Decentralized fault detection and diagnosis via sparse PCA based decomposition and Maximum Entropy decision fusion

► We propose a decentralized model for fault detection and diagnosis evaluated and compared to centralized model on Tennessee Eastman Process. ► The proposed Sparse PCA-based process decomposition leads to better fault detection and diagnosis results than the topology-based plant decompositions. ► The proposed Maximum Entropy algorithm is a much more effective tool for decision fusion than the baseline method. ► Sparse PCA decomposition and MaxEnt fusion form a decentralized model whose performance can approach that of a centralized model. ► The resulting decentralized model provides increased flexibility and fault tolerance.