Title:Drivetrain simulation using variational autoencoders

Abstract:This work proposes variational autoencoders (VAEs) to predict a vehicle's jerk from a given torque demand, addressing the limitations of sparse real-world datasets. Specifically, we implement unconditional and conditional VAEs to generate jerk signals that integrate features from different drivetrain scenarios. The VAEs are trained on experimental data collected from two variants of a fully electric SUV, which differ in maximum torque delivery and drivetrain configuration. New meaningful jerk signals are generated within an engineering context through the interpretation of the VAE's latent space. A performance comparison with baseline physics-based and hybrid models confirms the effectiveness of the VAEs. We show that VAEs bypass the need for exhaustive manual system parametrization while maintaining physical plausibility by conditioning data generation on specific inputs.