Models everywhere/How a fully integrated model-based test environment can enable progress in the future Mongi Ben Gaid, Romain Lebas, Morgan Fremovici, Grégory Font, Antoine Albrecht, Guénaël Le Solliec

Abstract The aim of this paper is to demonstrate how advanced modelling approaches coupled with powerful tools allow to set up a complete and coherent test environment suite. Based on a real study focused on the development of a Euro 6 hybrid powertrain with a Euro 5 turbocharged diesel engine, the authors present how a diesel engine simulator including an in-cylinder phenomenological approach to predict the raw emissions can be coupled with a DOC&DPF after-treatment system and embedded in the complete hybrid powertrain to be used in various test environments: -

-

coupled with the control software in a multi-model multi-core simulation platform with test automation features, allowing the simulation speed to be faster than the real-time exported in a real time hardware in the loop platform with the ECU and hardware actuators, embedded at the experimental engine test bed to perform driving cycles such as NEDC or FTP cycles with the hybrid powertrain management.

Thanks to these complete and versatile test platform suite xMOD/Morphée, all the key issues of a full hybrid powertrain can be addressed efficiently and at low cost compared to the experimental powertrain prototypes: consumption minimisation, energy optimisation, thermal exhaust management, NOx/soots trade off, NO/NO2 ratios... Having a good balance between versatility and compliancy of the model oriented test platforms such as presented in this paper is the best way to take the maximum benefit of the model developed at each stage of the powertrain development.

Kurzfassung Dieser Artikel beschreibt, wie mit modernen, leistungsstarken Simulationssystemen eine vollständige Testumgebung für einen Motorantriebstrang aufgebaut werden kann. Mit Hilfe von Prüfstandsergebnissen eines Euro 6 Hybrid Antriebstrangs ausgestattet mit einem turbogeladenen Euro 5 Diesel-Motor, zeigen die Autoren wie ein entsprechender Simulator, bestehend aus einem phänomenologisches Modell für einen Dieselmotor und einem DOC&DPF Abgasnachbehandlungssystem eingesetzt werden kann, um Schadstoffemissionen unter verschiedenen Einsatzbedingungen vorherzusagen. Die vorgestellte Plattform ermöglicht:

-

Parallele Vernetzung (multi-core) unterschiedlicher Simulationsmodelle (multimodel) in Echtzeit Möglichkeit der direkten Einbindung der Echtzeit-Rechenmodelle in die Hardware- in-the-Loop Mögliche Einbindung in die Regelung eines Motorprüfstandes, um verschiedenen Testzyklen (NEDC oder FTP) zu fahren.

Die hier vorgestellte Simulationsplattform xMOD/Morphée ermöglicht eine kostengünstige und effiziente Optimierung eines kompletten Hybrid-Antriebstrang, einschließlich Abgasnachbehandlung und kann zur Verbrauchsminimierung, zum thermischen Abgasmanagement oder Schadstoffreduzierung wie Russ oder NOxEmissionen genutzt werden. Der gewählte Kompromiss zwischen Modellflexibilität und –komplexität der vorgestellten Simulationsplattform hat sich sehr gut bewährt für die unterschiedlichen Auslegungsphasen in der Entwicklung eines Motor-Antriebstranges.

1.

Introduction

With the continuous improvement of models accuracy and computation performance, the strong cost killing process that is introducing models as a virtual test support is extended to the whole powertrain development cycle. Indeed, it is no longer possible to imagine the modern test configurations without embedded models to be able to address the complexity of the current powertrain technologies. Each day, models demonstrate a little bit more their interest in all the kind of test platforms, from the 100% virtual prototyping to the virtual powertrain coupled with an ECU or a hardware component. In a first part, this paper introduces a general description of the simulation-based engineering penetration in the powertrain development cycle and highlights how the simulation environments have an important role to play to support it. In a second part, this simulation tool issue is detailed through the example of the xMOD/Morphée suite concept which is developed in partnership by IFP Energies nouvelles and D2T. Finally, the implementation of these tool suite in the different stages of the development of a Euro 6 hybrid powertrain is presented with result examples.

2.

A look on the powertrain simulation penetration in the industrial practices

Among the developed tracks to face the increasing complexity of the powertrains while reducing costs and delays, the simulation-based engineering represents one of the most attractive options. Most of the powertrain industry actors have already initiated the introduction of this approach in their activities, with more or less determination and success. Foremost, the adoption of the simulation-based engineering in the powertrain development practices is a cultural change challenge. That is why one crucial issue is to take into account the current situation and the simulation tools themselves to have a big role to play in this process.

Figure 1: Powertrain development virtualisation This part presents some analysis elements to understand why the simulation is still not so much used in the powertrain development and to identify progress axes which may improve the situation, mainly around the simulation tools involved.

2.1. The context of simulation-based engineering growth in the powertrain development As a matter of fact, all the current trends indicate that the simulation-based engineering will be a key approach for the future of powertrain industrial development. The powertrains are more and more complex systems with strong coupling between heterogeneous components: mechanical, electrical, fluids, thermal, chemical... By offering a low cost and flexible way to study the whole system complexity at the early stages of the development process, the multi-domain system simulation is a very efficient approach to support technological decisions which can no longer be made only based on engineer experience or standard calculations. For example, the hybrid vehicles are expected to be able to deal with an increasing number of constraints and performances embedding advanced technologies. These new systems impose to take into account at the same time component sizing, powertrain architecture and energy management to be able to proceed efficiently in the vehicle development. This requires new powertrain development process and the multi-domain simulation is the cornerstone for this issue [1][2]. Another strong trend that set the virtual powertrain as a relevant support development is the cost and delay-killing politicies. To put high technology in the processes to shorten and reduce the cost of the development cycle is a good way to limit the low cost workforce location appeal while increasing the company competitiveness. If simulation can obviously not substitute the experimental supports, the virtual approach when used for relevant issues and with a rational complementarity with experiments can significantly help achieving these goals. On the first side, part of the experimental tests can be achieved by modelling. Furthermore, being able to set up virtual prototypes at low cost, more options can be investigated while less hardware prototypes would need to be built. On the second side, to take benefit of testing virtually the whole powertrain from the very beginning of the project and all along the development is a good way to reduce the time-expensive development iterations and to reach the right design at first time [3][4].

