxidation duration, described by Δφ 6. Turbulence generators include the fuel injection pressure (p rail ), the intake port shut-off (IPSO) and the engine speed (rpm). The reaction kinetics are described as a combination of the mean global in-cylinder temperature during the oxidation (T ox ) and the oxygen availability. The latter is divided into the stoichiometric

❸ Measured (red) and modelled (blue) exhaust soot mass, and measured (red) and modelled (blue) soot formation of the complete engine map (calibration) as well as for a single engine parameter variation between 5 and 8 bar BMEP at 1250 to 2500 rpm (validation) EGR-rate and λ. The parameters b1-b12 needs to be optimised for each different engine individually. The reference values are used to keep the single factors non-dimensional [7]. 3.2 STEADY-STATE RESULTS The model can be calibrated using a steady-state engine map only. Transient effects as well as different steady-state conditions are covered using the extrapolation of the model. ❸ shows the results of the calibration (engine map containing 71 operating points) and the steady-state validation, whereas single actuators (for example the EGR-valve) have been changed in order to change the engine operating conditions and the corresponding soot emissions. In addition, the figure shows a comparison between the modelled maximum formed soot and the maximum of the kl-value. The difference between soot formation and the soot remaining after expansion (end value) is the result of soot oxidation. The correlation coefficient R 2 of 0.85 for the end values was achieved and is considered sufficient for the intended use. The correlation between the modelled soot formation and kl max shows a value of 0.94. Both results indicate that the phenomenology of the soot formation and oxidation processes are well captured in the model. ❹ Top: results of VSS (blue) during a load step from 2.5 to 10 bar BMEP at 1300 rpm in comparison to kl end (black) and PASS (red, circa 2 s delay); bottom: related measurement of the fuel mass and λ 06I2014 Volume 75 45

RESEARCH ENGINE MANAGEMENT 3.3 TRANSIENT RESULTS In order to demonstrate the model s performance in transient engine operating conditions, the transient response under a load step has been tested. The soot emissions of each single cycle are calculated using the VSS and measured with kl-value at the end of combustion (kl end ). The evaluation of the kl end -value is not universally applicable and only valid in a relative regime in similar engine operating conditions. The absolute values are recorded ❺ Steps in the PM reference for emission control using the VSS (left: NO x (observer, measurement and reference); right: soot (VSS, measurement and reference); top: emissions, bottom relevant actuators (adapted from [8])) ❻ Comparison of conventional engine operation with VSS-based emission control on a part of the FTP-72 cycle 46

❼ Detail of the FTP-72 for VSS-based emission control (top: emissions; middle: relevant actuators; bottom: corresponding engine speed and load) using a PASS (photo acoustic soot sensor). The soot of a single cycle cannot be measured accurately using the PASS, since a dilution and filtration occurs during the gas transportation from the cylinder to the sampling point and measurement device, which lasts approximately 2 s. ❹ shows a comparison of the VSS and kl end for the cycle resolved values, and PASS for the steady-state values at 1300 rpm. The load step occurs from 60 to 240 Nm within 1 s. The lower part of the figure shows that during the transient, λ is lower than the steady-state level at 240 Nm. The resulting spike in soot emissions is visible in both the measurement methods. The kl end is accurately approximated by the VSS. The filtration and the dilution of PASS are clearly visible in the measurements. However, the steady-state values before and after the load step are also accurately reflected. More transient results can be found in [8]. 4 RESULTS WITH THE IMPLEMENTED VSS A comparison between the VSS-based emission control and conventional engine control on the first part of FTP-72 cycle is shown in ❻. The measured emissions of PM and NO x in the top plots show similar levels, while the cumulative emission mass flows in plots three and four indicate a lower NO x level and a similar PM level. The bottom plots show the trajectories of engine speed and load which indicate the operating point. Note that the actuation of the IPSO is based on the estimated PM level from the VSS solely. The VSS works without any information on the measured PM level. ❼ shows in detail a part of the FTP-72 with the reference emission trajectories, the estimated emission levels from VSS for soot emissions and observer for NO x -emissions respectively, and the measured emission levels, as well as the manipulated variable trajectories for SOI and IPSO. The fluctuating behaviour of the VSS is caused by fluctuations in the model inputs (for example EGR and Δφ 6 ) and due to the fast changing IPSO position, set by the controller. With the integration of the VSS in the PM- and NO x -controller of [1], the feedback soot value which was previously provided by the PASS in combination with an observer is replaced by the VSS. ❺ shows a PM reference step when using the VSS signal as the controlled variable. The top plots show the measured NO x and PM (blue) and the reference trajectories (black). The controlled variables (red) are the estimated NO x and PM from the observer and the VSS, respectively. The bottom plots show the manipulated variables SOI (start of injection) and IPSO. The NO x impact of the IPSO is compensated by the NO x -controller such that the observed and measured emissions remain close to the reference. The PM step following the reference PM is effectuated with the IPSO only. The estimated PM trajectory from the VSS has a slight offset to the measured trajectory. 5 CONCLUSIONS In the framework of the FVV project Soot Controlled Diesel Engine [2], a feedback control structure for PM and NO x engineout emissions has been designed [1]. Since suitable productiontype sensors for soot emissions are not currently available, and thus an observer cannot be used, a virtual soot sensor (VSS) has been developed. The VSS estimates soot emissions which are cycle resolved in real time using engine parameters that are available in the ECU, as well as combustion characteristics, which are calculated from in-cylinder pressure data. The model calculates the soot emissions in three consecutive steps, which correspond to the in-cylinder soot evolution, measured using three-colour pyrometry. Only twelve parameters of the VSS need to be calibrated 06I2014 Volume 75 47

