Digital Twin-Based Predictive Maintenance of Power Converters

Abstract

Power Electronics are considered the most important element in most of the modern applications which include the Renewable Energy sources like solar, wind, tidal and in Electric vehicles and industrial automation processes. Although it has great importance in these modern applications, the reliability of these power electronics converters is at very high risk due to the ageing and degradation of the critical components, such as IGBT/MOSFET switches and capacitors. Traditional Maintenance practices, which are time-based or reactive maintenance available in the market, are not able to predict the failure in advance, which results in an increase of downtime and poses operational risk to the operator and reliability is affected. To overcome the limitation faced by the traditional maintenance practices, this dissertation work introduces the Digital Twin-based predictive maintenance, specially employed for power converters to predict the health status and remaining useful life of critical components of power converters, which cause failures in power converters. In this research work, first, we discussed the role of power electronics converters in modern applications and the challenges faced that affect the reliability of power converters. Then, to overcome these challenges, how a digital twin-based predictive maintenance strategy is employed, its concept, architecture and its applications in other areas are discussed. Afterwards, the literature review about the applications to digital twin-based predictive maintenance of power converters is thoroughly studied and analyzed. Further, the literature gap is identified in the literature analyzed, which results in the identification of the problem. To bridge this literature gap, this dissertation work is carried out to provide a feasible solution to the identified problems, which bridges the literature gap in this maintenance strategy. The proposed methodology in this work generates synthetic run-to-failure data of the degradation component in the power converters and employs machine learning algorithms to identify the anomaly and predict the RUL of components. This work documents two case studies, firstly, the ageing severity of the power converter IGBT switch is identified with the classification models by generating synthetic data of ageing with the help of MATLAB, and secondly, the remaining useful life of the capacitor is predicted through a data-driven degradation model by generating degradation data in MATLAB. MATLAB simulations are utilized to develop and validate a predictive maintenance algorithm aimed at detecting ageing severity and estimating the remaining useful life (RUL) of a capacitor. The results show the effectiveness of the proposed methodology to predict the health status and RUL of power converter components and how it ensured reliability and extended the lifespan of converters. Future work will focus on integrating additional features like real-time data integration and many other technologies to enhance model accuracy and practical deployment.