Abstract
KEYWORDS – ECU Calibration, Model-Based, DoE, Gaussian Process, RDE
ABSTRACT –
Due to the high complexity of modern engines and powertrain systems, the calibration of the ECU parameter has a strong impact on project targets like fuel consumption, emissions and drivability as well as on development cost and duration. Simulation methods representing the system behavior by a model can support the calibration process considerably. But standard physical models are often not capable to describe all effects with sufficient accuracy and / or the effort to set up a detailed model is very high. Additionally, the calculation time of physical models is normally very high so the simulation models are not real-time capable. Well suited for ECU calibration are data driven models, combined with DoE (Design of Experiment) methodology. Here, the system to be calibrated is identified with few specific test bench or vehicle measurements. From these measurements, a mathematical regression model is built. This paper describes newly developed machine learning methods based on Gaussian processes. Contrary to standard polynomial or neural-net regression models, Gaussian processes are able to model even strongly nonlinear systems with high accuracy. In addition, the modeling is easy to apply since the model parameters are determined automatically by probabilistic principles. This model now replaces the real engine or vehicle and can be combined with optimization methods to identify the best ECU parameter with respect to the project tar-gets. The short response times of Gaussian process models allows their use in real-time environments, e.g. in Hardware-in-the-Loop (HiL) test systems or even in serial ECUs to replace complex physical models. The paper shows on different examples how the application of this new model based supports the ECU calibration process.