Research on Fault Diagnosis of Rolling Bearing Based on Machine Learning

As the core mechanical parts widely used in industrial production, the fault detection of rolling bearings is very important for production safety and efficiency. In this paper, the signal denoising algorithm combined with machine learning algorithm is used to monitor the bearing vibration signal. The specific description is as follows: The acquisition of bearing data is the premise of bearing fault monitoring research, this paper adopts the open bearing data set for experiment. In the study of bearing vibration signal denoising, the improved wavelet value denoising algorithm is used to denoise the bearing signal. Finally, the envelope spectrum is used to analyze the fault characteristic frequency of bearing vibration signal. In order to study the machine learning algorithm of bearing fault diagnosis, the sample feature space is constructed by using the methods of time domain and frequency domain, and the sample data feature space is cleaned by the box plot method. Finally, 13 features with high correlation are selected. In order to select the best machine learning algorithm and the best number of training sets, a pre-training frame is designed, and the learning curve is drawn through the pre-training framework for comparison. Finally, the random forest algorithm is used to conduct formal training on 12 kinds of detailed fault types, and the optimal number of formal training samples is determined. In order to deal with the unbalanced data categories, the “N-fold average classifier model” is constructed by using hierarchical random division cross-validation method. In order to verify the effectiveness of the improved wavelet de-noising algorithm, the original vibration signal is de-noised by threshold and then the sample feature space is made. Compared with no improved wavelet de-noising, the F1 score of the test set is greatly improved.