Figure 2: Simulation-based engineering as a first time right design process

2.2. A strong potential but still a lot of obstacles Even considering all the advantages of the simulation-based approach presented in the last section, it is easy to observe that the situation in the companies is not so simple. Most of the time, the top management can be easily convinced from a theoretical slide-based demonstration but the fact is that the simulation-based engineering is synonym of a strong cultural change for the operators. It represents large modifications in the way they use to work and need time and investment to be accepted by them. There are a lot of typical arguments to slow the simulation use. The easiest way is to only consider a complete substitution between experiment and simulation. The current powertrain models are not able to face all the issues expected from the experimental approach. For example, to predict by modelling the in-cylinder HC emission when the engine is under cold conditions is very difficult when one just have to use a standard gas analyzer to measure it. First of all, we have to be aware that the models accuracy is continuously increasing and this will continue thanks to the research effort. A lot of simulation results are now of the same accuracy as the measurement errors. But in this example, the most important thing to point out is that simulation is not assumed to completely substitute the experimental support with a plug-and-play process without adaptation at all. As examples, the simulation is relevant to be used: -

-

before the experiments to optimise the number of tests and hardwares to be tested at the bench, during the experiments to complete the campaign (non-measurable values, operating conditions difficult to be achieved experimentally) and to perform more realistic tests at the bench thanks to test bed-embedded models, after the experiments as an efficient project capitalisation and reuse bases.

Since all is done today to address the powertrain development with an experimentalbased process, it is very easy to point out the drawbacks of any simulation attempt. Some cases that allow to highlight the added value of simulation for a specific local application can be found, but the simulation investment is really able to demonstrate its full potential when applied in the whole development process, from the concept evaluation to the final validation. To be able to reach this whole cycle approach, the simulation has to offer an easy-touse and job-oriented environment for each one of the target engineers. This point is another important aspect that contributes to limit the simulation penetration. The current model development software environments are most of the time expert modelling environments. Once the simulator has been set up and is ready to be intensively exploited by as many people as possible, this expert user interface represents a strong limitation for the user who is not a modelling software expert. Today, most of the time, the one who uses a model is the one who developed it. To succeed in the simulation-based engineering deployment, this limitation has to be significantly overcame. The human dimension is not the only one that limits the simulation penetration in the powertrain development process. There are also strong technical difficulties to address. One of the main technical problems is the fact that the generated models for each subsystem of the powertrain have been developed with specific softwares. It seems obvious that the best tool to model a thermal engine is not the same as the one to model a battery or an air conditioning system. When coupling these models, it is necessary to take benefit of the virtual prototyping approach, it is most of the time not possible according to their heterogeneous formats. It is the reason why numerous cosimulation platforms have been developed. These platforms allow to manage the execution of each of the model development software in parallel and exchange data at determined timings. This kind of approach can be adapted for a small number of softwares and when the license cost and the execution delays are not crucial, but becomes prohibitive (in terms of license cost and time consumption) when it has to be applied to a big system.

2.3. The simulation environment as a simulation-based engineering support Having the strong potential of the simulation-based engineering approach introduced in the section 2.1 and the current limitations it faces to become a reality as described in the section 2.2, it seems that part of these limitations can be overcome thanks to specific simulation tools that have been designed taking into account this context. The current main issues that seem relevant that this simulation environment would address are: 1. The heterogeneous models of the powertrain simulation world The software corporate choice is a myth from the organisation acceptation point of view as well as from the technical efficiency point of view. The powertrain development mixes a lot of physics areas and technical issues and it is an illusion to think that a single modelling software would be able to address properly all the expecta-

tions. The best way to deal with simulation in development of such complex systems is therefore: -

to allow each engineer to use the best tool for his own application, to be able to integrate his models in the global system model for his own validation, to be able to easily transfer his models to his colleagues for them to take benefit of his development.

It is the reason why the simulation environment has to be able to integrate the models built in the softwares used by the engineers, through a specific importation process or thanks to a standard model format. 2. The execution time performances to match the industrial goals To be viable, the simulation approach has to offer a low cost way to perform tests but also to perform more tests in shorter delays with more and more complex models. Various computing features can help to fulfil these expectations. At first, most of the time, a system simulator is made of sub-systems which may not require the same numerical solver and the same time step. To be able to execute each subsystem model with a specific solver and a specific time step can speed up the execution time of the complete simulator. Another property of the system simulators is their systemic architecture which allows to easily compute each subsystem model separately. Combined with multi-core or multi-processor computers and with optimised load scheduling, to split the system simulator execution is a strong way to match faster CPU performance, notably when real-time computation is aimed. 3. The friendly environment to facilitate the simulation deployment It is very usual that the models built for a project are mainly used by the ones who developed them. To secure the return on the model development investment, at least ten people have to use each model. One of the big limitation to achieve this model developer/user ratio is the required skill to use the model. In fact, the modelling environments are expert environments and the potential users have to know the software to be able to take benefit of the available models. To facilitate the simulation access to non-experts, it is useful to give tools to customise the user interface. The goal is to be able to put the simulator in a specific exploitation environment for each job specificity. By this way it can reduce as much as possible the required adaptation of the day-to-day practice the user has to achieve to take benefit of the available simulators. 4. The combination with the current development supports and methodologies The last important issue to address to develop the simulation-based engineering is to take into account the current practices in the powertrain development process. It means that the simulation approach will be easily accepted if it does not require a brutal change in the work environment. That is why the simulation tool has to embed the necessary flexibility and compliancy to be put in a current process with as small perturbations as possible. From the development point of view, the compliancy with the standard environments is a very strong asset for a simulation platform. For the powertrain simulation, one of the main issue is to be able to connect or to export simulation elements with the standard Hardware-in-the-Loop (HiL) platforms and with

