基于声信号和一维卷积神经网络的电机故障诊断研究

针对电机故障诊断问题,设计一种新型的一维卷积神经网络结构(1D-CNN),提出一种基于声信号和1DCNN的电机故障诊断方法。为了验证1D-CNN算法在电机故障识别领域的有效性,以一组空调故障电机作为实验对象,搭建电机故障诊断平台,对4种状态的空调电机进行声信号采集实验,制作电机故障声信号数据集,并运用1DCNN算法对数据集进行分类,计算出基于该算法的电机故障识别准确率。实验结果表明,1D-CNN算法作为一种新型结构深度学习算法,能够对电机故障声信号进行很好分类,分类准确率高于FFT-BP、SVM、FFT-SAE等算法。为了探究1D-CNN算法内在机制,还对1D-CNN算法性能进行t-SNE可视化分析。

Motor Fault Diagnosis Using Microphones and One-Dimensional Convolutional Neural Network

A new structure of one-dimensional convolutional neural network(1D-CNN)is designed and a novel method for motor fault diagnosis based on acoustic signals and the 1D-CNN is proposed.In order to verify the effectiveness of the 1D-CNN algorithm in motor fault classification,an experiment is carried out on a set of air conditioner motors.A motor fault diagnosis platform is established to collect acoustic signals of the air conditioner motors in four states,and a dataset of faulted acoustic signals is established.The 1D-CNN algorithm is then applied to classify the dataset and the classification accuracy is evaluated and analyzed.The experimental results show that the 1D-CNN algorithm,as a new type of deep learning algorithm,can well classify motor faults using the acoustic signals,and the classification accuracy is higher than the other algorithms such as FFT-BP,SVM and FFT-SAE.In order to explore the internal mechanism of the 1D-CNN algorithm,this paper also conducts a t-SNE visualization analysis for the performance of the 1D-CNN algorithm.