Identification of linear repetitive processes using subspace algorithms

Abstract

A new approach to the identification of linear repetitive processes based on subspace algorithms is presented. The order of a linear repetitive process, the unknown process matrices, and the noise covariance matrices are determined based on sequences of the actual pass input and the previous pass output, and the actual pass output. The identification procedure can be restarted consecutively starting from the first pass data and boundary conditions. Therefore, the proposed approach can be very useful not only for time invariant linear repetitive processes but also for processes with dynamics that changes rapidly from pass to pass. Simulation example is provided to demonstrate the asymptotic convergence of parameter estimates and the effectiveness of the proposed approach.