基于多尺度卷积神经网络的滚动轴承故障诊断方法

New method for the fault diagnosis of rolling bearings based on a multiscale convolutional neural network

为提升滚动轴承在大噪声、变载荷及复杂工况下故障诊断的准确率,考虑被采信号具有时间多尺度特性,提出多尺度卷积神经网络(MTSC-CNN),开发一种“端到端”的故障诊断系统。为验证MTSC-CNN方法的有效性,通过实验数据,对11种含故障类型、损伤程度不同以及4种存在故障混合的轴承状态进行识别。结果表明:考虑单一时间尺度提取时,因信息缺失导致模型性能欠佳;过多的时间尺度将产生信息过提取,继而增加模型复杂度,且弱化模型诊断能力。与现有方法相比,MTSC-CNN模型在复杂环境下性能更佳。此外,基于可视化技术,表明由于不同尺度所学习特征存在互补性,而使模型具有较强的鲁棒性。

In order to improving the accuracy of rolling bearings fault diagnosis in complex working environments,and considering the multiple time scale characteristic of the measured signals,a new algorithm named multiple time scale characteristic extracted convolutional neural network,(MTSC-CNN)was proposed to develop an end-to-end fault diagnosis system.The proposed MTSC-CNN was used to realize fault identification under 11 working conditions of the tested bearings,including different fault types and damage degrees for verifying the effectiveness of the proposed model.The results show that when only single time scale is considered,the performance of the model is poor due to lack of information.Too larger time scale will lead to over-extraction of information,which will increase the computational time and weaken the diagnostic capability of the model.Compared with the existing methods,the MTSC-CNN model has better performance under variable load and noise conditions.In addition,the results of neural network visualization also show that the features learned at different scales are complementary to improve the robustness of the model.