【摘要】 针对传统滚动轴承故障诊断模型存在的特征提取不充分，故障诊断准确率低的问题，提出了一种基于多特征输入和多通道并行残差卷积网络（MFI-MCP-ResNet18）的滚动轴承故障诊断方法。将轴承振动信号分别转换为相应的格拉姆角场、马尔科夫变迁场和欧氏距离矩阵，对这3种矩阵进行逐行交叉组合得到一个二维矩阵并作为神经网络的输入，通过多通道并行的ResNet18网络实现对轴承故障特征的自动提取和分类。借助公开数据集以及自建试验平台数据进行MFI-MCP-ResNet18模型的有效性和泛化性验证，结果表明MFI-MCPResNet18模型能够有效提取轴承振动信号中的故障特征，实现对轴承故障的高效诊断，具有比单输入单通道方法更高的判断精度以及更好的泛化性。更多还原

【Abstract】 Aimed at the problems of insufficient feature extraction and low fault diagnosis accuracy in traditional fault diagnosis models for rolling bearings, a method for fault diagnosis of the bearings is proposed based on multi-feature input and multi-channel parallel residual convolutional network(MFI-MCP-ResNet18). The bearing vibration signal is converted into corresponding Gram angle field, Markov transition field and Euclidean distance matrix respectively.These three matrices are cross-combined row by row to obtain a two-dimensional matrix as input of neural network.The multi-channel parallel ResNet18 network is used to realize the automatic extraction and classification of bearing fault features. The effectiveness and generalization of MFI-MCP-ResNet18 model are verified by public data sets and self-built test platform data. The results show that the MFI-MCP-ResNet18 model can effectively extract the fault features from bearing vibration signals, achieve the efficient diagnosis of bearing faults, and have higher judgment accuracy and better generalization than single-input single-channel methods.更多还原