Bearing Fault Diagnosis Method Based on Deep Learning and Health State Division

Abstract

As a key component of motion support, the rolling bearing is currently a popular research topic for accurate diagnosis of bearing faults and prediction of remaining bearing life. However, most existing methods still have difficulties in learning representative features from the raw data. In this paper, the Xi’an Jiaotong University (XJTU-SY) rolling bearing dataset is taken as the research object, and a deep learning technique is applied to carry out the bearing fault diagnosis research. The root mean square (RMS), kurtosis, and sum of frequency energy per unit acquisition period of the short-time Fourier transform are used as health factor indicators to divide the whole life cycle of bearings into two phases: the health phase and the fault phase. This division not only expands the bearing dataset but also improves the fault diagnosis efficiency. The Deep Convolutional Neural Networks with Wide First-layer Kernels (WDCNN) network model is improved by introducing multi-scale large convolutional kernels and Gate Recurrent Unit (GRU) networks. The bearing signals with classified health states are trained and tested, and the training and testing process is visualized, then finally the experimental validation is performed for four failure locations in the dataset. The experimental results show that the proposed network model has excellent fault diagnosis and noise immunity, and can achieve the diagnosis of bearing faults under complex working conditions, with greater diagnostic accuracy and efficiency.

An identification method of compound faults of rolling bearings blending variational mode decomposition and vector bispectrum

“…When a rolling bearing develops a fault, it’s usually accompanied by intensified impulse impacts, leading to significant changes of signal amplitude; RMS is particularly sensitive to bearing faults. 19 Mi et al 20 employed mean, variance, and RMS values to design a novel threshold for anomaly detection. This method exhibits heightened sensitivity to anomalies indicative of bearing degradation, outperforming classic detection methods with a lower error rate.…”

“…A prominent frequency at 854.5 Hz, which corresponds to eight times the inner ring fault characteristic frequency subtract twice the rotational frequency: ((854.5 + 2f r )/8 = 113 19…”

Gaussian-Linearized Transformer with Tranquilized Time-Series Decomposition Methods for Fault Diagnosis and Forecasting of Methane Gas Sensor Arrays

“…Recently, DL [20][21][22] has been used to extract high dimensional features of fault diagnosis data and classify them directly, avoiding the shortcoming of requiring handcrafted features designed by engineers [23,24]. Therefore, a significant amount of DL methods have been extensively used in fault diagnosis [25,26]. Wen et al [27] used a convolutional neural network (CNN) for fault diagnosis.…”