Fault diagnosis methods for multi-layer edge computing systems

Abstract

Industrial Internet of Things (IIoT) advances the fault diagnosis of multi-tier edge computing systems into the realm of big data. However, a significant challenge arises in terms of privacy leakage due to the necessity for nodes in edge systems to share local private data. Additionally, a centralized architecture is susceptible to causing single-point failures in edge systems. Moreover, the fault data from nodes in IIoT edge systems are often nonindependent and nonidentically distributed (nonIID), posing difficulties for convergence. Furthermore, there is a lack of corresponding defenses to prevent malicious node fault attacks. To tackle these challenges, this work introduces a fault diagnosis framework and method for multi-tier edge computing systems based on blockchain smart contracts. The framework primarily adopts a decentralized model to ensure the privacy and security of fault data in multi-tier edge systems. Within this framework, a feature comparison loss function is designed to address the nonIID nature of faults. Additionally, a Byzantine fault-tolerant method is developed to prevent fault threats. Furthermore, we design an incentive method based on reputation to assess the rewards that nodes should receive. Experiments demonstrate that the proposed method achieves robust overall performance without compromising the privacy of local data.