Mahalanobis semi-supervised mapping and beetle antennae search based support vector machine for wind turbine rolling bearings fault diagnosis

Abstract

Intelligent fault diagnosis of wind turbine rolling bearings is an important task to improve the reliability of wind turbines and reduce maintenance costs. In this paper, a novel intelligent fault diagnosis method is proposed for wind turbine rolling bearings based on Mahalanobis Semi-supervised Mapping (MSSM) manifold learning algorithm and Beetle Antennae Search based Support Vector Machine (BAS-SVM), mainly including three stages (i.e., feature extraction, dimensionality reduction, and pattern recognition). In the first stage, Multiscale Permutation Entropy (MPE) is utilized to extract the feature information from rolling bearing vibration signals at multiple scales, while a high-dimensional feature set is constructed. Second, the proposed MSSM algorithm, combining the advantages of Mahalanobis distance, semi-supervised learning and manifold learning, is applied to reduce the dimension of high-dimensional MPE feature set. Subsequently, low-dimensional features are input to the BAS-SVM classifier for pattern recognition using the BAS algorithm to search the best parameters. The performance of the proposed fault diagnosis method was confirmed by conducting a fault diagnosis experiment of wind turbine rolling bearings. The application results show that the proposed method can effectively and accurately identify different states of wind turbine rolling bearings with a recognition accuracy of 100%.