基于MFCC特征与卷积神经网络的托辊故障诊断方法

A fault diagnosis method of roller based on MFCC features and convolutional neural network

针对现有托辊诊断方法诊断装置安装复杂且分析过程复杂等问题,提出一种基于Mel频率倒谱系数(Mel frequency cepstrum coefficient,MFCC)和卷积神经网络(convolutional neural network,CNN)的故障诊断方法。首先对托辊运行声音信号的时频特征进行提取,通过Mel变换构造MFCC特征参数;然后建立卷积神经网络模型,对构造的MFCC特征向量进行训练,利用多层特征提取网络将MFCC特征向量逐层变换形成抽象的深层特征;最后使用训练好的卷积神经网络进行分类,输出故障识别结果,从而实现托辊故障诊断。实验结果表明:该算法对托辊故障声音诊断的准确率达到99.64%,检测性能较好且具有较好的鲁棒性,验证了所提方法的可行性和有效性。

Focusing on the issues that complicated device installation and complicated analysis process of the existing roller diagnostic methods,this paper proposes a fault diagnosis method based on Mel frequency cepstrum coefficient(MFCC)and convolutional neural network(CNN).First,the time-frequency features of the roller running sound signal are extracted,and the MFCC feature parameters are constructed through Mel transformation.Then,a convolutional neural network model is established to train the constructed MFCC feature vector,and the multi-layer feature extraction network is used to transform the MFCC feature vector layer by layer to form the abstract deep features.Finally,the trained convolutional neural network is used to classify and output the fault recognition results,so as to realize the fault diagnosis of the roller.The experimental results show that the accuracy of the proposed algorithm for the roller faults sound diagnosis reaches 99.64%,the detection performance is good,and the robustness is also good,which verifies the feasibility and effectiveness of the proposed method.