Block-FDT: Blockchain-enhanced federated learning approach to secure DT-assisted IIoT networks

AbstractThe Industrial Internet of Things (IIoT) has transformed modern industries by enhancing automation, efficiency, and connectivity. However, this advancement has introduced critical cybersecurity challenges that may not be addressed using conventional security measures. Intrusion detection systems (IDSs) leveraging machine learning (ML) are increasingly adopted to address IIoT security concerns. However, centralized ML models face significant privacy and security concerns. Federated Learning (FL) addresses these privacy concerns, yet FL is susceptible to Byzantine attacks that can poison global model updates. To address these challenges, this paper proposes Block-FDT, a blockchain-enhanced asynchronous FL framework with Cyber Digital Twin (CDT) for threat detection in IIoT networks. The system uses a long short-term memory (LSTM)-based CDT to predict gateway behavior across six temporal features (gradient norm, loss reduction, anomaly score, gradient variance, latency, staleness), enabling intelligent client selection through Adaptive Participation Control (APC). Blockchain integration provides tamper-proof audit trails of model aggregation, client selection, and Byzantine rejections using SHA-256 hashing with asynchronous writes. We evaluate Block-FDT on 20 % of the Edge-IIoTset dataset (88,768 samples, 6-category classification) across 20 distributed gateways, with 40 % Byzantine attacks. Experiments demonstrate that Block-FDT achieves 91.15 % detection accuracy with staleness-aware asynchronous aggregation. The blockchain introduces a minimal overhead (0.357 % latency), providing transparency without compromising system performance.