A Fault Diagnosis Method for Variable Condition Bearings Based on Fast Kurtogram and Deep Residual Learning

Abstract

The invention belongs to the technical field of fault diagnosis of mechanical parts, and in particular relates to a fault diagnosis method for variable working condition bearings based on fast kurtogram and deep residual learning. In this method, the vibration acceleration signal is converted into a series of spectral kurtosis maps through the fast spectral kurtosis map method, and processed into a data set including a training set, a test set and a verification set, and a residual block and a deep convolutional neural network are constructed. The deep residual network uses the data set to train the deep residual network to obtain a trained bearing health status classification model, and the bearing health status classification model can be used to diagnose the bearing health status of the signal to be tested. The feature generated by the method of the present invention, that is, the spectral kurtosis diagram is basically free from noise interference, and the spectral kurtosis diagram of the same fault under different operating conditions has considerable similarity; Learning technology, using its powerful feature self-learning feature, further enhances the robustness of the method to noise and different working conditions.

Description

technical field [0001] The invention belongs to the technical field of fault diagnosis of mechanical parts, and in particular relates to a fault diagnosis method for variable working condition bearings based on fast kurtogram and deep residual learning. Background technique [0002] Rolling bearings are one of the key components of rotating machinery and are also the most vulnerable components (more than 40% of failures are caused by bearings). Therefore, monitoring the health status of bearings in real time is of great significance to avoid unexpected failures. In order to achieve this goal, many methods have been proposed around classical signal processing methods (such as spectrum analysis, cepstrum analysis, wavelet packet decomposition), but these methods have a common defect: lack of robustness to changing operating conditions. These methods can realize real-time diagnosis under known working conditions. However, in actual industrial production, the workload and speed...

AI Technical Summary

Problems solved by technology

In order to achieve this goal, many methods have been proposed around classical signal processing methods (such as spectrum analysis, cepstrum analysis, wavelet packet decomposition), but these methods have a common defect: lack of robustness to changing operating conditions

These methods can realize real-time diagnosis under known working conditions. However, in actual industrial production, the workload and speed are often different under different working conditions. When the working conditions change, the defect characteristic frequency will also occur. Therefore, the method has to be adjusted to adapt to the new working conditions, and if the new working condition parameters are unknown, it is difficult to complete the diagnostic task

Since then, some multi-working condition diagnosis methods have been proposed one after another: Feng Zhipeng et al. proposed to use the joint time-varying amplitude-frequency demodulation spectrum to reveal the time-varying fault characteristic frequency, but this method needs to know the experienced working conditions in advance; P.Borghesani proposed a new method of diagnosis using envelope analysis, but it needs to know the defect frequency of the bearing in advance, but in practical applications, the working conditions change frequently, and it is difficult to obtain specific working condition parameters in most cases , so these multi-condition diagnosis methods are also difficult to deal with

In addition, unavoidable noise in the signal can also have an impact on the diagnosis

Images