Abstract A systematic and unified approach to monitor performance and to predict fault is proposed based on a robust Linear Predictive Coding Algorithm (LPCA) implemented as part of an expert system. An Auto-Regressive and Moving Average (ARMA) model of the measured output is estimated in real-time and the model estimate is used to monitor performance and predict faults. The expert system is comprised of a knowledge-base which is represented in the form of frames and rules, and the rules are fired in the order of decreasing importance and increasing computation. The proposed scheme is evaluated both on simulated and physical control systems.