Virtual RDE: Ensuring RDE conformity of hybridized powertrains in the early stage of the development process

Abstract

With the introduction of "Real Driving Emissions" (RDE) powertrain simulation has become indispensable to identify critical operating conditions for the exhaust aftertreatment system at an early stage in the development process and to analyze possible corrective actions. Therefore a rudimentary virtual RDE calibration is also necessary in the early concept phase. For doing so, it makes a lot of sense to use 1D flow simulation to model effects such as boost pressure built-up, high/low pressure EGR travel times or thermal inertia. There are two challenges regarding 1D simulation of RDE:  Combustion system development is usually done at single cylinder engines. It is necessary to integrate the results of the single cylinder engine in an effective way into a virtual full engine for RDE investigations and first virtual calibration  Moreover, it is typically necessary to investigate a vast number of RDE driving patterns, taking up much more computational time than it was the case previously with a single type-approval driving cycle. Consequently, "Fast Running Models" (FRM) that reduce the computational effort needed for flow simulation have been used to counteract this increase. However, it is inevitable that such an approach reduces the accuracy of the boundary conditions at intake valve closing (IVC), e.g. EGR rate, temperature and pressure, and crucially, the predictive power of quasi-dimensional burn rate, emission or knock models is very sensitive to these parameters. Such an approach thus yields results of questionable reliability, with the risk of overlooking critical conditions that have to be addressed later on at much higher costs. Existing data-based approaches avoid these challenges, but fail to depict relevant physical processes in the flow path predictively (boost pressure built-up etc.), making it hardly possible to assess technologies such as variable valve trains. Therefore Robert Bosch GmbH and FKFS developed a new simulation approach. The new tool presented in this paper thus combines a physics-based model for the gas exchange (intake and exhaust system) with a data-based model for the high pressure-part, using mean effective pressure and emission values derived from the test bench or detailed 1D flow simulation. This solves the aforementioned dilemma and provides at the same time an easy way to use test bench data for RDE simulations and powertrain analysis. The implementation of such an approach will be presented in this paper, along with some exemplary results showing how the new tool can be used to generate accurate and reliable results for RDE investigations at minimal computational cost. VDI-Berichte Nr. 2373, 2020 344 https://doi.org/10.51202/9783181023730-I-343 Generiert durch IP '54.244.78.42', am 12.11.2021, 10:42:26. Das Erstellen und Weitergeben von Kopien dieses PDFs ist nicht zulässig. RDE Challenge and Existing SimulationTools From 2017 on new passenger cars in the EU have to fulfil exhaust gas emission limits also under RDE conditions. While the emission limits will tighten from current Euro 6d-TEMP to Euro 6d in 2020 and Euro 7 beyond the RDE regulation includes many drive requests that go beyond the current pre-defined driving cycles. E.g. on the one hand vehicle speeds up to 160 km/h and severe accelerations on inclining roads can cause high raw emissions that have to be converted reliably by exhaust gas aftertreatment (EAT). On the other hand low ambient temperatures (down to -7°C), long single stop durations (up to 5 minutes) and downhill drives can lead to an underrun of the lower operational temperature limits of the EAT components. In the near future the latter issue might be even strengthened by two striven efficiencyenhancing measures – hybridization with longer ICE standstill periods and reduction of fuel demanding cold start heating strategies. A series of the RDE requests as well as the emission calculation itself are related to extensive drive sections or even the full RDE drive. As a consequence the substitution of RDE drives by a small number of representative powertrain operation sequences is only possible to a very limited extent. The consideration of RDE performance within powertrain and EAT concept evaluation causes thereby a high testing effort which might even be complicated by missing hardware during early concept phase. 