Neural Network Enabled Intermittent Fault Diagnosis Under Comparison Model

Intermittent faults are common in daily life and industrial manufacture, which have been drawing much attention from both academia and industry. In practice, intermittent faults will pose a great threat to system performance and equipment safety. Because of the randomness and unpredictability of intermittent faults, it is a great challenge to diagnose them. The fault diagnosis strategy under system-level diagnostic model plays a very important role in measuring the endogenous network security without prior knowledge, which can significantly enhance the self-diagnosing capability of network. However, as the networks become large-scale and complicated, the fault diagnosis using full syndromes from a system-level diagnostic model seems to reach its bottleneck. In this article, we first determine that the intermittent fault diagnosability of a general r-regular network G under comparison model is (t^{\text{Intermittent}}(G))^{M}=r-2. This results can be directly applied to 18 well-known networks. Then, we propose a reliable neural network enabled intermittent fault diagnosis algorithm RNNIFDCom to solve the problem of fault identification with partial syndromes for a general r-regular network G under comparison model. Finally, we implement our proposed algorithm RNNIFDCom in different networks and analyze its performance under different number of faulty nodes in terms of true positive rate, true negative rate, false positive rate, and false negative rate. The experimental results verify the theoretical results and show the advantage of our proposed algorithm RNNIFDCom.