Abstract: Aiming at the challenge of multi-position and multi-type fault diagnosis of rolling bearing,a method of rolling bearing fault diagnosis based on spatio-temporal feature fusion was proposed.The Long Short Term Memory network (LSTM) was adopted to extract the time series features of the bearing data set,and the improved one-Dimensional Full Convolution Network (1D-FCN) was used to extract the spatial features of the vibration acceleration signal of the rolling bearing.Then,the innovative full connection layer algorithm was designed to fuse the spatial and temporal features and update the network parameters.Finally,the proposed multi-classification algorithm was applied to identify different positions and different fault types of the rolling bearing.Experimental results showed that the proposed method had more significant feature extraction ability than Convolutional Neural Network (CNN),Support Vector Machine (SVM) and other methods,and the final classification accuracy was better than the above traditional methods,which proved the effectiveness and superiority of this method.

Key words: rolling bearing, multi-fault classification, spatio-temporal features, one-dimensional full convolution network, long short term memory network

摘要： 针对滚动轴承多位置、多类型故障诊断面临的挑战,提出一种基于时空特征融合的滚动轴承故障诊断方法。首先利用长短时记忆网络(LSTM)提取轴承数据集的时间序列特征,利用改进的一维全卷积网络(1D-FCN)提取滚动轴承振动加速度信号空间特征,再使用创新全连接层算法融合时空特征、更新网络参数,最后利用所提多分类算法实现对滚动轴承不同位置和不同故障类型的识别。实验结果表明,所提方法和卷积神经网络(CNN)、支持向量机(SVM)等方法相比,具有更显著的特征提取能力,最终的分类准确率优于上述传统方法,证明了该方法的有效性和优越性。

关键词: 滚动轴承, 多故障分类, 时空特征, 一维全卷积网络, 长短时记忆网络