The Industrial Internet of Things (IIoT) presents significant challenges for training machine learning models due to data privacy concerns, heterogeneous data distributions, and limited bandwidth. This paper proposes HPoT (Hierarchical Proof-of-Trust), a novel federated learning-based consensus algorithm for IIoT infrastructure that addresses these limitations while ensuring data integrity and model fidelity. HPoT integrates blockchain technology with federated learning to create a decentralized, privacy-preserving framework enabling collaborative model training without raw data sharing. The algorithm features: (1) robust aggregation with weighted averaging and anomaly detection for heterogeneous datasets, (2) adaptive weighting prioritizing updates from trustworthy devices, (3) reputation scoring to detect malicious nodes, and (4) hierarchical three-tier architecture optimized for IIoT constraints. Using Byzantine fault tolerance principles, HPoT tolerates up to Formula: see text malicious participants while maintaining convergence. Experimental evaluation demonstrates Formula: see text reduction in communication rounds, Formula: see text decrease in per-node energy consumption, and Formula: see text lower consensus latency compared to traditional federated learning, while maintaining comparable accuracy (Formula: see text vs Formula: see text). The algorithm effectively handles data heterogeneity, communication constraints, and adversarial attacks common in IIoT environments. HPoT provides a scalable, secure, and energy-efficient solution for privacy-preserving machine learning on resource-constrained IIoT devices, advancing practical deployment of collaborative AI in industrial settings.
Chaurasia et al. (Tue,) studied this question.