Counterfactual Inference for Early Parametric Drift Fault Diagnosis in Analog Circuits

Le résumé fourni par la source

In analog circuits, early-stage faults often exhibit subtle features, making their symptoms difficult to effectively identify and distinguish, which poses significant challenges for fault recognition. To address these issues, this paper introduces counterfactual reasoning into early fault diagnosis of analog circuits, using “if-then” logical inference to explore the causal relationship between faulty components and observed fault symptoms. Signal data at the circuit output under various soft fault conditions are used as inputs to a causal Bayesian network. To reduce the complexity of the network model, we incorporate a Leaky Noisy-OR node. Additionally, a noise parameter λ is introduced to represent the uncertainty in fault propagation from component failures to signal anomalies. During the fault diagnosis process, a counterfactual diagnosis algorithm is derived based on intervention operations to compute the counterfactual probabilities of fault nodes. The faulty component is identified quantitatively based on probability inference by comparing the obtained counterfactual probabilities. The proposed method is evaluated using two experimental circuits: a Sallen-Key band-pass filter and a quad op-amp biquadratic filter circuit. Experimental results demonstrate that the proposed approach achieves higher diagnostic accuracy than existing early fault diagnosis methods for analog circuits.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.