0D/1D powertrain simulation including the prediction of the thermal behaviour of ICE and EAT as well as EAT conversion rates could help to virtualize RDE tests in order to save time and costs of powertrain development. Additionally simulation can be used for synthetic RDE driving profile generation to create “worst-case” drives, either by combining a variety of cut up measured real driving sequences or by utilizing stochastic driving parameter distributions derived from them. Fundamental condition for harnessing the potentials of virtual RDE tests is a simulation tool that constitutes an optimal compromise between predictability, flexibility and simulation time. Assessing existing simulation approaches shows that all of them have severe drawbacks regarding this demand profile (see Fig. 1): 1. Detailed 1D flow simulation model coupled with quasi-dimensional models Physical modeling has proven itself to be a valuable tool in the development of engine technologies: especially 1D-CFD allows the modeling of the whole combustion engine with great flexibility and moderate effort. Air system dynamics, EGR-mixing and the gas exchange can be simulated accurately. Combustion models are available in a broad variety from measured burn rates over simple Vibe approximations up to phenomenological approaches that allow the prediction of the rate of heat release ([1]-[3]). The application ranges from very earVDI-Berichte Nr. 2373, 2020 345 https://doi.org/10.51202/9783181023730-I-343 Generiert durch IP '54.244.78.42', am 12.11.2021, 10:42:26. Das Erstellen und Weitergeben von Kopien dieses PDFs ist nicht zulässig. ly stages in the development where the engine hardware is not necessarily defined or available until the support of function development and software calibration in the final development stages. Virtual hardware components like turbochargers can be matched to the respective requirements or EGR control strategies can be evaluated. All in all it represents a very powerful tool, albeit with a crucial drawback with regard to RDE boundary conditions: computational times (real time factor 50...200, [4]), although rather low compared to 3D-CFD simulations, are still quite high – too high for the high amount of different operating conditions that has to be tested for RDE development. To a limited degree, this can be mitigated by using smart features like master/slave modes for the cylinder objects in full engine models, allowing to reduce the computational effort for the high pressure part distinctly (e. g. almost 75% in a four-cylinder engine). However, as the main part of the computational time is consumed by the flow simulation, the overall reduction is insufficient for RDE demands. 2. Fast-running 1D flow simulation model coupled with quasi-dimensional models This approach addresses exactly the already mentioned, high computational effort for 1D CFD flow simulation. The basic idea is to lump various flow volumes together, reducing thus the number of required calculations per time step while enabling a larger time step size at the same time. A considerable reduction in simulation time can be reached in this way, coming close to real time capability depending on the level of simplification (a factor of two compared to real time can be considered as a typical value), which is definitively enough to qualify for the "fast" tag. However, this approach inevitably changes the model's ability to predict pressure waves in the flow part – actually one of the most important benefits of 1D simulation compared to a pure 0D approach – leading to significant changes in the boundary conditions at IVC for the high pressure part. By nature quasidimensional models are very sensitive to these starting conditions (not unlike the real engine), so they should only be used with flow models that can deliver accurate boundary conditions for the combustion. 3. Data-based models/mean value models Data-based approaches are the tools of choice when it is required to quantify characterize existing systems accurately. Here, former map-based interpolations are increasingly replaced by statistical models that are able to describe the desired result value in dependence of more than just one to three input parameters, which are typical for maps. This can either be necessary to describe results depending on the degrees of freedom of operation that modern engines provide or to represent the deviations from stationary operation an engine faces while operated under highly transient conditions. Besides the proven fulfillment of acVDI-Berichte Nr. 2373, 2020 346 https://doi.org/10.51202/9783181023730-I-343 Generiert durch IP '54.244.78.42', am 12.11.2021, 10:42:26. Das Erstellen und Weitergeben von Kopien dieses PDFs ist nicht zulässig. curacy demands, trained data models can be evaluated with nearly no computational effort. Here the limitation is the extrapolation capability – data based models can only provide trustful information where training data was available. In particular this means that changes in the intake or exhaust system compared to the original configuration cannot be taken into account in the simulation model, making it highly inflexible and unsuitable for tasks like function development and calibration. Fig. 1: Positioning of different simulation set-ups along the three basic requirements for RDE simulations (DET: detailed 1D flow simulation model coupled with quasidimensional models; FRF: Fast-running 1D flow simulation model coupled with quasidimensional models, DAT: data-based models/mean value models) Basic Idea for New Tool To get a fast, accurate and flexible simulati