Predictive Model of Acceleration Response during Automotive Tip-In Operation

In recent years, the complexity of automotive control systems has been increasing, leading to a wider adoption of Model-Based Development (MBD). One crucial aspect of vehicle dynamic performance is drivability, where the evaluation of longitudinal acceleration vibrations in response to driver inputs is essential. To shorten development time, predictive models based on time-series data have been utilized. However, it is difficult for conventional prediction methods to accurately capture the steep transient changes in longitudinal acceleration response that occur during tip-in input. In particular, in the evaluation of prediction waveforms using MSE, the weight of steep changes is relatively small in order to minimize the average point-to-point error. In this study, we develop a gradient information loss function that captures steep changes in the front/rear acceleration response of a car and a loss function that evaluates the similarity of the entire waveform, and construct a prediction model that integrates the two loss functions, which are local and global, respectively, to verify their usefulness.