Circuit fault diagnosis method based on NEITD-IDBO-VME and hybrid neural network

Abstract To address the challenges in modern circuit system fault diagnosis, including strong signal nonlinearity, significant noise interference, and poor stability, an intelligent diagnostic method based on noise-enhanced intrinsic time-scale decomposition (NEITD)-improved dung beetle optimizer (IDBO)-variational mode extraction (VME) and a hybrid neural network is proposed. The methodology comprises a dual-stage signal processing framework: first, the NEITD algorithm preprocesses raw fault signals through adaptive auxiliary signal enhancement and geometric filtering to improve noise immunity; second, the IDBO algorithm optimizes VME parameters to achieve adaptive decomposition of multiple IMF components. A multidimensional feature set integrating time-domain, frequency-domain, and nonlinear characteristics is subsequently constructed. Finally, a spiking-probabilistic hybrid neural network with attention mechanism enables high-precision fault classification. Experimental results demonstrate a diagnostic accuracy of 99.7%, significantly outperforming conventional methods and providing an effective solution for complex circuit system fault diagnosis.