Cloud Fault Diagnosis Method of a Service Robot Motion System Based on SE-CNN

A fault diagnosis system based on a service robot cloud platform is studied and designed. Since the traditional algorithms only focus on the state data of the service robot at a particular time, the extracted feature information is limited. Therefore, it is difficult for the traditional algorithms to complete the fault diagnosis task well. In this context, a fault diagnosis method based on the correlation features of the time series is proposed. Firstly, the collected service robot data are pre-processed by normalization and backward difference to eliminate the data dimension and obtain the data change characteristics. Secondly, the sliding window is employed to generate time series samples to ensure that each sample contains sufficient feature information. Then, convolutional neural network (CNN) is applied to extracting correlation features of time series, and channel attention networks (squeeze-and-excitation network, SENet) are introduced to construct an SE-CNN model, which can adaptively adjust the importance of feature channels and focus on more effective feature channels, thereby improving diagnosis accuracy. Comparative experiments and comprehensive test in the actual scene prove the feasibility and effectiveness of the fault self-diagnosis method proposed.