Digital Twin and AI-Based Fault Diagnosis of Power Electronic Converters for Smart Grid Applications

Abstract

Power electronic converters form the backbone of renewable energy integration, electric mobility, energy storage and flexible smart-grid operation; however, converter switching devices, sensors, capacitors, and control circuits remain vulnerable to electrical and thermal faults. This paper proposes a framework for monitoring a broad range of converter operating states continuously using new digital twin (DT) and artificial intelligence (AI) technology to detect faults. Measured voltage, current, and control inputs are fed into the DT to reproduce converter dynamics and residual signals are generated based on a comparison of physical outputs with model-predicted responses. Abnormal state detection and classification of healthy operation, two open-circuit conditions in the switches, current-sensor failure and direct current (DC)-link capacitor degradation are performed using an AI-based classifier. The proposed framework integrates data acquisition, prediction and residual processing based on physics-informed concepts, domain feature extraction and automated diagnostic action directly within a unified loop of monitoring. Measured and predicted DC-link voltages are in close agreement during normal operation, followed by a clear residual increase after the initiation of fault, between points A and B. Root-mean-square residual signatures are well-separated across operating classes, which helps maintain excellent discrimination. The confusion matrix shows very high class-wise recognition, and then the benchmarks show better diagnostic performance, response time, and instability. This approach covers a wide range of techniques, leading from data to inferences that form the basis for predictive maintenance. This approach decreases the stability. The stability of the system is particularly important, especially in renewable energy and smart grid applications, as converters are not expected to operate under disruptions.