Bearing fault detection method based on machine learning

Abstract

Traditional fault diagnosis for bearing usually requires a large amount of labeled data for training and deliberate selection of features and the diagnostic results are often too scattered to provide a thorough fault diagnosis procedure. To address these issues, we use data mining techniques to analyze the raw vibration signals of rolling bearings in various fault states under different operating conditions and construct labeled training and test datasets. By introducing the attention mechanism, we utilize convolutional neural networks and bidirectional long and short-term memory prediction models to diagnose deep fault categories and improve prediction accuracy from the perspective of machine learning. In addition, to intuitively describe the operating state of equipment bearings, we construct a visualization interface based on the prediction model. After example analysis, the model constructed in this paper has certain feasibility and effectiveness.