the typical testbed equipments to have a versatile and compliant coupling between virtual and real systems or to be able to easily achieve mixed simulation/experimental works. Finally, based on high technologies and versatility, the simulation platforms have to be the bridge to support the powertrain development from the today practices to the tomorrow techniques.

Figure 3: The xMOD/Morphée suite

3.

The xMOD/Morphée suite: a concept to enhance simulation use in powertrain development

The previous part has underlined the role of the simulation environment to develop the simulation-based engineering approach, pointing out the relevant specificities to address. This part presents a tool suite concept developed by IFP Energies nouvelles and D2T, which takes into account most of these benefits (Figure 3).

3.1 xMOD: from Software-in-the-Loop to ECU Hardware-in-the-Loop xMOD aims to use the concept of "gray to black box" as a main basis for model exchange [5]. The exchanged models are instrumented using virtual instruments dashboards, allowing to abstract their modelling language. xMOD does not intend to replace the original modelling and simulation tools, but aims at promoting their coexistence. The xMOD concept also relies on separating the phases of model building and model exploitation of the model life cycle. It focuses on the latter phase. The main idea of xMOD is to combine within the same platform: -

a heterogeneous model integration environment a know-how protected, standalone and optimal execution platform a virtual experimentation laboratory

The heterogeneous model integration environment xMOD provides a heterogeneous model integration environment for models built by different persons using different languages and tools and working within different entities. Currently, xMOD allows the integration of a lot of system models such as Matlab/Simulink, AMESim, Dymola, SimulationX, GT Power, ... among others. It can also import the homemade C/C++ models. By this way, xMOD allows to exploit the forces of each modelling and simulation tool (modelling language or environment suitability with an engineering field, solvers efficiency, language expressivity, modeller proficiency, existence of adapted libraries...). Consequently, thanks to xMOD, modelling language choice has less impact on users, using a heterogeneous model integration environment allowing to congregate the benefits of each tool for each specific application, without having to leave its own tools. The xMOD target for importing Simulink models relies on Matlab Real-Time Workshop code generator. For the others softwares, the RTW code generator can also be used but direct export from the original softwares to xMOD are also available for some of them. Furthermore, xMOD is also compatible with the functional mock-up interface (FMI) standard that was developed within the MODELISAR ITEA2 project [6]. This compatibility allows xMOD to be able to import models from the modelling and simulation environments that are able to export models as functional mock-up units (FMUs: software files that are compatible to the FMI interface specification [7]).

Figure 4: xMOD import features The know-how protected, stand-alone and optimal execution platform xMOD executes models in binary form. When a model is converted to the xMOD imported format, two files are generated: a model interface description xml file (whose content is customisable) and a dynamic link library, which is compiled directly from C code. This provides a generally acceptable level of model details protection. The xMOD target allows specifying exactly the signals and the parameters that have be included in the model interface and made accessible during model exploitation and co-simulation. In xMOD, the models are then seen as gray boxes, whose "darkness"

is customisable by the model owner. In this representation, models are characterised by inputs, outputs, parameters and signals. xMOD offers an integration, co-simulation and virtual prototyping platform, which is able to run models without any need of the existing modelling and simulation tools. xMOD allows the integration and exploitation of models without requiring the installation, on the same computer or network, of the original modelling tools (Simulink, AMESim, Dymola ...) that have been used to produce these models (in opposition to other collaborative co-simulation environments, which are based on tools coupling and which ensures the co-simulation by running the models on their original simulation environments and exchanging data using DDE or DCOM protocols). Imported models embed all the needed data: no data files are referred to. This significantly eases model exchange and increases simulation performance. xMOD runs binary models compiled from their associated C code, activating the full code optimisation during compilation. For this reason, execution times are several times faster than those obtained while simulating interpreted modelling languages. xMOD associates an execution thread to each integrated model. Mechanisms for guaranteeing the synchronisation between the different threads were implemented. This allows xMOD to execute models embedding different solvers at different steptimes on multi-core and multiprocessor architectures. Co-simulation benefits: in comparison to a global simulation of a complete system using a single solver, a cosimulation where the system is conveniently decomposed, for example to isolate stiff parts, provides a more efficient execution and better simulation times [8]. In fact, isolating stiff components of the system model avoids constraining the whole model with their solving requirements. Non-stiff parts may be integrated with lighter solvers, and with greater step-sizes. The research and methodology that allowed to run the engine, powertrain and vehicle models, described in this paper, in xMOD with a simulation speed faster than real-time, are described in [9]. The virtual experimentation laboratory xMOD allows creating virtual dashboards, containing virtual instruments, which may be linked to the different parameters or signals of the model that have to be displayed (and that the model designer allowed to make accessible). xMOD also allows to create automated virtual test benches using different scripting languages. These features enable validating the system model that was built through the integration of different components, dimensioning components (design parameters optimisation using simulation), pre-calibrating control algorithms parameters or running robustness tests. These virtual experimentation features allow model exploitation by non-experts. For this reason, proficiency in simulation design tools is no longer the "entry ticket" for exploiting simulation. The ECU Hardware-in-the-Loop platform xMOD can currently be used as a virtual instrumentation and automation software for models running under xPC operating system from The Mathworks. A hard-real time version of the xMOD kernel is also available, allowing it to run models in hard realtime under the operation system RTX from IntervalZero. This version is able to use numerical communications with the external world : using Controller Area Network

