A GOA-MSVM based strategy to achieve high fault identification accuracy for rotating machinery under different load conditions

• Published in: Measurement : journal of the International Measurement Confederation Vol. 163; p. 108067
• Main Authors: Zhang, Jianqun, Zhang, Jun, Zhong, Min, Zheng, Jinde, Yao, Ligang
• Format: Journal Article
• Language: English
• Published: London Elsevier Ltd 15.10.2020; Elsevier Science Ltd
• Subjects: Algorithms; Bearings; Design optimization; Diagnostic systems; Fault detection; Fault diagnosis; Fault identification; Gearboxes; Grasshopper optimisation algorithm; Machinery; Mixed kernel-support vector machine; Pattern recognition; Roller bearings; Rotating machinery; Sample entropy; Strategy; Support vector machines; Variational mode decomposition; Sample entropy; Variational mode decomposition; Mixed kernel-support vector machine; Grasshopper optimisation algorithm; Rotating machinery; Fault identification
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2020.108067

•GOA-MSVM method can identify both fault type and fault severity of rolling bearings.•GOA-MSVM is proved to be able to identify weak faults of planetary gearbox.•VMD SE is used as a part of feature vector and compared with VMD AE and EMD SE.•GOA-MSVM is more suitable to identify rotating machinery faults than other 5 SVM-based methods. Identifying fault of rotating machinery under different load conditions with high accuracy is a remaining challenge for vibration signal based fault diagnosis. Aiming at this challenge, this paper proposes a comprehensive strategy of combining mixed kernel-support vector machine (MSVM) with grasshopper optimisation algorithm (GOA) to identify typical faults of rotating machinery subject to different load levels. The basic idea of the proposed strategy can be summarized as the following three steps. Firstly, a feature vector that uses multi-domain indexes containing the sample entropy (SE) of variational mode decomposition (VMD) is constructed to characterize the fault information. Secondly, a MSVM model containing six design variables is established and then optimized by GOA. Finally, the optimized MSVM model is adopted to train the fault feature vectors to fulfill fault pattern recognition. In order to verify the identification accuracy of the proposed strategy, two sets of fault signal generated from a rolling bearing test rig and a laboratory planetary gearbox operating under different load conditions are analyzed. The diagnostic results manifest that the proposed strategy can fully identify different level faults of rolling bearing and weak faults of planetary gearbox as well. More than 99% identification accuracy of the proposed strategy is further highlighted by comparisons with other five SVM-based methods. The present method provides a promising solution for high fault identification accuracy in rotating machinery working under different loads.