Fault Detection in a Batch Process Using a Bayesian Model

Abstract

Application of Bayesian dynamic modeling to fault detection is developed for a nonstationary batch process. In the modeling, the observed time series are expressed in several specific components such as local polynomial trend, observation noise and globally stationary autoregressive component.
To illustrate the method, detection of a fault in an operation of a stirred vessel with a heater is presented. From the sequential probability ratio test of the model estimation error, the fault can be detected successfully with high sensitivity.