Investigation of Industrial Bearing Fault Diagnosis Based on 1D-Cnn-Lstm
Eduard Muratbakeev, D. Novák, Yuriy Kozhubaev
Abstract
This paper investigates a hybrid model method based on one-dimensional convolutional neural network (1DCNN) and long short-term memory (LSTM) network for fault diagnosis of industrial bearings. Based on a bearing dataset provided by Southeastern University, feature extraction and fault classification based on vibration signal data under different operating conditions were performed. The 1D-CNN module extracts local features from the signal in the time domain, while the LSTM module additionally captures the time dependence of the signal to improve the accuracy of fault diagnosis. Experimental results show that the model can effectively distinguish different types of bearing faults, exhibits good classification performance on the test set, and is highly robust to noise. This method provides an effective deep learning-based solution for industrial bearing fault diagnosis and has good application prospects.