Early fault diagnosis for DC/DC converters based on digital twins and transfer learning

Abstract

DC/DC converters are widely used as power supplies in various power supply systems. Their faults can lead to improper system operation. Building intelligent fault diagnosis models for ‘highly reliable and long-life’ DC/DC converters is challenging due to the high data acquisition costs and insufficient fault data. In this paper, the digital twin (DT) technique is utilized to provide extensive and reliable data to address the issue of insufficient data for diagnosing DC/DC converter faults. However, establishing multidisciplinary, multi-physical quantity, multi-scale, and multi-probability virtual models (VMs) can be challenging. Thus, a cloud model-based interpretable transfer model is designed to overcome the limitations of VMs in fully deducing the full-state output properties. The proposed method effectively diagnoses DC/DC converters based on DT technology by providing a more comprehensive and uniformly distributed source domain data to suit the samples in the target domain. Comparison with other algorithms shows the effectiveness of the proposed method. It has the potential to provide accurate and real-time diagnosis for DC/DC converter faults and enable timely maintenance strategies for power monitoring in equipment, such as submarine observation networks and space power supply systems.