(CAN) or real-time Ethernet. Both versions allow to run models in real-time in standard PC hardware. For future versions, it is planned to make xMOD compatible with future hardware-inthe-loop standards, especially the ASAM HiL API standard. The main objective of the ASAM HIL API 1.0 is to decouple test automation software from test hardware. This achieves the reuse of test cases within the same automation software on different test hardware systems. This API also provides important building blocks allowing the portability of test cases. This is the objective of the next version (2.0) of the API.

3.2 Morphée: from component test bed to system test bed with embedded models Morphée is a real-time test automation and monitoring software. Using exclusively industrial hardware from the market, Morphée may be used for complete vehicle development, from components and emission testing, up to the final ECU calibration. Morphée allows to perform several types of engine test applications: advanced development work, engine performance, endurance, emissions (transient tests), catalyst ageing, ECU / TCU mapping and development, friction losses, quality control, and production. The same automation system may be used for component tests. The editing of test procedures may be carried out without any hardware on a simple desktop PC. The pre-emptive real-time kernel, based on the RTX real-time operating system, ensures the execution of the critical test bed tasks, controlling devices and I/O through protocol and communication drivers in accordance with the test bed configuration. The version 2 of Morphée has the ability to run simulation models in real-time on the test bed. For Morphée 2, a model is considered as a black box whose inputs and outputs are linked to Morphée channels [10]. In this setting, the model will perform all necessary calculations, and Morphée 2 will perform the required communication with the test bed. The most important advantage of this solution is that it allows distinguishing the two tasks which are: • The model development. • The model integration on the test cell. With this principle, the simulation engineer can focus on model development ignoring the communication process between the model and the several test bed equipments (this communication will be available through Morphée). When the model is integrated into Morphée, the test engineer just has to link the model inputs/outputs with his test quantities. This methodology also allows to execute the same model on different test cells without performing modifications, and to keep the model as confidential for the user in case of model redistribution. During the test execution, each model has its own real time task, this principle will allow to beneficiate of the powerful calculation capacity of multi-processor platforms. During the test execution, models can exchange data between them, and models’ parameters can be tuned by Morphée. For the vehicle simulation on the engine test bed study, we have developed two models; one in Simulink to simulate the vehicle driver and another in AMESim to simulate the vehicle. During the test, the two models are executed in real time and

exchange data through Morphée . The figure 5 illustrates the data exchange between the models, Morphée and the test bed.

Figure 5: Model execution principle in MORPHEE

3.3 The xMOD/Morphée compatibility Dashboards and automation scripts used in xMOD are compatible with test bench automation software Morphée 2. Thanks to this compatibility, the Model-in-the Loop (MiL) validation performed with xMOD allows to prepare and to significantly reduce the time of the subsequent phases, like experimental validation which are more expansive. Using an automated procedure, xMOD simulation descriptions may be converted to Morphée 2 test descriptions. The transition from a fully virtual to a semivirtual/semi-real test is extremely quick. It is sufficient to remove the blocks that represent virtual components in xMOD, to apply the conversion wizard, and finally to connect the ports of these blocks in Morphée to the physical channel in order to complete the conversion process.

4.

Application

This last section presents several results from a Euro 6 hybrid powertrain project which has taken benefit of the simulation-based engineering approach implemented in the xMOD/Morphée suite.

4.1. Model description The powertrain simulator The powertrain model set up for this study is based on two softwares: Mathworks Matlab/Simulink and LMS Imagine.Lab AMESim. LMS Imagine.Lab AMESim is a simulation platform which offers a modelling approach that allows to take into account all the powertrain aspects and simulate multiphysics system thanks to the Bond Graph approach. It also proposes adapted interface and export facilities with the others tools involved in the applications (Matlab/ Simulink, xMOD, Morphée 2, ...). To be able to support all the simulation expectations, LMS Imagine.Lab AMESim provides an efficient variable step numerical solver to capture accurately all the system dynamics and a fixed step solver to be able to significantly reduce the CPU time after all the relevant mechanisms have been identified. In this study, the conventional part of the powertrain has been modelled with this tool: the phenomenological thermal engine, the mechanical transmission, the wheel dynamics and the vehicle external forces. Mathworks Matlab/Simulink is the inescapable tool to develop control algorithm and is also widely used to develop physical models. In this powertrain simulator, main of the hybrid-oriented models have been implemented using Simulink: the batteries and the electric motors. Furthermore, all the control laws and the energy management have been developed in the Mathworks environment.

