BCE-FL: A Secure and Privacy-Preserving Federated Learning System for Device Fault Diagnosis Under Non-IID Condition in IIoT

Traditional device fault diagnostic methods in Industrial Internet of Things (IIoT) require nodes to upload local data to the cloud, which, however, may lead to privacy leakage issues. Although Federated learning (FL) methods can protect the privacy of data, many challenges still need to be addressed. For example, the nonindependently and identically distributed (non-IID) issue in FL prevents the convergence of global model. Moreover, FL lacks detection mechanism to resist poisoning attacks from malicious nodes, and it requires incentive mechanism to encourage nodes to share their resources. To address these challenges, this article proposes a secure and privacy-preserving FL system that leverages blockchain and edge computing technology. Specifically, a feature-contrastive loss function is constructed to train an unbiased global model under the non-IID condition. Additionally, a Byzantine-tolerance scoring mechanism is designed to resist poisoning attacks, and a reputation-based incentive algorithm is developed to estimate the rewards or penalties owed to nodes. The proposed method is applied to two case studies: 1) chiller fault diagnosis for heating, ventilation and air conditioning systems and 2) gearbox fault diagnosis for wind turbines in IIoT. Experimental results show the superior performance of the proposed method.