Structured model identification algorithm based on constrained optimisation

This article presents an identification algorithm dedicated to the estimation of the coefficients of physically-structured models. This algorithm is based on a constrained gradient-based optimisation that minimises a non-convex cost function to identify a structured model, starting from an initial black-box model. The suggested method is able to handle linear relations between the model coefficients so as to identify a model with the desired physical structure. Since the minimised function is non-convex, the initial model must be in the neighbourhood of its global minimum. The initialisation issue is discussed and a total-least-squares-based solution is suggested. These new developments are evaluated on a simulation benchmark, illustrating the identification of a mechanical system.