Conceptual and reduced vehicle models performance enhancement through parameter estimation and neural-networks coupling

摘要

A detailed 3D vehicle dynamics benchmark case has been defined, which has been modeled in LMS Virtual.Lab Motion, The simplified models are created in the 1D multiphysics environment of LMS Imagine. Lab Amesim, in which several mathematical vehicle representations have been adopted, tested and functionally correlated with the reference case. Primary importance has been given to the reliability of results together with numerical efficiency of the finalized models. For this purpose, the reduced models have been enhanced by implementing a Neural Network model in parallel, which can be trained to extend the model's capability to reproduce the complex dynamics of the real system, while remaining as simple as possible for online computation. This new methodology is validated on numerical examples, including an industrial-level vehicle dynamics application case.