Low-cost Digital Twin Design for Power Electronics using Deep Neural Networks

Deep learning-based digital twin (DT) models have been used in many real-time applications in the literature showing a superior behavior to traditional models. In particular, power electronics (PE) industrial applications demand a very fast response with little room for error. The superiority of such deep learning models comes at the cost of increased complexity in time and space; thus, higher hardware requirements and higher cost. This paper shows detailed guideline on the methodology of building DT models using Deep neural networks (DNNs) for PE applications (PEDTD) using low-cost microcontrollers. PEDTD models are analyzed in this paper with respect to their space and time empirical complexity while showing the limitations of the models based on the available hardware specifications. A use case scenario of switch fault localization and detection in T-type three level rectifiers is presented showing that the proposed technique is applicable in practical applications.