Multiple Model Approach for Inferential Instruments Design

Abstract

In this paper, an application of a multiple model approach for the design of inferential instruments is reported. The multiple model of interest presents a decoupled structure in the sense that the sub-models do not share the same state variable. A two-stage identification procedure is developed for the model identification and a soft sensor application is later conducted with the decoupled multiple model structure. The soft sensor aims at predicting a quality variable for an industrial separation unit. The decoupled multiple model structure allows obtaining a dynamical model for the soft sensor despite the presence of practical constraints related to multi-rate sampling problem. Real-time implementation results are presented.