Compressed Channel-Based Edge Computing for Online Motor Fault Diagnosis With Privacy Protection

Abstract

Edge computing technology has been increasingly used in remote and real-time motor fault diagnosis. However, majority of the diagnostic algorithm is deployed onto a single processor on the edge node. These configurations require extensive computational resources on the edge node, resulting in data security and privacy problems. To address this issue, this work develops a compressed channel-based edge computing (CCEC) framework for online motor fault diagnosis. First, a compressed channel attention neural network (CCANN) model based on the CCEC framework is designed and trained by using the distributed motor condition signals. Subsequently, the well-trained CCANN model is split and deployed onto the distributed edge computing nodes by transferring the model parameters to those of the independent sub-models. Such a procedure reduces the transmitted data to one-eighth and eliminates the security issues in the communication level by leveraging the black-box nature of the neural networks. The effectiveness and superiority of the designed model have been validated on an edge computing system by using the data from a motor test rig and compared with the state-of-art methods. The proposed method can reach to 100% diagnosis accuracy in recognition of 10 types of motor statuses, and the model size and inference time are 377 KB and 0.22 ms, respectively. The results prove the several advantages, including low size, high recognition accuracy, strong robustness, and high security with privacy protection. Thus, this method shows great potential in remote and real-time motor fault diagnosis.