Figure 6: the powertrain simulator in xMOD (dark grey models are from Simulink, light grey models are from AMESim)

Figure 6 presents an overview of the complete powertrain system in the xMOD environment, once each model has been exported from one of these two softwares. This system diagram is called a MIPS and refers to a global model which is independent of the modelling softwares. The xMOD interface presented in Figure 7 is an example of the dashboards set up for the study, to use the powertrain simulator as a virtual development support. Huge quantities of parameters and variables can be modified during the simulation which offers a flexible way to perform a wide range of tests, from component sizing to control tuning. The same platform can then be used for real time Hardware-in-the-Loop issues and for high dynamics tests at the bench thanks to the xMOD/Morphée tool suite.

Figure 7: xMOD user-interface

Combustion and pollutant emissions models' description This section presents some details about the phenomenological combustion and pollutant emission models embedded in the thermal engine model, which are advanced modelling approaches to be executed in real time applications. A combustion model based on the Barba approach [11],[12] is chosen to compute the combustion heat release rate. In this approach, the combustion process is divided into two parts. In a first step, the fuel is burnt using a pre-mixed model with the hypothesis of flame propagation in the pre-mixed zone. When the pre-mixed zone is burnt, the remaining fuel is oxidised using a mixing controlled combustion model. The different hypothesis and equations of Barba's combustion model are presented in [12] Using a simple example with a pilot and a main injection, different steps can be isolated during the combustion. At the beginning of the pilot injection, the combustion model is initialised and the different variables are computed. During injection, the auto-ignition delay is computed and the injected fuel is introduced in the premixed zone. The auto-ignition delay is defined at the beginning of the injection. In 0D Diesel combustion models, the auto-ignition delay is generally defined with one step chemistry [11],[12]. These simple models can be used for conventional Diesel combustion in

which the evolution of the logarithm of the auto-ignition delay is relatively proportional to the inverse of the temperature. The auto-ignition delay is thus computed using a simple Arrhenius law. The use of a simple auto-ignition delay model was chosen so as to have a Diesel combustion model useful for engine control definition under conventional operating conditions. When the auto-ignition delay is reached at the end of the phase, the pre-mixed combustion starts. In a first step, the ascendant part of the combustion heat release rate (burning mode 1) is generated by the flame propagation in a turbulent field. The equations to compute the burning rate are defined in Barba et al. [12], and are similar to the equations presented in [13] and in [14] for the propagation of a flame in a homogeneous mixture. Then, because of the multiplication of the auto-ignition sites, the hypothesis of the propagation of a single flame in the pre-mixed zone is no longer valid and the flame interactions must be modelled (burning mode 2), leading to a reduction of the total flame surface. When the main injection starts, the model creates a new pre-mixed zone for the main injection and computes the auto-ignition delay. The definition of the pre-mixed zone is similar to that of the pilot injection. As for the first injection, the auto-ignition delay of the main injection is computed from the beginning of the injection, leading to the beginning of the combustion. The autoignition model is the same as for the pilot injection, but thanks to a higher pressure and temperature, the auto-ignition delay is smaller and reached before the end of the injection. As the auto-ignition delay is reached before the end of the injection, the fraction of fuel available in the premixed zone is lower than for the pilot injection. The combustion process is the same than for the first injection. In parallel to the pre-mixed combustion mode, the remaining injected fuel starts to burn progressively with a mixing controlled combustion model. In this burning mode, the combustion is piloted by the mixture speed between the injected fuel and the surrounding air. Generally, the mass of fuel burnt in this mode is defined with a simple equation depending on the turbulent kinetic energy. In order to enlarge the range of applications of this combustion model, we introduce the possibility to compute more injections and combustions. In parallel to the development of the combustion model, the simulation of the pollutant emissions formation is proposed. In a first step of development, the formation of CO, NO and soot are modelled. The unburnt hydrocarbons formation will be studied in details in a future work. For now, the unburnt hydrocarbons come from the remaining fuel which is not consumed at the end of the combustion cycle. The CO is formed during the combustion process, in the rich zones. In addition to the formation of CO due to the combustion process, it is also important to take into account kinetic reactions. These kinetic reactions are modelled thanks to Arrhenius laws. In a similar way, the NO formation can be modelled following kinetic reactions, illustrating the Zeldovich model [15]. Modelling the soot formation in a 0D approach is a difficult task due to a lack of information concerning geometrical phenomena that occur in the combustion chamber. A first step is proposed here, introducing a model for the formation and the postoxidation of soot particles, thanks to the approach of Bayer and Foster [16].

4.2 Thermal engine In a first step of the development project, the engine control is adapted to take into account the powertrain goal specificities. In Figure 8, a load transient at a constant engine speed of 2000 RPM is performed in two configurations: - the xMOD HiL with the four-cylinder diesel engine simulator with airpath and EGR loops and the rapid prototyped ECU, - the Morphée 2 test bed with the real thermal engine.

10

50

9

45

8

40

7

35

6

30 NOx [g/h]

IMEP [bar]

In the presented test, the IMEP varies from 2 bar up to 8 bar. The first part of the transient conditions represents small variations of IMEP (steps of 1 bar) whereas the second part of the test consists in simulating strong variations (2 to 8 bar or 5 to 9 bar), representing a kind of "kick-down" behaviour of the driver's pedal. Injected mass and air mass flow rate are regulated by the engine control in order to obtain correct IMEP and air mass flow rate.

