ROBUST MODEL-BASED SOFT SENSOR: DESIGN AND APPLICATION

Abstract

A model-based robust soft sensor is proposed here. The soft sensor is based on a Kalman filter (KF), which is designed to estimate the inaccessible variables using the output measurements in the face of the measurement noise, disturbances, and model perturbations. The performance of the soft sensor critically depends upon the reliability and accuracy of the identified system model it is based on. To overcome the degradation due to model mismatch, a reliable offline identification scheme, based on the powerful concept of emulator which significantly improved the accuracy of the proposed scheme, is proposed here. It involves performing a number of experiments using emulators, which are transfer function blocks connected to the system input or the output, and which are used to introduce model perturbations to mimic likely operating scenarios. It is shown that the KF residual is a function of the product of the model perturbation and the control input sensitivity function. The proposed new soft sensor is successfully evaluated on a simulated and laboratory-scale physical velocity control system.