A collaborative assurance framework for federated intrusion detection in industrial cyber-physical system

To address the challenges posed by non-independent identically distributed (Non-IID) data and adversarial attacks in federated intrusion detection for industrial cyber-physical systems (CPS), a collaborative assurance framework based on quality of service (QoS) and hyperledger Fabric (HLF) consortium blockchain technology is proposed, termed as QoS-HLF. Specifically, a QoS-awareness/evaluation (QoS-A/E) collaborative model selection mechanism is developed to quantitatively assess the collaborative service quality of intrusion detection models deployed in different regional subnetworks of industrial CPS. By selecting QoS-consistent and functionally homogeneous third party models for federated training, the proposed approach mitigates convergence degradation caused by heterogeneous data distributions. In addition, by integrating the permissioned distributed architecture of HLF with the collaborative modeling protocol, a decentralized cross-domain federated modeling platform is constructed to ensure that federated learning participants can securely initiate and engage in collaborations without relying on a trusted central authority. A QoS-A/E -driven federated collaboration chaincode is further designed to enable automatic and reliable execution of agreed protocols. The mechanism supports timely detection of participants whose model performance suddenly deteriorates, fails, or exhibits adversarial behavior, and dynamically replaces them with qualified alternatives to maintain collaboration stability. Experimental results and theoretical analysis on real industrial CPS datasets demonstrate that the proposed QoS-HLF framework achieves an F1-score of 98.29% and reduces the false positive rate to 1.28%. Meanwhile, blockchain-based evidence storage and smart contract mechanisms significantly enhance the security, trustworthiness, and automation capability of the collaborative process.The proposed framework offers an efficient and reliable solution for secure collaborative detection in industrial CPS.