Fault detection in wastewater treatment plants using distributed PCA methods

This paper proposes a distributed fault detection and diagnosis method based on Principal Component Analysis (PCA) in a whole plant monitoring scheme. The method is based on the decomposition of the plant into multiple blocks using plant topology. A local PCA based fault detection method is applied in each block and the results are sent to the central node to fuse the information and to detect and diagnose faults in the global plant. This method is compared with the centralized PCA method and some distributed principal component analysis (DPCA) methods in a wastewater treatment plant (WWTP). The objective is to check which of the distributed methods implemented is the best one in terms of detecting faults and minimizing the communication cost between the blocks. Empirical results on the WWTP show that the DPCA method based on local models has very good results.