Study on nature of crossover phenomena with application to gearbox fault diagnosis. (15th January 2017)

• Record Type: Journal Article
• Authors: Jiang, Xingxing; Li, Shunming; Wang, Yong
• Is Part Of: Mechanical systems and signal processing. Volume 83(2017)
• Page Start: 272
• Page End: 295
• Publication Date: 2017-01-15
• Subjects: DFA -- Feature extraction -- Crossover phenomenon -- Piecewise-linear regression model -- Gearbox -- Fault diagnosis; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2016.06.012
• Languages: English
• ISSNs: 0888-3270

Abstract

Abstract: Detrended Fluctuation Analysis (DFA) is a robust tool for uncovering long-range correlations hidden in the non-stationary data. Recently, crossover properties of the scaling-law curve obtained by DFA have been applied to diagnose gearbox faults. However, the nature of the crossover phenomena has not been well- explained. In this paper, an explanation for the nature of crossover phenomena is specifically given, which is conducive to discovering novel features for gearbox fault diagnosis. Firstly, an explicit exposition of the crossover phenomena is provided by analyzing the gearbox vibration signal. Secondly, the nature of crossover phenomena is specifically disclosed. Thirdly, the features with clear physical meaning are proposed to describe operating conditions of a gearbox. Then, to overcome the deficiency of feature extraction through visual observation, a piecewise-linear regression model is utilized to extract the features automatically. Lastly, several combinations of these features are used to classify the fault types. As a consequence, the proposed novel features are verified that they can well- distinguish the gearbox operating conditions with different fault types and severities, and deliver a better performance than the existing method depending on the sensitive index (SI). Highlights: The nature of crossover phenomena of gearbox vibration data is explained. A set of novel features containing clear physical meaning are proposed. A piecewise-linear regression model is used to extract the features automatically. The proposed method can discriminate types and severities of gearbox faults.