基于改进残差神经网络的滚动轴承故障检测

Fault Detection of Rolling Bearing Based on Improved Residual Neural Network

针对在矿井等特殊环境下在面对运算量大的复杂算法时,传统深度学习算法由于运算量大,现场检测设备由于需要消耗大量的资源无法完成现场检测的问题,提出了一种基于改进残差神经网络的滚动轴承故障检测方法。方法通过在卷积残差块和恒等残差块中加入跳跃连接,尽可能地减少了信息的损失,并且将部分残差块中的普通卷积替换成深度可分离卷积,大大降低了运算量。实验表明,改进残差神经网络能够有效地提取数据的特征信息,提高运算的速度,在解决恶劣环境下大数据量难以现场运算的同时对滚动轴承故障检测的准确率有很大提高,准确率可达99.97%。

In view of the problem that the traditional depth learning algorithm can not complete the on-site detection due to the large amount of computation when facing the complex algorithm with a large amount of computation under special environments such as mines,this paper proposes a rolling bearing fault detection method based on improved residual neural network.This method reduces the loss of information as much as possible by adding jump connections in convolutional residual blocks and identical residual blocks,and replaces the ordinary convolution in partial residual blocks with depth separable convolution,which greatly reduces the computational complexity.The experiment shows that the improved residual neural network can effectively extract the feature information of data,improve the speed of operation,and greatly improve the accuracy of rolling bearing fault detection,with an accuracy of 99.97%,while solving the problem that large amounts of data are difficult to operate on-site in harsh environments.