基于SALR优化NM算法的轴承故障信号智能诊断

Intelligent Diagnosis of Bearing Fault Signals Based on SALR Optimized NM Algorithm

考虑到采用深度学习方法进行机械设备故障诊断时无法准确识别所有故障类型的缺陷,综合运用Nesterov动量法(Nesterov LSDA,NM)和独立自适应学习率算法(Self-individual adaptive learning rate,SALR),设计了一种SALR-NM优化算法,从而实现高维度与复杂数据在深度学习过程中获得更快速率并提升学习结果准确性。研究结果表明:以RVM进行处理时,各载荷工况呈现较低精度,随着载荷增大后,信号精度也逐渐降低。各载荷工况下形成的振动信号都可以通过SALR-NM准确识别故障特征,测试10次得到的准确率都是100%。各载荷状态下形成的振动信号,以SALR-NM与PCA处理时达到了100%的最高精度,Adam则达到了最小精度,只有68.82%。本研究通过Nesterov动量法构建得到具备自适应功能的深度信念网络SALR-NM,能够直接获取频域信号的深层数据特征,从而高效诊断滚动轴承故障。

Considering that the deep learning method for mechanical equipment fault diagnosis cannot accurately identify all types of defects,Nesterov LSDA(NM)and self-individual adaptive learning algorithm are used in combination to design a SALR-NM optimization algorithm to achieve higher speed and improve the accuracy of the learning results in the process of deep learning for high-dimensional and complex data.The results show that when RVM is used for processing,the accuracy of each load condition is low,and with the increase of load,the signal accuracy gradually decreases.The vibration signals formed under various load conditions can be accurately identified by SALR-NM fault characteristics,the accuracy is 100%after 10 times of testing.When SALR-NM and PCA are used to process the vibration signals under various load states,the highest accuracy is 100%,while Adam reaches the minimum accuracy,only 68.82%.In this study,the deep belief network SALR-NM with adaptive function is constructed by Nesterov LSDA method,which can directly obtain the deep data features of frequency domain signals,so as to effectively diagnose rolling bearing faults.