Model-based identification of a vehicle suspension using parameter estimation and neural networks

Abstract A two-step scheme for identification of a vehicle suspension is presented which combines parameter estimation and neural networks for approximation. At first, the parameters of the discrete time transfer function are estimated using a RLS-algonthm. These parameters are nonlinear functions of the physical coefficients, but a direct calculation of these is often not possible or leads to large errors due to the nonlinear amplification of noise. Therefore, to approximate the coefficients, a nonlinear mapping using a RBF network is performed. For training of the network and to test generalization abilities, the coefficients of a vehicle suspension were varied. The study shows that an approximation of the physical coefficients by application of the presented scheme is possible. The method was tested by simulated data and measurements from a test rig at the Technical University of Darmstadt.