EPE Association: EPE 2023 - LS6e: Data Analysis, Artificial Intelligence and Communication

EPE 2023 - LS6e: Data Analysis, Artificial Intelligence and Communication
You are here: EPE Documents > 01 - EPE & EPE ECCE Conference Proceedings > EPE 2023 ECCE Europe - Conference > EPE 2023 - Topic 10: Data Analysis, Artificial Intelligence and Communication > EPE 2023 - LS6e: Data Analysis, Artificial Intelligence and Communication

   [return to parent folder]

   A Data-Driven Thermal Digital Twin of a 3-Phase Inverter using Hi-Fidelity Multi-Physics Modelling
By Sachin Kumar BHOI, Mohamed Amine FRIKHA, Gamze Egin MARTIN, Farzad HOSSEINABADI, Sajib CHAKRABORTY, Mohamed EL BAGHDADI, Omar HEGAZY [View]
[Download]

Abstract: The assessment of Power Electronics Converters (PECs) reliability is a crucial research area forthe development of robust PECs. The key parameter for reliability assessment is the real-time junction temperature (Tj ) profile for semiconductor power switches of PECs. While complex high-fidelity simulation models with electro-thermal models can accurately predict Tj during the offline design phase, their computational requirements make them impractical for real-time applications. This paper proposes a methodology to develop a real-time hardware deployable model for estimating the junction temperature of an Insulated Gate Bipolar Transistor (IGBT) power device. A data-driven reduced-order model(ROM) of an industrial inverter setup based on a highfidelity multiphysics simulation model is presented in the article. This work also contributes to the realization ofvirtual sensors and digital twins for PECs.

   Concept Validation of Digital Twin-Based Power Losses Estimation Method for Traction Inverter Applications
By Mohamed Amine FRIKHA, Sachin Kumar BHOI, Sajib CHAKRABORTY, Niels DIVENS, Reginald DILTOER, Mohamed EL BAGHDADI, Omar HEGAZY [View]
[Download]

Abstract: This paper presents a digital twin concept to accurately estimate the power losses of a traction inverter operating under different dynamic driving cycles by utilizing analytical models. The paper illustrates the different stages involved in developing the digital twin, including data collection, optimisation techniques and training. A physical system is prepared to validate the concept, including a back-to-back motor inverter setup, required voltage and current sensors, control and DAQ device. Finally, the results show a high degree of correlation between the digital twin and the physical system, which verifies the validity of the proposed concept.

   Machine Learning Model for High-Frequency Magnetic Loss Predictions Based on Loss Map by a Measurement Kit
By Xiaobing SHEN, Wilmar MARTINEZ [View]
[Download]

Abstract: Accurately forecasting the losses in highfrequencymagnetic materials is a significant challengewhen optimizing the design of high-frequency (HF) magneticcomponents. Existing models do not adequatelyconsider the intricate interactions among geometry, andtemperature factors, which have distinct and substantialimpacts on core losses. A new method is introduced, whichutilizes a Deep Neural Network (DNN) model to constructparameterized models for high-frequency magnetic coreloss based on measurement data. The DNN employs theGaussian Error Linear Unit (GELU) activation functionand Huber loss function, and its performance is comparedto that of a conventional Rectified Linear Unit (ReLU) activationand Mean Squared Error (MSE) loss function. Theproposed DNN demonstrates significantly higher accuracyand improved robustness.

EPE 2023 - LS6e: Data Analysis, Artificial Intelligence and Communication
You are here: EPE Documents > 01 - EPE & EPE ECCE Conference Proceedings > EPE 2023 ECCE Europe - Conference > EPE 2023 - Topic 10: Data Analysis, Artificial Intelligence and Communication > EPE 2023 - LS6e: Data Analysis, Artificial Intelligence and Communication
	[return to parent folder]

	A Data-Driven Thermal Digital Twin of a 3-Phase Inverter using Hi-Fidelity Multi-Physics Modelling By Sachin Kumar BHOI, Mohamed Amine FRIKHA, Gamze Egin MARTIN, Farzad HOSSEINABADI, Sajib CHAKRABORTY, Mohamed EL BAGHDADI, Omar HEGAZY	[View] [Download]
Abstract: The assessment of Power Electronics Converters (PECs) reliability is a crucial research area forthe development of robust PECs. The key parameter for reliability assessment is the real-time junction temperature (Tj ) profile for semiconductor power switches of PECs. While complex high-fidelity simulation models with electro-thermal models can accurately predict Tj during the offline design phase, their computational requirements make them impractical for real-time applications. This paper proposes a methodology to develop a real-time hardware deployable model for estimating the junction temperature of an Insulated Gate Bipolar Transistor (IGBT) power device. A data-driven reduced-order model(ROM) of an industrial inverter setup based on a highfidelity multiphysics simulation model is presented in the article. This work also contributes to the realization ofvirtual sensors and digital twins for PECs.

	Concept Validation of Digital Twin-Based Power Losses Estimation Method for Traction Inverter Applications By Mohamed Amine FRIKHA, Sachin Kumar BHOI, Sajib CHAKRABORTY, Niels DIVENS, Reginald DILTOER, Mohamed EL BAGHDADI, Omar HEGAZY	[View] [Download]
Abstract: This paper presents a digital twin concept to accurately estimate the power losses of a traction inverter operating under different dynamic driving cycles by utilizing analytical models. The paper illustrates the different stages involved in developing the digital twin, including data collection, optimisation techniques and training. A physical system is prepared to validate the concept, including a back-to-back motor inverter setup, required voltage and current sensors, control and DAQ device. Finally, the results show a high degree of correlation between the digital twin and the physical system, which verifies the validity of the proposed concept.

	Machine Learning Model for High-Frequency Magnetic Loss Predictions Based on Loss Map by a Measurement Kit By Xiaobing SHEN, Wilmar MARTINEZ	[View] [Download]
Abstract: Accurately forecasting the losses in highfrequencymagnetic materials is a significant challengewhen optimizing the design of high-frequency (HF) magneticcomponents. Existing models do not adequatelyconsider the intricate interactions among geometry, andtemperature factors, which have distinct and substantialimpacts on core losses. A new method is introduced, whichutilizes a Deep Neural Network (DNN) model to constructparameterized models for high-frequency magnetic coreloss based on measurement data. The DNN employs theGaussian Error Linear Unit (GELU) activation functionand Huber loss function, and its performance is comparedto that of a conventional Rectified Linear Unit (ReLU) activationand Mean Squared Error (MSE) loss function. Theproposed DNN demonstrates significantly higher accuracyand improved robustness.