Abstract The Industrial Internet of Things (IIoT) is a network of interconnected sensors, devices, and control systems in the oil and gas sectors that has been developed to make the industries automated and continuously monitored. However, there are challenges in fire detection in such environments, including the unreliable nature of the sensor data, privacy issues, communications delays, and the lack of a generalized model across locations in a distributed solution. To overcome the above problems, BFLAFD, Blockchain-assisted Federated Learning framework for Adaptive Fire Detection is introduced. Unlike centralized methods, BFLAFD makes use of Federated Learning (FL), where local edge servers train models directly on-device in which data confidentiality is upheld and less data is transmitted. Hierarchical aggregation process is effective in maximizing global performance, in addition to being sensitive to sensor drift and device heterogeneity. To make sure of trust and resilience, BFLAFD combines the permissioned blockchain with smart contracts providing access control, transparency of logs and no tampering of models or insider manipulation. Furthermore, Personalized Federated Learning (PFL) makes it possible to create a customized fire detection model, effectively enhancing the accuracy in varying conditions. Experimental evaluations have shown that BFLAFD has 98.2% detection accuracy, false alarm rate of 2.7%, and a 100–150 ms inference latency, and blockchain validation time of 1–2 s. In addition, the cost of communication was reduced by 82.3% compared to centralized training. Overall, BFLAFD offers critical IIoT environments fast, accurate, and secure fire detection solutions.
Desikan et al. (Tue,) studied this question.