基于双通道特征融合的CNN-LSTM轴承故障诊断方法

CNN-LSTM Bearing Fault Diagnosis Method Based on Dual Channel Feature Fusion

针对滚动轴承故障诊断过程中特征提取复杂、难以捕获时间序列数据之间的长期依赖问题,提出了双通道特征融合的CNN-LSTM故障诊断模型。首先将原始振动数据同时输入到CNN中和LSTM通道中,其次利用CNN和LSTM的各自优势分别提取原始振动数据的空间特征以及时序信息特征,将提取到的特征信息进行融合,最后将融合后的特征输入到softmax完成故障分类。经实验证明,与CNN-LSTM、CNN、LSTM 3种故障诊断模型相比,所提出的模型直接以原始信号进行故障诊断的准确率可达99.64%,且融合后的特征更容易区分不同故障状态。在不同噪音背景下,所提模型也保持了95%以上的故障诊断准确率,具有较好的抗噪性。

Aiming at the complexity of feature extraction and the difficulty in capturing the long-term dependence between time series data in rolling bearing fault diagnosis,a CNN-LSTM fault diagnosis model based on dual-channel feature fusion was proposed.Firstly,the original vibration data are introduced to CNN and LSTM channels at the same time.Secondly,the spatial features and timing information features of the original vibration data are extracted respectively by using the respective advantages of CNN and LSTM.The extracted feature information is then fused.Finally,the fused features are input into softMax to complete fault classification.Comparing with the CNN-LSTM,CNN and LSTM fault diagnosis models,experimental results demonstrate that the accuracy of the model proposed in this paper can reach 99.64%by directly using original signals for fault diagnosis.Moreover,it is easier to distinguish the fault states from the fused features by using the developed model.Under different noise background,Additionally,the proposed model maintains more than 95%of fault diagnosis accuracy under various noise background thus presents good noise resistance.