DiffDA-Net: diffusion-augmented domain adaptive network for analog circuit fault diagnosis under imbalanced and...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Accurate analog circuit fault diagnosis is vital for electronic system reliability.
• Data-driven methods face challenges with imbalanced data and varying operating conditions.
• The combined impact of these challenges on analog circuit diagnosis is under-explored.

Purpose of the Study:

• To propose DiffDA-Net, a unified two-stage framework addressing data imbalance and cross-condition generalization in analog circuit fault diagnosis.
• To systematically benchmark generative augmentation and domain adaptation techniques in this joint setting.
• To improve the reliability and accuracy of automated fault diagnosis systems.

Main Methods:

• Utilized a conditional Denoising Diffusion Probabilistic Model for class-conditioned fault signal generation to balance datasets.
• Employed a domain adaptive network with adversarial training and Maximum Mean Discrepancy for knowledge transfer across operating conditions.
• Conducted experiments on a 13-class cross-severity transfer task (50%→25% deviation).

Main Results:

• The dual-alignment domain adaptation module achieved high target accuracy, significantly outperforming single-mechanism baselines.
• Under label-free conditions, the method doubled the accuracy compared to no adaptation.
• The full DiffDA-Net framework attained substantial accuracy in the joint imbalanced and cross-domain setting.
• Controlled experiments confirmed domain alignment as the primary performance driver, with diffusion augmentation providing further gains.

Conclusions:

• DiffDA-Net offers a unified solution for simultaneous data imbalance and domain adaptation in analog circuit fault diagnosis.
• Diffusion augmentation proved effective and competitive with other generative methods.
• The framework demonstrates significant potential for enhancing the robustness and accuracy of fault diagnosis systems in real-world applications.