A rotor fault diagnosis method based on BP-Adaboost weighted by non-fuzzy solution coefficients

• Published in: Measurement : journal of the International Measurement Confederation 2022-06, Vol.196, p.111280, Article 111280
• Main Authors: Liu, Yang, Zhao, ChenCheng, Liang, HaiYing, Lu, HuanHuan, Cui, NingYuan, Bao, KuiYuan
• Format: Article
• Language: English
• Subjects: Classification; Classifiers; Decision making; Fault diagnosis; Features sequence; Fuzzy logic; Mechanical systems; NFSW-BP-AdaBoost; Rotors; Signal processing; Vibration
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2022.111280

•Analyzed the drawbacks of binary classification algorithms in multi-classification problems.•A comprehensive evaluation index for fault features is proposed.•An improved machine learning method for rotor fault diagnosis. To ensure the long-term safe and reliable operation of mechanical equipment, it is necessary to carry out fault diagnosis of key components in the mechanical system. A rotor fault diagnosis method based on non-fuzzy solution weighted BP-AdaBoost (NFSW-BP-AdaBoost) is proposed in this paper. First, the rotor vibration signal is transformed into a high-dimensional feature set. Then, the VC value is used to obtain the high-dimensional feature sequence. Finally, the NFSW-BP-AdaBoost method is used to diagnose the type and degree of rotor fault. The core of this article is to analyze the fuzzy solution of the combined binary classifier applied to multi-classification problems in decision-making, and put forward the non-fuzzy solution coefficient to weigh the conclusion of the combined multi-classifier, highlighting the sensitivity to the non-fuzzy solutions of each multi-classifier. The experimental analysis results show that the recognition rate of the classification method for fault types can reach more than 97%.