5

25

4

20

3

15

2

10

HiL Bench

1 0

HiL Bench

5

20

40

60

80 Time [s]

100

120

140

0

20

40

60

80 Time [s]

100

120

140

Figure 8: Results of IMEP and NOx mass flow rate for 2 different platforms (HiL and test bed) for a load transient at 2000 RPM

The right hand side of Figure 8 presents the results obtained thanks to the model, in comparison with measurements obtained thanks to a high-frequency NOx sensor. As it can be observed, the results are good and close to the experiments. Limits of the model appear at high loads, between 40 / 60 s and 120 / 140 s but the trends are matched. Difficulties have been encountered with measurements of NOx between 80 / 95s and 100 / 115s. In fact, these operating points correspond to very low loads (under 2 bar) and high levels of EGR. Levels of NOx do not match, compared to the other operating points at low loads during the transient test: the simulation predicts very low mass flow rates of NOx (as expected) whereas the measurements indicate levels of NOx comparable to IMEP of 4 bars.

4.3 Conventional powertrain Engine-based conventional powertrain development for Euro 6 In this section, the conventional powertrain aims to match Euro 6 standards focusing on the engine-out emissions. It presents a comparison between two configurations for the complete conventional powertrain: • the xMOD SiL configuration where all the components of the system are modelled thanks to the approach described in section 4.1 and compiled under xMOD environment, • the Morphée 2 test bed configuration with the thermal engine as the hardware part of the system, simulating all the other parts (gearbox, vehicle, control, etc.) under Morphée 2 environment. Figure 9, Figure 10 and Figure 11 present respectively Indicated Mean Effective Pressure (IMEP), NOx and CO2 mass flow rates, comparing the two configurations described above for a New European Driving Cycle (NEDC). 14 Bench HiL IMEP Cyl. 1 Bench HiL IMEP Cyl. 2 Bench HiL IMEP Cyl. 3 Bench HiL IMEP Cyl. 4 SiL IMEP mean value

12

10

IMEP [bar]

8

6

4

2

0

-2

0

200

400

600 Time [s]

14

1000

1200

14 Bench HiL IMEP Cyl. 1 Bench HiL IMEP Cyl. 2 Bench HiL IMEP Cyl. 3 Bench HiL IMEP Cyl. 4 SiL IMEP mean value

12

10

Bench HiL IMEP Cyl. 1 Bench HiL IMEP Cyl. 2 Bench HiL IMEP Cyl. 3 Bench HiL IMEP Cyl. 4 SiL IMEP mean value

12

10

8

8 IMEP [bar]

IMEP [bar]

800

6

6

4

4 2

2 0

0 -2 600

620

640

660

680

700 Time [s]

720

740

760

780

800

-2

800

850

900

950

1000 Time [s]

1050

1100

1150

Figure 9: Results of IMEP for SiL and test bed platforms (at the test bed, all the four cylinder were equipped with a pressure sensor for IMEP estimation) during NEDC cycle – last UDC and EUDC are plotted at the bottom

One can observe that the 100% virtual configuration (SiL) seems to well reproduce the behaviour observed at the high dynamics testbed even if it can be notified that the SiL approach tends to overestimate the NOx mass flow rate during the EUDC part of the driving cycle. However, comparisons between experiments and numerical results are always a difficult task, without taking into account the delay observed experimentally during the convection of the species at the exhaust line, between the exhaust manifold and the gas analyser. This delay is not taken into account here, due to the lack of a deconvolution function for the pre-processing of the experimental signals. 40 Bench HiL SiL 35

30

NOx [g/h]

25

20

15

10

5

0

0

200

400

600 Time [s]

800

1000

1200

Figure 10: Results of NOx mass flow rates for SiL and test bed platforms during NEDC cycle 3.5

x 10

4

Bench HiL SiL

3

2.5

CO2 [g/h]

2

1.5

1

0.5

0

0

200

400

600 Time [s]

800

1000

Figure 11: Results of CO2 mass flow rates for SiL and test bed platforms during NEDC cycle

1200

Table 1 sums up the results obtained with both of these configurations, comparing NOx emissions and fuel consumption / CO2 emissions during the whole NEDC cycle. The emissions are really comparable and very close, with deviations lower than 2.5%, which are acceptable in terms of fuel consumption and very interesting as far as the NOx emissions are concerned. Configuration

Platform

NOx [mg/km]

CO2 [g/km]

Fuel cons. [L/100km]

Test bed

Morphée 2

123.5

133.2

5.1

SiL

xMOD

126.7

133.8

5.2

Table 1: Comparison between test bed (considered as the reference) and SiL platforms in terms of NOx, CO2 and fuel consumption during NEDC cycle

After-treatment-based conventional powertrain development for Euro 6 In this section, the conventional powertrain aims to match Euro 6 standards using a Euro 5 thermal engine and focusing on after-treatment performances. In such a context, sizing a DPF and/or a SCR/NOx-trap are difficult tasks before dealing with the whole vehicle, including the calibrated thermal engine. Thanks to a SiL platform, it is possible to obtain a first estimation of raw emissions such as NOx and Smoke mass flow rates, at the exhaust manifold, during NEDC cycles (see Figure 12 and Figure 13). 80

70

Bench SiL

60

NOx [g/h]

50

40

30

20

10

0

100

200

300

400

500

600 Time [s]

700

800

900

1000

1100

Figure 12: Results of raw NOx mass flow rates for a SiL platform and chassis dynamometer obtained with the conventional powertrain

20 SiL 18

16

Smoke before DPF [g/h]

14

12

10

8

6

4

2

0

100

200

300

400

500

600 Time [s]

700

800

900

1000

1100

Figure 13: Results of raw smoke mass flow rates for a SiL platform obtained with the conventional powertrain Table 2 sums up raw emissions in terms of NOx and Smoke, comparing the results obtained thanks to the SiL platform to the measured ones. Values are very close and prove a good reliability of the approach. Configuration

NOx [mg/km]

Smoke before DPF [mg/km]

Chassis dyno.

