Fault Diagnosis of Rolling Bearing Based on Multidimensional Feature Extraction and Optimized LSSVM Algorithm

• Published in: 2023 14th International Conference on Reliability, Maintainability and Safety (ICRMS) pp. 1291 - 1297
• Main Authors: Liu, Wei, Mu, Huina, Wei, Shijie, Yan, Hongmei
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.08.2023
• Subjects: Accuracy; Classification algorithms; Data mining; Entropy; fault diagnosis; Fault location; Feature extraction; least squares support vector machine; multidimensional feature extraction; rolling bearing; Rolling bearings; Support vector machines; Vectors; Wavelet packets
• ISSN: 2575-2642
• DOI: 10.1109/ICRMS59672.2023.00221

Herein, a multidimensional feature extraction method based on permutation entropy, wavelet packet decomposition, and empirical mode decomposition (EMD); and a stepwise fault diagnosis method based on particle swarm optimization (PSO), artificial bee colony algorithm (ABC) and least squares support vector machine (LSSVM) are proposed for rolling bearings with multiple fault states. First, the normalized bearing initial signal is decomposed by wavelet packet decomposition and EMD. Next, 31-dimensional feature vectors are extracted as input for fault diagnosis classifier. Finally, a three-step fault diagnosis is constructed on the basis of optimized LSSVM classifiers, consisting of fault existence detection, fault localization, and fault degree judgment. According to different steps, the corresponding feature vectors are used as input for the classifiers. Through the simulation analysis of bearing data, the five-fold cross-validation accuracy of the three-step fault diagnosis LSSVM classifier are 100%, (96.3%, 95.2%, 90.9%), and (97.5%, 92.5%, 91.7%). The identification and diagnosis of the fault location and degree of the rolling bearing is realized. Additionally, the validity of the proposed method is verified.