Vehicle Suspension Identification Via Algorithmic Computation of State and Design Sensitivities

Abstract

It is common in mechanical simulation to not know the value of key system parameters. When the simulation is very sensitive to those design parameters and practical or budget limitations prevent the user from measuring the real values, parameter identification methods become essential. Kalman filter methods and optimization methods are the most widespread approaches for the identification of unknown parameters in multibody systems. A novel gradientbased optimization method, based on sensitivity analyses for the computation of machineprecision gradients, is presented in this paper. The direct differentiation approach, together with the algorithmic differentiation of derivative terms, is employed to compute state and design sensitivities. This results in an automated, generalpurpose and robust method for the identification of parameters. The method is applied to the identification of a reallife vehicle suspension system (namely of five stiffness coefficients) where both smooth and noisy reference responses are considered. The identified values are very close to the reference ones, and everything is carried out with limited user intervention and no manual computation of derivatives.