Fault Diagnosis based on Improved Piecewise Linear Representation for Three-Capacity Water Tank

In this paper, a fault diagnosis method based on improved piecewise linear representation (PLR) for Three-capacity water tank system is proposed. Compared to conventional PLR algorithm, trend operations including trend judgement and trend merger are introduced and conducted on all segments. From trend judgement, every segment is labeled with '-2', '-1', '0', '1' and '2', representing the rapid decreasing, decreasing, steady, increasing and rapid increasing, respectively. After trend judgement, trend merger is utilized to correct segmentation, resulting in new segmentation reflecting data characteristics better. Then the K-means clustering algorithm is used to cluster the segments, the number of segments in each cluster is an important feature, which is used to complete fault detection. The label value differences of all measurable variables form a table, which can be used to achieve fault isolation. Finally, simulations are conducted to prove the effectiveness of the proposed method.