Trustworthy Federated Learning for Industrial IoT: Balancing Robustness and Fairness via Blockchain‐Based Reputation

ABSTRACT The integration of artificial intelligence into the industrial Internet of Things is pivotal for predictive maintenance and autonomous control. However, deploying AI in safety‐critical engineering contexts faces a dual challenge: the vulnerability to malicious data attacks and the performance disparity among heterogeneous legacy devices. To address these issues, this paper proposes Block‐FairFL , a Trustworthy Federated Learning framework empowered by Blockchain. Specifically, we design a smart contract‐based gradient auditing mechanism that detects and penalises malicious updates without reliance on a centralised authority, thereby ensuring system robustness. We introduce gradient magnitude clipping as a defence against sophisticated scaling attacks that attempt to exploit fairness mechanisms. Furthermore, we introduce a fairness‐regulated aggregation strategy that dynamically amplifies the contributions of minority clients with high local loss, effectively mitigating algorithmic bias. Extensive experiments on the NASA C‐MAPSS turbofan degradation dataset demonstrate that our approach reduces the root mean square error by approximately 12% under poisoning attacks compared with standard baselines, including q‐FedAvg, while simultaneously decreasing performance variance across clients by over 60%. These results validate that Block‐FairFL offers a resilient and ethical AI solution for complex engineering systems.