173.9

27.2

SiL

175.9

28.8

Table 2: Comparison between experiments and SiL in terms of NOx and smoke emissions before the Diesel Particulate Filter

4.4 Hybrid powertrain SiL as a hybrid powertrain architecture virtual testing environment Thanks to a SiL approach, a comparison between several architectures integrating different levels of hybrid strategies can be easily performed at a very early stage to achieve a first selection of the most interesting architectures, in regard to predetermined criteria such as a NOx/Smoke trade-off or the fuel consumption, for example. Four configurations are compared in Figure 14, Figure 15 and Figure 16: • • • •

conventional: thermal engine without any hybrid technology (as the reference), micro-hybrid: thermal engine with a stop and start technology, hybrid 1 (HEV): thermal engine with a stop and start technology and a full hybrid strategy coupled with an electric motor of 20 kW – Equivalent Consumption Minimisation Strategy (ECMS) N°1, hybrid 2 (PHEV): thermal engine with a stop and start technology and a full hybrid strategy coupled with an electric motor of 20 kW – Equivalent Consumption Minimisation Strategy (ECMS) N°2 (“plug-in – like” beha viour).

Table 3 sums up the results obtained with these four configurations, in terms of NOx & Smoke mass flow rates and fuel consumption. Architecture

NOx [mg/km]

Smoke [mg/km]

Fuel consumption [L/100km]

Conventional

126.7

46.8

5.2

Micro-hybrid

117.2

46.3

4.9

Hybrid (HEV)

53.8

36.6

4.1

Hybrid (PHEV)

42.0

26.2

2.7

Table 3: Comparison between four powertrain architectures in terms of NOx, smoke and fuel consumption – results obtained thanks to SiL platforms 25 SiL conventional SiL micro-hybrid (stop & start) SiL full hybrid (HEV) SiL full hybrid (PHEV) 20

NOx [g/h]

15

10

5

0

0

200

400

600 Time [s]

SiL conventional SiL micro-hybrid (stop & start) SiL full hybrid (HEV) SiL full hybrid (PHEV)

20

1000

1200

SiL conventional SiL micro-hybrid (stop & start) SiL full hybrid (HEV) SiL full hybrid (PHEV)

20

15

NOx [g/h]

NOx [g/h]

15

10

10

5

0 600

800

5

620

640

660

680

700 Time [s]

720

740

760

780

800

0

800

850

900

950

1000 Time [s]

1050

1100

1150

Figure 14: Comparison of NOx mass flow rates between four powertrain architectures using a SiL platform – last UDC and EUDC are plotted at the bottom

Smoke and NOx mass flow rates decrease while increasing the hybrid level, as expected. A drawback of the stop and start configuration observed thanks to the SiL is the higher peaks of NOx after re-starting the thermal engine, compared to the conventional configuration. These peaks come from a lack of Burnt Gases Ratio in the cylinders, due to the EGR loop inertia. 30

SiL conventional SiL micro-hybrid (stop & start) SiL full hybrid (HEV) SiL full hybrid (PHEV)

25

Smoke [g/h]

20

15

10

5

0

0

200

400

600 Time [s]

30

1200

SiL conventional SiL micro-hybrid (stop & start) SiL full hybrid (HEV) SiL full hybrid (PHEV)

25

20 Smoke [g/h]

20

15

15

10

10

5

5

0 600

1000

30 SiL conventional SiL micro-hybrid (stop & start) SiL full hybrid (HEV) SiL full hybrid (PHEV)

25

Smoke [g/h]

800

620

640

660

680

700 Time [s]

720

740

760

780

800

0

800

850

900

950

1000 Time [s]

1050

1100

1150

Figure 15: Comparison of smoke mass flow rates between four powertrain architectures using a SiL platform – last UDC and EUDC are plotted at the bottom

The two hybrid configurations correspond to two different ECMS, the PHEV one enabling to consume more battery than the HEV one. The State Of Charge (SOC) presented in Figure 16 is close to 20% for the PHEV one at the end of the NEDC cycle. This strategy corresponds to the maximum use of the electric motor, simulating a vehicle coming back at home without thermal engine solicitation.

0.55

0.5

0.45

SOC [%]

0.4

0.35

0.3

0.25

0.2

0.15

SiL conventional SiL micro-hybrid (stop & start) SiL full hybrid (HEV) SiL full hybrid (PHEV)

0.1

0.05

0

200

400

600 Time [s]

800

1000

1200

Figure 16: Comparison of States Of Charge between four powertrain architectures using a SiL platform – last UDC and EUDC are plotted at the bottom of the figure

Engine bench with virtual powertrain as an hybrid architecture testing environment It has been possible to benefit from the investigations carried out thanks to SiL platforms to propose a first selection of powertrain architectures. A second step consists in testing variant configurations from the first selection, thanks to the test bed configuration with the thermal engine hardware and the remaining parts of the powertrain architecture using the same models as the SiL but under the Morphée 2 environment. Four configurations are compared in this part and sum up in Table 4: • • • •

conventional: thermal engine without any hybrid technology (as the reference), micro-hybrid: thermal engine with a stop and start technology, hybrid 1 (HEV): thermal engine with a stop and start technology and a full hybrid strategy coupled with an electric motor of 20 kW – Equivalent Consumption Minimisation Strategy (ECMS) N°1, hybrid 2 (HEV): thermal engine with a stop and start technology and a full hybrid strategy coupled with an electric motor of 20 kW – Equivalent Consumption Minimisation Strategy (ECMS) N°2,

