Improved Identification of Mechanical Multibody Systems From Discrete-Time Data. Differential Identification of Linear Mechanical Systems

The parameter estimation problem of linear mechanical systems evaluating discrete-time data is considered. In principle it can be solved by the algebraic back-transformation of an estimated discrete time model. This approach means that a-priori-knowledge concerning some physical parameters or zero positions in the matrices of the mechanical system cannot be used. An algorithm is presented which offers a possibility to use this information. It requires initially an approximate discrete-time model-parametrization, generated e.g., with known identification schemes. The new method combines data transformation based on linearized algebraic links between continuous and discrete-time models with a side-constrained instrumental variable estimator. Simulation results from a centrifuge model, illustrating the improved accuracy of the estimates if continuous built-in a-priori information are available.