Adaptive federated sensor clustering and inference framework for real time fault prioritization in industrial Internet of Things systems

Industrial Internet of Things (IIoT) environments need fast, reliable, and context-aware fault detection in order to keep the high-hazard-rate manufacturing systems safe. Conventional cloud-centric architectures are a significant source of scalability limitations, latency, and privacy concerns when handling various sensor streams. To address these issues, Adaptive Federated Sensor Clustering and Inference Model (AF-SCIM) is proposed in this study, where solid state sensor processing is the core of the decentralized learning framework. Solid state sensors provide stable and high-fidelity signals, which result in improved local feature extraction and enhancement of the clustering and inference steps within the model. AF-SCIM applies dynamic clustering using the entropy algorithm to group sensors based on state changes, enabling real-time detection of subtle changes in system operation. It also adopts an attention-weighted federated gradient aggregation strategy that prioritizes updates from state-deviant or fault-critical sensors. Experimental results show that AF-SCIM achieves a fault-detection accuracy of 92.6% and reduces communication overhead by 37% relative to the evaluated federated-learning baselines.