ECMS 1 & 2 correspond respectively to two levels of CO2/NOx trade-offs, the first one favouring the fuel consumption and the second one the NOx emissions. The two hybrid configurations include a SOC at 50% at the end of the NEDC cycle and correspond to optimised ECMS based on a method described in details in [17]. These comparisons enable to observe that a penalty of 0.1L/100km of fuel consumption permit a decrease of 17mg/km of NOx. It is an example of key values that can be

determined thanks to this methodology. Dealing with these tools, a control engineer is able to propose an ECMS optimisation NOx/EURO6 – oriented. Architecture

NOx [mg/km]

CO2 [g/km]

Fuel consumption [L/100km]

Conventional

123.5

133.2

5.1

Micro-hybrid

113.3

123.1

4.7

Hybrid 1 (HEV)

92.9

101.5

3.9

Hybrid 2 (HEV)

76.0

103.6

4.0

Table 4: Comparison between four powertrain architectures in terms of NOx, CO2 and fuel consumption – results obtained at the test bed with the hardware thermal engine in a virtual powertrain

5.

Conclusion

Some analysis elements about the penetration of the simulation-based engineering approach in the powertrain development cycles have been presented. Based on this analysis, the simulation-oriented environments xMOD and Morphée developed in partnership by IFP Energies nouvelles and D2T have been introduced. This integrated tool suite has been designed to address the current powertrain development expectations in order to take advantage of the simulation taking into account the current practices and implementing the most advanced technologies. Finally, the development of a Euro 6 hybrid powertrain has been used as an example to illustrate how simulation-based engineering when combined with adapted tools can be a powerful way to reduce cost and delay while optimizing the products.

Acknowledgements The authors want to especially acknowledge Frédéric Lippens, control engineer at D2T, Olivier Grondin, Philippe Moulin and Laurent Thibault, control engineers at IFP Energies nouvelles, Jörg Anderlohr, powertrain simulation engineer at IFP Energies nouvelles, and Florent Perez, software development engineer at D2T, for their significant contribution to this paper.

Literature [1]

A. Del Mastro, A. Chasse, P. Pognant-Gros, G. Corde, F. Perez, F. Gallo and G. Hennequet (2009), "Advanced hybrid vehicle simulation: from "virtual" to "HyHiL" test bench", SAE paper 2009-24-0068.

[2]

D. Denger, A. Hirsch, G. Hochmann and J. Zehetner (2010), "The virtual vehicle at the engine test bed", MTZ worldwide, 9/2010, v71, p.32-37.

[3]

O. Imberdis (2011), "Hybrid electric vehicles – development processes & challenges", 12th SIA – CNAM Conferences Series, March 8th 2011.

[4]

C. Schyr, S. Jakubek and G. Stempfer (2009), "A new method of coupling HiLsimulation and engine testing based on AUTOSAR-compliant control units", SAE paper 2009-01-1521.

[5]

M. Ben Gaid, G. Corde, A. Chasse, B. Léty, R. De La Rubia and M. Ould Abdellahi. Heterogeneous Model Integration and Virtual Experimentation Using xMOD: Application to Hybrid Powertrain Design and Validation. EUROSIM'10, Prague, Czech Republic, September 2010.

[6]

http://www.itea2.org/public/project_leaflets/MODELISAR_profile_oct-08.pdf

[7]

http://functional-mockup-interface.org/

[8]

J. Larsson, P. Krus and J.O. Palmberg, Power Transmission and Motion Control, chapter Methods for Organising Co-simulation Among Several Participants, Professional Engineering Publishing Limited, London and Bury St Edmunds, UK.

[9]

C. Faure, M. Ben Gaid, N. Pernet, M. Fremovici, G. Font and G. Corde. Methods for real-time simulation of Cyber-Physical Systems: application to automotive domain. In Proc. First IEEE Workshop on Design, Modelling and Evaluation of Cyber Physical Systems (CyPhy'11), Istambul, Turkey, July 2011.To appear.

[10] F. Perez, Ph. Moulin and A. Del Mastro. Vehicle simulation on an engine test bed. In Proc. SIA Conference on Diesel Engines – The low CO2 & Emissions Reduction Challenge, Rouen, France, Mai 2008. [11] R. Lebas, G. Mauviot, F. Le Berr and A. Albrecht, A Phenomenological Approach to Model Diesel Engine Combustion and In-Cylinder Pollutant Emissions Adapted to Control Strategy, IFAC Paper, ECOSM'09, 2009. [12] C. Barba, C. Burkhardt, K. Boulouchos and M. Bargende, A phenomenological combustion model for heat release rate prediction in high speed DI Diesel engines with common-rail injection, SAE Technical Paper, 2000-01-2933, 2000. [13] J. B. Heywood, Internal Combustion Engine Fundamentals, McGraw-Hill, New York, 1988. [14] O. Colin and A. Benkenida, The 3-zones extended coherent flame model (ECFM3Z) for computing premixed/diffusion combustion, Oil & Gas Science and Technology – Rev. IFP Energies nouvelles, Vol. .59, No. 6, pp. 593- 609, 2004. [15] J. Zeldovich, The Oxidation of Nitrogen in Combustion and Explosions. Acta Physicochimica U.R.S.S. 21 (4), pp. 577 - 628,1946. [16] J. Bayer and D. E. Foster, Zero-Dimensional Soot Modelling. SAE Technical Paper, 2003-01-1070, 2003. [17] O. Grondin, L. Thibault, P. Moulin, A. Chasse and A. Sciarretta, Energy Management Strategy for Diesel Hybrid Electric Vehicle, 7th IEEE Vehicle Power and Propulsion Conference, 2011.