A method is described for choosing a model structure which can describe the dynamics of a manipulator. The model structure is based on a black-box approach, so the model does not attempt to determine physical parameters such as moments of inertia or link masses. The model structure is used to estimate parameters which best fit the selected model structure to the observed input/output data sequence. The manipulator dynamics are written as a Fourier series expansion. The Fourier coefficients are identified by a least-squares algorithm. A parameter selection criterion which will determine the model structure one should choose is given. The tracking accuracy of controllers based on different model structures is compared.

Proceedings of the 1989 American Control Conference
Department of Systems and Computer Engineering

Schwartz, H.M. (1989). Model structure selection for manipulator trajectory tracking. In Proceedings of the 1989 American Control Conference (pp. 2366–2371).