Advanced Fault Diagnosis Method for DC-DC Converters: Leveraging the Temporal Continuity of Electrical Signals

Abstract

This article focuses on the crucial role of reliable dc–dc converter operation for the stability of modern power electronic devices. Addressed is a common issue in the fault diagnosis of dc–dc converters: the tendency to rely on local feature fitting while the temporal continuity of electrical signals is neglected. An innovative diagnostic method that utilizes an adaptive wavelet transform from a data processing perspective is proposed. This technique can dynamically adjust the scale and translation parameters to adapt to the continuous changes in electrical signals caused by varying circuit conditions. From the standpoint of model improvement, the extended convolutional capsule network model is designed. Through multiscale feature extraction, integration of global-local attention mechanisms, and global vector analysis, this model effectively diagnoses fault features. It is demonstrated that our method is effective in extracting the time-continuity features of electrical signals, and exhibits significant advantages in diagnostic accuracy, performance metrics, and application generalization capability. Consequently, this study presents a holistic and effective approach for fault diagnosis in dc–dc converters.