Parameter identification of multibody vehicle models using neural networks

In this study, a methodology for the identification of parameters of multibody vehicle models using neural networks is proposed. A neural network is trained based on the first parameter identification and then acts as a metamodel for all further parameter identifications. This approach is applied to two industrial examples in this study. The methodology is first used to identify the exact loading situation of a vehicle, taking into account the pitch characteristics and the normal force distribution of the wheels. Furthermore, the methodology is also used to identify the preload of suspension bushings for given kinematics and compliance test rig properties. Using the neural network as a metamodel for the parameter identification process, for both examples, the computational time can be significantly reduced from 4.5-6 hours to 1-3 minutes and thereby this methodology contributes to a more efficient virtual development process.