A two-stage fault diagnosis methodology for rotating machinery combining optimized support vector data description and optimized support vector machine. (15th August 2022)

• Record Type: Journal Article
• Authors: Zhang, Jianqun; Zhang, Qing; Qin, Xianrong; Sun, Yuantao
• Is Part Of: Measurement. Volume 200(2022)
• Publication Date: 2022-08-15
• Subjects: Fault diagnosis -- Rotating machinery -- Support vector data description -- Support vector machine -- Refined composite multiscale fuzzy entropy; SVDD Support vector data description -- MSE Multiscale sample entropy -- MPE Multiscale permutation entropy -- CWRU Case Western Reserve University -- RCMSE Refined composite multiscale sample entropy -- RCMFE Refined composite multiscale fuzzy entropy -- IMPE Improved multiscale permutation entropy -- MWPE Multiscale weighted permutation entropy -- PSO Particle swarm optimization -- GOA Grasshopper optimization algorithm -- MSMR Maximum stability and minimum redundancy -- IMS Intelligent Maintenance Systems -- COM The complex index -- AdaKNN2GIHS Adaptive KNN2 coupled with the global imbalance handling scheme -- SVM Support vector machine -- ME Multiscale entropy -- PE Permutation entropy -- RMS Root mean square -- SE Sample entropy -- FE Fuzzy entropy -- KNN K-nearest neighbor -- MFE Multiscale fuzzy entropy -- GA Genetic algorithm -- MI Mutual information -- FA Firefly algorithm -- GWO Gray wolf optimization -- CSO Chicken swarm optimization -- MPA Marine predators algorithm -- AO Aquila optimizer; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2022.111651
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A two-stage intelligent fault diagnosis methodology combing GOA-SVDD and GOA-SVM is proposed for rotating machinery. MSMR is proposed for feature selection when only normal samples are used for training. The feature extraction capabilities and robustness of eight entropy-based indicators are compared by three case studies. Abstract: Most intelligent fault diagnosis methods of rotating machinery generally consider that normal samples and fault samples as equally important for pattern recognition training. It ignores that the rotating machinery is in a normal state most of the time, and the collected normal samples are far more than the fault samples. Considering this situation, this paper proposes a two-stage intelligent fault diagnosis methodology for rotating machinery combing optimized support vector data description (SVDD) and optimized support vector machine (SVM). Specifically, SVDD is applied for fault detection, and SVM is applied for fault identification. The parameters of SVDD and SVM are optimized by the grasshopper optimization algorithm (GOA). Multiscale entropy (ME) is used for feature extraction, which is the input feature vector of SVDD and SVM. Under the framework of the proposed methodology, the strengths and weaknesses of 8 different entropy-based indicators in feature extraction are explored. The analysis results of three cases prove the availability and universality of the proposed methodology. According to the analysis results of three case analyses, the permutation entropy (PE)-based indicators are not promising for fault detection. Refined composite multiscale fuzzy entropy (RCMFE) is recommended for fault detection and fault identification to generalize the universality. Based on RCMFE, the fault detection accuracy in the three cases is more than 99%, and the fault identification accuracy is more than 94%. This methodology, therefore, provides a powerful tool for rotating machinery fault diagnosis, which is beneficial to practical application.