An Adaptive Multihop Branch Ensemble-Based Graph Adaptation Framework With Edge-Cloud Orchestration for Condition Monitoring

Abstract

Condition monitoring plays a crucial role in securing smooth production, which has been facilitated into a cyber-physical system (CPS) integration paradigm with the new information and communication technologies and data-driven intelligence. However, traditional methods limit its successful deployment from the different distribution of training data and testing data, the missing relationships between the input signals, and the insufficient data size. To overcome these limitations, a novel graph-based adaptation framework with edge-cloud orchestration is proposed. A three-stage edge-cloud orchestration mechanism is encapsulated with CPS architecture. The proposed graph-based approach mainly consists of an adaptive multi-hop branch ensemble module to intelligently aggregate the node information, a distance metric learner to autonomously align the data distribution, and a classifier module to automatically generate the pseudo-labeled data to guide the edge-cloud orchestration and output results. Finally, a real-life case study and extensive experiments are conducted to prove the effectiveness of the proposed approach.