BFLAFD: blockchain-enabled federated learning framework for adaptive fire detection in IIoT networks.

Area of Science:

• Computer Science
• Cybersecurity
• Industrial Internet of Things (IIoT)

Background:

• The Industrial Internet of Things (IIoT) in oil and gas sectors faces fire detection challenges: unreliable sensor data, privacy concerns, communication delays, and lack of generalized models.
• Centralized fire detection methods in IIoT environments are insufficient due to data confidentiality issues and communication inefficiencies.

Purpose of the Study:

• To introduce a novel Blockchain-assisted Federated Learning framework for Adaptive Fire Detection (BFLAFD) to address IIoT fire detection limitations.
• To enhance fire detection accuracy, security, and efficiency in distributed IIoT environments, specifically within the oil and gas industry.

Main Methods:

• Utilized Federated Learning (FL) for on-device model training, preserving data confidentiality and reducing data transmission.
• Implemented a hierarchical aggregation process for optimized global model performance, addressing sensor drift and device heterogeneity.
• Integrated a permissioned blockchain with smart contracts for secure access control, log transparency, and tamper-proof model integrity. Employed Personalized Federated Learning (PFL) for customized models.

Main Results:

• BFLAFD achieved 98.2% fire detection accuracy with a 2.7% false alarm rate.
• Inference latency was recorded between 100-150 ms, with blockchain validation at 1-2 seconds.
• Communication costs were reduced by 82.3% compared to centralized training methods.

Conclusions:

• BFLAFD provides a robust, secure, and efficient solution for fire detection in critical IIoT environments.
• The framework successfully overcomes challenges of data reliability, privacy, and model generalization in distributed industrial settings.