基于DMD和改进胶囊网络的变工况轴承故障诊断方法

Fault diagnosis method of variable working condition bearing based on DMD and improved capsule network

针对轴承单一传感器所测数据存在不全面性和简单利用多传感器数据造成数据过多的问题,引入动态模态分解(DMD)的信号分解方法,将多传感器信号分解为多个模态,并提出利用能量值最大的模态对原始信号进行重构;针对变工况的问题,首先引入含有矢量神经元的胶囊网络(CN),然后提出在胶囊网络中加入转置卷积,构建改进的胶囊网络模型(ICN)充分提取输入数据的空间信息,对故障特征进行智能识别。基于DMD和ICN的轴承故障诊断方法既可以利用多传感器信号,同时也不会造成数据冗余;此外,ICN可以充分提取不同数据的空间信息,并通过动态路由算法计算胶囊层的相关度,实现在变工况下对轴承故障的精确诊断。实验结果表明,基于DMD和ICN模型的轴承智能故障诊断方法,比传统卷积神经网络(CNN)和未改进的胶囊网络具备更强的变工况故障诊断能力。

In order to address the issues of incomplete data measured by a single bearing sensor and the problem of excessive data caused by using multiple sensors,a signal decomposition method called dynam-ic mode decomposition(DMD)was introduced.The multi-sensor signal was decomposed into multiple modes,and the mode with the highest energy value was proposed to reconstruct the original signal.To solve the problem of variable working conditions,capsule network(CN)containing vector neurons was introduced,and then transposed convolution was added to the capsule network to construct improved cap-sule network(ICN)to fully extract the spatial information of input data.Intelligent identification of fault characteristics was conducted.The bearing fault diagnosis method based on DMD and ICN not only lever-ages multi-sensor signals but also avoids data redundancy.Furthermore,ICN effectively extracts spatial information from different data and calculates the correlation of the capsule layer using the dynamic rou-ting algorithm,enabling accurate diagnosis of bearing faults under varying working conditions.Experimen-tal results demonstrate that the intelligent bearing fault diagnosis method based on DMD and ICN models outperforms traditional convolutional neural networks(CNN)and unimproved capsule networks in terms of its ability to diagnose faults under variable working conditions.