An intermittent fault diagnosis method of analog circuits based on variational modal decomposition and adaptive dynamic density peak clustering

Analog circuits are widely used in industrial systems and avionics. Intermittent faults (IFs) as a special type of fault in circuits are difficult to diagnose. Due to the short duration of IFs, it is first necessary to obtain IF samples from the original signal. Therefore, variational modal decomposition (VMD) and autoencoder are proposed to capture the appearing and disappearing moments of IFs. Then, IF samples can be extracted from the original signal by the detected moments of appearance and disappearance. Finally, the adaptive dynamic density peak clustering (ADDPC) method is proposed for automatically identifying IF categories. The dynamic nature of ADDPC is reflected in the fact that the density kernel is not a fixed scanning radius, but a dynamic radius density kernel based on the k nearest neighbor. The adaptability is reflected in the fact that the parameters of ADDPC can be selected automatically by particle swarm optimization according to the change of data distribution. The intelligent diagnosis method is subsequently applied to a typical analog filter circuit under three noise levels. The results show that the proposed framework has a better diagnostic performance in the presence of noise.