基于混合时间序列卷积神经网络的轴承故障诊断

在传统滚动轴承故障诊断中,绝大多数方法采用了从振动信号提取特征的诊断模式,但是这种模式必然会使原始信号降维进而导致故障信息的丢失。卷积神经网络(CNN)通过权重共享和稀疏连接直接对原始信号进行操作,实现自适应特征提取,最大化保留故障信息。受CNN原理启发,开发出了一种基于工业振动信号特征的新型诊断框架,称之为混合时间序列CNN(HTS-CNN)。首先,利用估计总体比例的方法自适应确定模型训练样本数目;其次,通过对时间序列片段进行随机组合的方式,使模型能够提取非相邻信号特征;最后,利用Softmax激活函数在模型输出端执行多分类任务。通过对凯斯西储大学及CUT-2平台轴承数据进行分析,实验结果表明:该方法能够准确、有效的对滚动轴承故障进行分类。

Bearing Fault Diagnosis Based on Mixed Time Series Convolutional Neural Network

In the traditional fault diagnosis of rolling bearings,extracting features from vibration signals is most methods diagnostic mode,but this mode will inevitably reduce the original signal and may lead to the loss of fault information.The Convolutional Neural Network(CNN)directly operates the original signal through weight sharing and sparse connection to achieve adaptive feature extraction and maximize the retention of fault information.Inspired by the CNN principle,we developed a new diagnostic framework based on the characteristics of industrial vibration signals,called hybrid time series CNN(HTS-CNN).Firstly,the number of model training samples is adaptively determined by the method of estimating the overall proportion.Then,by randomly combining the time series segments,the model can extract non-adjacent signal features.Finally,the Softmax activation function is used to perform multi-classification tasks at the model output.The bearing datas of Case Western Reserve University and CUT-2 platform are analyzed in the experiment.The experimental results show that the method can accurately and effectively classify rolling bearing faults.