Closed-loop subspace identification and fault diagnosis with optimal structured residuals
The development of system identification and fault diagnosis theory is of great practical significance. Systems are concerned with a broad spectrum of human-made machinery, including industrial production facilities (power plants, chemical plants, oil refinery, semiconductor fabrication plants, steel mills, paper mills, etc.), transportation vehicles (ships, airplanes, automobiles) and household appliances (heating/air conditioning equipment, refrigerators, washing machines, etc.). This dissertation is focused on subspace identification algorithms and optimal structured residuals approach for processes modeling and diagnosis. Main contributions of this work include: 1. Novel subspace identification methods (SIMs) with enforced causal models are implemented. It has been shown that proposed algorithm has lower estimation variance compared to traditional SIMs. Meanwhile the rigorous analysis shows that the proposed algorithms are consistent under certain assumptions. 2. The feasibility of closed-loop subspace identification is investigated. Novel closed-loop subspace identification methods with innovation estimation are proposed. The new algorithms are shown to be consistent under closed-loop conditions, while the traditional SIMs fail to provide consistent estimates. 3. A new optimal structured residuals (OSR) approach for unidirectional fault diagnosis is proposed. The necessary and sufficient conditions for unidirectional fault isolability with OSR approach are introduced. 4. The OSR for unidirectional fault diagnosis is extended to multidimensional fault diagnosis. The sufficient condition for deterministic multidimensional fault isolability is investigated.