A Hybrid Fault Diagnosis Framework Based on Test Stimulus Optimization and MPCNN for Analog Circuits

Hybrid fault diagnosis in analog circuits remains a challenging task, particularly due to the overlapping characteristics of different component faults, which often degrade the discriminative power of conventional diagnostic features. To overcome this limitation, we propose a novel diagnosis framework that integrates optimized test stimulus generation with a multiperiodic convolutional neural network (MPCNN) for accurate and robust fault identification. In the first stage, a bio-inspired optimization algorithm selects test stimuli that amplify the differences between normal and faulty circuit responses in both amplitude–frequency and phase–frequency domains. In the second stage, the MPCNN adaptively learns discriminative temporal–spatial features from the optimized responses, while a calibrated cross-entropy (CE) loss function improves classification confidence. The proposed approach is evaluated on multiple benchmark analog circuits, where the diagnostic accuracy rates reached 99.62%, 98.27%, and 91.46%, respectively, on the three standard circuits. The experimental results reveal statistically significant improvements in diagnostic accuracy, enhanced noise robustness, and superior generalization performance compared to state-of-the-art methods.