Open AccessBased on Improved CNN Bearing Fault Detection
Author(s) -
Yang Xu,
Lei Yang
Publication year - 2022
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/2171/1/012073
Subject(s) - convolutional neural network , discriminative model , fault (geology) , bearing (navigation) , computer science , pattern recognition (psychology) , artificial intelligence , signal (programming language) , vibration , artificial neural network , feature extraction , fault detection and isolation , mechanism (biology) , deep learning , sample (material) , actuator , acoustics , philosophy , chemistry , physics , epistemology , chromatography , seismology , programming language , geology
In recent years, the problem about the fault detection of rolling bearings in mechanical equipment has gradually become an important research direction.Among them, the diagnostic method based on vibration signal analysis is widely used in the fault detection of rolling bearings.Since the one-dimensional convolutional neural network (1D-CNN) has certain limitations on the processing of vibration signal data, the solution to this problem in this paper is to integrate the attention mechanism and bi-directional long and short-term memory neural network (BiLSTM) on the basis of the one-dimensional convolutional neural network, using the attention mechanism to give different weights to different feature dimensions in the sample data and extract key and important information, thus the sample data can be further optimized.On the other hand, BiLSTM can automatically extract the deep information of the bearing vibration signal, which makes up for the deficiency of artificial extraction features to a certain extent, and strengthens the discriminative property of high-level features.Subsequently, the improved CNN bearing fault detection model was experimentally validated using the Case Western Reserves University bearing dataset, and it was concluded that the attention mechanism acting on the model obtained by adding BiLSTM to the 1D-CNN could achieve a fault identification accuracy of about 98.9% and the loss degree was reduced to about 0.17%, thus achieving an effective diagnosis of the fault state.