A Distributed Data-Driven Optimal Fault Detection Approach for Complex Interconnected Systems

In this paper, a distributed data-driven optimal fault detection approach for complex interconnected systems is proposed, which uses sensor networks to collect process variable data. Recall that the average consensus algorithm is generally adopted for distributed fault detection, which would inevitably result in detection delay. To deal with this issue, an alternative iterative approach is developed to achieve the optimal detection in each iteration step in this paper. To be specific, covariance coefficient matrix in the iterative process of the average consensus algorithm is first decomposed into the matrix product with the same structure, which lays the foundation for further distributed fault detection. Based on it, a distributed data-driven optimal fault detection is developed which consists of the offline training and online detection. In the offline training stage, the detection statistic function is constructed through the obtained covariance in the iterative fashion. In the online detection stage, the optimal fault detection scheme is carried out to solve detection delay problem in the iterative process of the average consensus algorithm. A case study on the Tennessee Eastman benchmark are used to demonstrate the proposed approach.