Machine learning-based bearing failure research

Abstract

To ensure the safe operation of aircraft engine bearings under extreme conditions such as high temperatures, high pressures, and high-speed rotation, and to address their susceptibility to failure, this study explores a machine learning-based bearing fault diagnosis method. The core of the research lies in enhancing diagnostic accuracy through effective feature engineering strategies: first, multidimensional features are extracted from both the time and frequency domains of bearing vibration signals; subsequently, key features are selected using variance analysis and the Gini coefficient, with Principal Component Analysis employed for dimensionality reduction to retain core information. Performance comparisons of models including One-Dimensional convolutional neural networks, logistic regression, random forests, and gradient-boosted trees demonstrated that random forests combined with Gini coefficient feature selection achieved optimal results. This approach attained an exceptionally high accuracy of 0.9872 on the test set while exhibiting robust generalisation capabilities. This research confirms that traditional machine learning models, optimized through manual feature engineering, can provide a ‘high-precision, low-risk’ solution for bearing fault diagnosis. It offers significant reference value for the intelligent operation and maintenance of aero-engines and other industrial equipment.