Influence of noise on wear fault diagnosis based on recurrence plot. (December 2022)

• Record Type: Journal Article
• Authors: Zhou, Yuankai; Zhu, Rui; Zhao, Huan; Zuo, Xue
• Is Part Of: Measurement. Volume 205(2023)
• Publication Date: 2022-12
• Subjects: Recurrence plot -- Recurrence quantification analysis -- Noise -- Fault diagnosis accuracy; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2022.112158
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: The vertical and diagonal structures of RP become isolated points with the decrease of SNR. RQA quantifiers RR, DET, L, ENTR, LAM and TT decrease with SNR. Noise leads to the enhancement of signal randomness and the weakening of stability and predictability. The accuracy of wear fault diagnosis based on RP can be enhanced to 99.2% when using the denoised signal. Abstract: Recurrence plot (RP) and recurrence quantification analysis (RQA) have been widely applied in fault diagnosis of tribo-systems. The influence of noise on the RP, RQA quantifiers and RP-based wear fault diagnosis was studied by analyzing theoretical signals and vibration signals. With the decrease of signal-to-noise ratio (SNR), the vertical and diagonal structures of RP of noisy Lorenz and Rössler signals are discretized into isolated points, and RQA quantifiers decrease gradually. This effect is due to the interference of noise on the phase trajectory of original signals. Noise makes some phase points escape the hypersphere, which leads to the disappearance of recurrence relationship. Recurrence plot-convolutional neural network (RP-CNN) can diagnose wear forms with 99.2% accuracy by analyzing the denoised vibration signals, which is 2.2% higher than that without denoising. The results will help to improve the reliability and accuracy of recurrence analysis applied in fault diagnosis of tribo-systems.