A quality-related distributed fault detection method for large-scale sequential processes

Process industries are usually composed of several coupled sub-processes, which are distributed in different positions, connected and transmitted in the form of quality flow and information flow. The long process, large-scale, dynamic coupling variables, and quality inheritance among sub-processes for process industries have brought new challenges to traditional quality-related fault detection. In this paper, a novel distributed fault detection method based on quality-related modified regularized slow feature analysis (QMRSFA) is proposed to deal with dynamics, connection relation, and outliers in large-scale sequential processes. First, robust preprocessing methods are devised to eliminate outliers, and process knowledge is utilized as a constraint to decompose the whole production process into different sub-processes. Then, a new dynamic QMRSFA method is developed as the local monitoring model. After that, an expression of the connection relation between sub-processes is given, where the quality-related slow features extracted from the previous sub-process are used as part constraint conditions of the current sub-process. In addition, the local and global indexes are established based on Bayesian fusion for quality-related fault detection. Finally, a typical large-scale sequential process, the hot strip mill process is taken as an example for verification, and the results show the practicability and feasibility of the proposed method.