Fault diagnosis using pattern classification based on one-dimensional adaptive rank-order morphological filter

Pattern classification is one of the major methodologies used for fault diagnosis. Employing trend modeling technique, such as nonlinear signal processing tools, to construct new pattern classification method often provides unique advantage for detecting and recognizing faults. In this paper, a novel supervised pattern classification algorithm applied to fault diagnosis is proposed on the basis of one-dimensional adaptive rank-order morphological filter (1DARMF). The algorithm adopts 1DARMF to process noised signal under supervision of each reference signal and compares output signal against corresponding reference one to find out which pairs match each other the best. Based on the procedures, it manages to recognize different noised signals. Parameters of the proposed algorithm are subject to random choice and adaptively tuned, which makes the algorithm readily be adopted in many applications. Important implementing issues such as enhancement of correct classification rate and the trade-off balance between algorithm convergence rate and computational cost are also discussed in details. Fault diagnosis for real faulted rolling bearings and Tennessee Eastman Process as case studies are presented to underline the efficacy and advantages of proposed algorithm.

► We propose fault diagnosis using adaptive rank-order morphological filter.
► It recognizes and classifies faults by signal geometry matching.
► Major parameters are subject to random choice and suitable for various applications.
► Faulted bearing diagnosis demonstrates its benefits and advantages.
► Desirable TEP fault classification validates its performances in complex process.