基于融合特征的双通道CNN滚动轴承故障识别

Dual Channel CNN Bearing Fault Identification Based on Fusion Feature

针对传统的滚动轴承故障识别方法效果较差,对专家经验依赖较高的问题,提出一种基于融合特征的双通道CNN滚动轴承故障识别方法。该方法首先将原始信号采用小波分解方法生成时频图,再将时频图和原始故障信号融合输入到Lenet5网络中,进一步对故障特征进行准确提取,在输出层对数据进行融合,使用Softmax分类器对轴承故障进行分类。实验结果表明,该方法对不同种类的滚动轴承故障的识别均能做出准确的判断,识别准确率高。

The traditional fault identification method has poor effect and relies more on expert experience,so to solve this problem,a twochannel CNN rolling bearing fault identification method based on fusion features was proposed.In this method,the original signal was first generated by Morlet wavelet method,and then the timefrequency diagram and the original fault signal were fused and input into lenet5 network,and the fault features were further accurately extracted.The data were fused at the output layer,and the faults were classified by Softmax classifier.The experimental results show that this method can make accurate judgment on the identification of different kinds of rolling bearing faults with high accuracy.