AI-Based Fault Diagnosis of Electrical Equipment Using Quantum-Inspired Dynamic VMD and a Lightweight CNN

Condition monitoring and fault diagnosis of electrical equipment are essential for ensuring the reliability and safety of modern energy and industrial systems. This paper proposes an artificial intelligence-based method for equipment fault diagnosis by integrating Quantum-Inspired Dynamic Variational Mode Decomposition (QD-VMD) with a lightweight convolutional neural network (L-CNN). QD-VMD adaptively determines the number of modes (K) and the penalty factor (α) through a quantum-state probability model with phase-guided updates, thereby enabling stable decomposition of nonstationary signals under noisy conditions. Statistical features derived from the decomposed modes in both time and frequency domains are combined into multi-scale feature tensors, which are then classified by an L-CNN employing depthwise separable convolutions and channel attention. The proposed method achieves 99.95% accuracy on the CWRU bearing dataset with approximately 55,000 parameters and 1.04 × 10⁻³ GFLOPs, while maintaining an average inference latency of about 2.5 ms on embedded hardware, supporting real-time edge deployment. Cross-domain evaluations on datasets from power systems and new energy vehicles yield precision, recall, and F1-scores all greater than or equal to 0.987, demonstrating strong robustness across heterogeneous operating conditions. Reconstruction experiments further show that QD-VMD preserves critical fault-related structures with a low global mean squared error (MSE). These results confirm that combining robust signal decomposition with compact learning architectures provides an effective AI-based solution for fault diagnosis of electrical equipment, with potential applicability to insulation monitoring and related reliability tasks.