On-line Parameter Identification from Noisy Measurement Data and Its Application : Construction of an On-line Soft Sensor

Abstract

An on-line identification problem is investigated, as one of the inverse problems, for estimating unknown model parameters which should be determined to establish a valid mathematical model. The system considered in this paper is the linear dynamical system which is subjected to random disturbance. The on-line identification is realized by splitting the procedure into state estimation and parameter identification. The former is curried out by the Kalman filter, and the latter is accomplished by modifying the conventional Gauss-Newton-based algorithm. The basic idea of the identification is to introduce the principle of contraction mappings in order to guarantee the stability of the algorithm. The resulting algorithm consists of the two procedures which are processed on-line alternately. In order to show the efficacy of the proposed algorithm, simulation studies and experiments are shown by applying the theory to an inverse problem of determining physical parameters of the cantilevered beam model.