RESEARCH ENGINE MANAGEMENT using a steady-state engine map. The VSS provides accurate results in steady-state as well in transient conditions and also outside of the calibrated range without any adjustment of the model parameters. The phenomena of soot formation and oxidation are reflected in a realistic manner. The VSS has been integrated in the emission feedback controller of [1]. The engine operating strategy can be adjusted by adjusting the emission set points only. Furthermore, the feedback control loop has been successfully tested on a driving cycle, showing improved results compared to the original feed-forward control strategy. REFERENCES [1] Tschanz, F.; Amstutz, A.; Steuer, J.; Guzzella, L.: Regelung der Schadstoffemissionen von Dieselmotoren. In: MTZ 75 (2014), No. 2, pp. 80-86 [2] Barro, C.; Tschanz, F. et al.: Rußgeregelter Dieselmotor. FVV-Abschlussbericht zum Vorhaben No. 986, Frankfurt a. M., 2012 [3] Tschanz, F. et al.: Cascaded multivariable Control of the Combustion in Diesel Engines. IFAC, E-COSM12, Paris, 2012 [4] Schubiger, R. A.: Untersuchungen zur Rußbildung und -oxidation in der dieselmotorischen Verbrennung. Zürich, Eidgenössische Technische Hochschule, Dissertation, 2001 [5] Warth, M.: Comparative Investigation of Mathematical Methods for Modeling and Optimization of Common-Rail DI Diesel Engines. Zürich, Eidgenössische Technische Hochschule, Dissertation, 2007 [6] Kirchen, P.: Steady-State and Transient Diesel Soot Emissions: Development of a Mean Value Soot Model and Exhaust-Stream and In-Cylinder Measurements. Zürich, Eidgenössische Technische Hochschule, Dissertation, 2008 [7] Barro, C.; Obrecht, P.; Boulouchos, K.: Development and Validation of a Virtual Soot Sensor. Part 1: Steady State Engine Operation. In: International Journal of Engine Research, 2014, published online [8] Barro, C.; Obrecht, P.; Boulouchos, K.: Development and Validation of a Virtual Soot Sensor. Part 2: Transient Engine Operation. In: International Journal of Engine Research, 2014, for publication accepted THANKS The project #986 Soot Controlled Diesel Engine has been supported by the FVV and by the Swiss Federal Office for the Environment. The authors would like to thank the supporting organisations for enabling this project. Furthermore we would like to thank the FVV research group with the other chairmen Dr. Josef Steuer (Daimler AG), Mr Christian Dengler and Mr Johan Eldh (both Daimler AG) for their help on engine issues. Further the authors would like to thank Dr. Frédéric Tschanz and Dr. Alois Amstutz from the Institute for Dynamic Systems and Control of ETH Zurich. 48

The Whole Spectrum of Media at Your Fingertips. Best Ad Media spreads your message via professional journals, online solutions and textbooks Specialised books in print and e-book form, events and other media formats tailored to your individual needs. As part of the renowned specialised publisher Springer Fachmedien Wiesbaden GmbH we help you reach decision-making target groups in business, technology and society. All the solutions for your communication needs - at your fingertips! www.best-ad-media.de Find the right gear. 06I2014 Volume 75 49