深度卷积神经网络在滑动轴承转子轴心轨迹识别中的应用

针对传统旋转机械智能识别方法需要人为提取特征及诊断精度低的问题,基于深度学习的强大学习能力,提出一种深度卷积神经网络故障诊断模型(Deep Convolutional Neural Network Fault Diagnosis Model,DCNN-FDM)用于轴心轨迹识别。该模型包括输入模块、特征提取模块及分类模块三部分。原始图像输入模型后,经过输入模块的二值化处理及最近邻插值,统一变为尺寸大小为32×32的单通道图像;经特征提取模块中两组交替的卷积层和池化层作用,得到图形特征;最后,这些特征经全连接层的扁平化处理而张成一维向量,输入到softmax分类器中进行分类。利用奇异值差分谱方法,对实测轴心轨迹进行提纯,得到4类轴心轨迹样本集用于DCNN-FDM的训练与预测。结果表明:所提模型较传统的浅层学习模型的识别效果好,可实现转子故障的精确诊断,识别率达到97.09%。最后通过全连接层的主成分可视化分析,验证了模型具备自适应特征学习能力。

Application of deep convolution neural network in identification of journal bearing rotor center orbit

Here,aiming at the problem of traditional rotating machinery intelligent identification method requiring artificial feature extraction with low diagnostic accuracy,based on deep learning having strong learning ability,a deep convolutional neural network fault diagnosis model(DCNN-FDM)was proposed to identify rotor axis center orbit.The model included 3 parts of input module,feature extraction one and classification one.After the original images were input into the model to do binary processing and nearest neighbor interpolation of input module,they were unified into a single channel images with size of 32×32.Then,through interaction of two groups of alternating convolution layers and pooling layers in feature extraction module,the graphic features were obtained.Finally,through flattening treatment of the full connection layer,these features were expanded into one-dimensional vectors to be input into SOFTMAX classifier for classification.By using the singular value difference spectrum method,four kinds of axis center orbit samples were obtained for training and prediction of DCNN-FDM.The results showed that the recognition effect of the proposed model is better than that of the traditional shallow learning model to realize the accurate diagnosis of rotor faults,and the recognition rate reaches 97.09%;the visual analysis of main components of the full connection layer verifies the model having the adaptive feature learning ability.