Efficient Experimental Design Framework for Reliable Parameter Identification of Electro-Thermal Models

Electro-thermal models commonly rely on experimentally identified parameters to predict temperatures in power electronic systems. However, even with a physically reasonable model structure, arbitrarily selected experimental cases may leave some parameters unidentifiable and degrade prediction capability. This paper proposes an experimental design framework with an information-sufficiency criterion linking prescribed parameter variations to measurable temperature responses. Singular value decomposition (SVD) of the parameter-sensitivity matrix identifies weak parameter directions, whose corresponding temperature responses are compared with a practical detectability threshold. The framework is demonstrated using a nonlinear electro-thermal model with thermal coupling and validated on four independently power-controlled MOSFETs representing multiple devices. Unguided and frameworkguided selections yield fitting errors of 2.161 °C and 1.413 °C, but prediction errors of 28.236 °C and 0.462 °C at the same validation case. A broader analysis over experimental case selections also shows an increase in prediction error as temperature response falls below the proposed criterion. These results show that the proposed framework can guide experimental case selection to improve electro-thermal model prediction reliability.