The Digital Twin Concept in Power Electronics for Parameters Identification and Health Monitoring

The increasing integration of power electronic converters in modern power systems, such as renewable energy generation and electric vehicles, highlights the critical need for reliable operation and efficient maintenance strategies. In this scenario, the idea of monitoring the main components constituting a power electronic converter in order to evaluate their health status started to arise. This thesis focuses on developing a robust methodology for real-time parameter identification in an indirect way to not introduce additional hardware in the physical asset. To meet this aim, the digital twin technology has been investigated. The proposed approach combines digital twin models with advanced optimization algorithms to estimate key parameters such as capacitance, inductance, and switches on-state resistance that are indicative of the operational health of the converter's components. By integrating real-time sensor data, the digital twin is dynamically updated, enabling continuous monitoring and, therefore, a suitable predictive maintenance strategy. This research activity has begun by investigating topics already widely addressed in the literature, such as digital twins of DC-DC converters. In this way, the methodology has been acquired and drawbacks and difficulties have been found. Subsequently, this knowledge has been transferred to converter topologies less present in the literature regarding the parameters’ identification through the digital twin method, such as three-phase AC-DC converters. In this context, a study of the digital model, the objective function and a comparison of the best optimization algorithms to realize the digital twin of a three-phase AC-DC converter are presented. The experimental results are shown to validate the investigated method. Furthermore, a discussion is carried out relating to the difficulties and limitations of the proposed method. Finally, the development of a recurrent neural network for the estimation of the degradation parameters of a power electronic converter is proposed. Comparing the results obtained with the latter method to those obtained with the digital twin technique, the advantages from the point of view of accuracy are immediately clear. For this reason, a study regarding the feasibility and real-time applicability of complex algorithms, such as neural networks on simple digital signal processors, is then carried out in this work. Hence, the research aims to bridge the gap between parameter identification performed with the digital twin technique and neural networks for more complex conversion systems, such as three-phase converters. Moreover, this thesis contributes to enhance neural network applicability for the identification of the deterioration parameters in power electronic converters.