Performance-Driven Subsystems Decomposition for Distributed Fault Detection

Fast and accurate fault detection is essential to the safe and economical operation of the manufacturing industry. Traditional centralized large-scale system fault detection has many limitations, such as the inability to guarantee timeliness. An effective method to detect such system faults is to divide the operating unit of the system into subsystems and implement the fault detection method synchronously in each subsystem. This paper proposes a special distributed fault detection for fusing mechanisms and data, an optimal decomposition method based on fault detection performance is developed, which is called performance-driven cluster decomposition. The optimization strategy is constructed with the optimal detection performance as the objective function and the model mechanism as the constraint condition. The detection performance is optimized based on graph-theoretic clustering algorithm and greedy search strategy. Experimental results show the effectiveness of the proposed method.