A Novel Fault Diagnosis Method for Multistage Conversion Circuits Based on Data Fusion

Abstract

This article addresses the research gap on fault diagnosis of multistage conversion circuits within analog circuit fault diagnosis. A diagnostic system is introduced, in which multipoint data fusion is combined with deep feature analysis, leading to the integrated dual-axis vision transformer system. Initially, signals from multiple monitoring points are fused through the integrated wavelet transform algorithm. Following this, deeper secondary data fusion is achieved by the dual-axis vision transformer algorithm, which utilizes a dual-axis observation encoder and an axial data decoder to interact between time-domain and frequency-domain features. This approach effectively analyzes signal characteristics, improving the accuracy of fault diagnosis. In experiments with the LLC series resonant converter, both soft and hard faults were reliably diagnosed by the system, showing excellent accuracy, recall, and F1 score metrics.