Qualitative trend clustering of process data for fault diagnosis

In this paper, a qualitative trend clustering (QTC) algorithm is developed for fault diagnosis of industrial process. The clustering procedure mainly involves three steps. Firstly, the process data are segmented into consecutive episodes to which qualitative primitives will be assigned. Secondly, Smith Waterman algorithm for local sequence alignment in bioinformatics is utilized to derive the similarity measure of any two qualitative representations. Thirdly, pattern clustering analysis of process data is carried out to discriminate normal and abnormal conditions. The advantages of the proposed method include requiring less prior knowledge, being more robust to process noise and variation of signal characteristics. The application of QTC in industrial processes is illustrated on a real blowing down and recovery process of blast furnace iron making process, which shows its potentials in fault diagnosis and process monitoring tasks.