Investigation of Black Box Modeling Approaches for Representation of Transient Gearshift Processes in Automotive Powertrains with Automatic Transmission

In this investigation two different nonlinear dynamic black box modelling approaches are compared. The purpose of the models is to reproduce the transient gearshift process. The models are used to compute the torque at the sideshafts, which is highly correlated to the gearshift comfort. The first model is a Gaussian process (GP) model. The GP is a probabilistic, non-parametric approach, which is additionally capable to compute the confidence interval of the simulated output signal. The second black box model uses the artificial neural net (ANN) approach. In addition to training algorithms the resulting model configurations for both black box approaches are shown in this investigation. Furthermore the empirical error of both modeling approaches is compared to the predictive variance of the GP model and to the intrinsic uncertainty of the gearshift process.

This research demonstrates that black box models are capable of representing the torque at the sideshafts of a transient gearshift process. The results show that both investigated black box approaches yield a sufficient accuracy. Nonetheless the models have their own characteristic advantages, so that the appropriate model approach should be chosen according to the individual intention.