International Journal of Mechanical and Production Engineering Research and Development (IJMPERD) ISSN (P): 2249–6890; ISSN (E): 2249–8001 Vol. 10, Issue 3, Jun 2020, 3111-3124 © TJPR Pvt. Ltd.
GEAR FAULT DIAGNOSIS BASED ON ADAPTIVE TIME-FREQUENCY FEATURE EXTRACTION AND BSA-SVM METHOD LONG HOANG1, VANTRONG THAI2 & TUANLINH NGUYEN3 1
School of Mechanical Engineering, Hanoi University of Science and Technology, Hanoi, Vietnam 2,3
Faculty of Mechanical Engineering, Hanoi University of Industry, Hanoi, Vietnam
ABSTRACT There are many types of approaches to the detection of gear faults. The method of a novel adaptive feature ex-traction and optimized machine learning is applied to diagnose a gear faults. Firstly, the vibration signal produced by local faults is decomposed into intrinsic scale components (ISCs) by using local characteristic-scale decomposition (LCD) method and initial feature vector matrices are obtained. Then, singular value decomposition (SVD) techniques to obtain singular values extract fault feature values, while avoiding the selection of reconstruction parameters. Secondly, the Backtracking Search Optimization Algorithm (BSA), an evolutionary algorithm, is pro-posed and demonstrated to be effective though various benchmark problems. The paper proposes an optimization method for the SVM parameters based on BSA, being so called BSA-SVM. This is a new approach method applied in diagnosing gear faults. The experimental results prove this method, the results will be more accurately and shorter time cost. KEYWORDS: Signal Processing, Fault Detection, Gears, Artificial Neural Networks & Backtracking Search Optimization Algorithm
Received: May 15, 2020; Accepted: Jun 05, 2020; Published: Jul 11, 2020; Paper Id.: IJMPERDJUN2020295
Original Article
that the proposed method operates highly effective and mostly feasible for identifying gear faults in practice. By applying
1. INTRODUCTION Gears are widely applied in mechanical machines as the most important parts. Recently, the gear vibration monitoring for fault detection and diagnosis has been unceasingly investigated due to its useful application. Many researchers investigated the dynamic modeling of gearbox vibration aiming at quantifying the effect of different types of gear train damages on the resultant gear case vibration [1]. In the field of gear fault diagnosis, the method of vibration analysis plays as a crucial part of the diagnosis technologies [2]. The vibration analysis could be summed up in three steps: acquisition and preprocessing of a vibration signal; fault feature extraction [3]; and fault pattern recognition. If a pair of gears runs in faults, then the dynamic behaviour will appear to be non-stationary and nonlinear, and the vibration signals will reveal the same features. Accordingly, a method of signal decomposition is investigated to handle signals before extracting feature parameters. The original vibration signals are disintegrated on some components[3]. Among them there is alternative feature information of the original vibration signal with different bands. By doing so, the interference or coupling of the signal characteristic information can be reduced. More interestingly, the effective separations of fault features of the signal can also reflect the essence of fault information better. Fault diagnosis of rotating machinery mainly includes two key steps. One key step is feature extraction. Fault features could be generally extracted by applying a suitable vibration signal processing approach[4]. Recently,
www.tjprc.org
SCOPUS Indexed Journal
editor@tjprc.org