A novel virtual sensor approach to estimate junctions temperature of power electronics components for electrical/hybrid vehicles.

As modern technology advances, the demand for compact and complex electronic components has increased, bringing new challenges in managing heat generation within integrated circuits (ICs). Effective thermal management has become essential, as excess heat can impair performance and reduce the operational lifespan of components through thermal wear. In high-stakes applications like electric vehicle (EV) inverters, accurate temperature estimation is vital for efficiency and reliability. This thesis addresses these needs by developing a mathematical thermal model to estimate the junction temperature of MOSFET transistors in EV inverters, utilizing both simplified and complex modeling methodologies with a case study provided by Ideas & Motion Torino. This thesis comprises two main units, each exploring a different method for creating a model capable of estimating the junction temperature of a PCB. Unit 1 introduces a simplified lumped parameter model (Cauer), chosen for its computational efficiency and suitability for real-time applications. This model, represented in a Linear Time-Invariant (LTI) system framework, approximates the thermal behavior by discretizing the system into manageable components, facilitating simulations in both continuous and discrete-time domains via MATLAB and Simulink. Using temperature data from an onboard NTC thermistor within the feedback loop, the model enables recalibration for greater prediction accuracy. Despite structural assumptions, validation against reference data underscores the model’s flexibility and adaptability, enabling reliable design-specific calibration through curve-matching techniques to address any data inconsistencies. In Unit 2, the study extends the model's predictive accuracy by constructing a more detailed assumed model with the help of bond graph representation. This approach captures the multi-domain energy dynamics across the system and is particularly suited for integrating complex thermal and electrical interactions within the inverter. The refined model includes matrix building and parameter adjustments, enhancing its capability to reflect the system’s thermal behavior under varying load conditions. Through MATLAB simulations and iterative modifications, the model is tested and then finally compared to the one done in Unit 1. This thesis highlights the value of grey-box modeling, which blends theoretical and empirical data for balanced and effective thermal estimation. The lumped parameter approach reduces model complexity, while the bond graph method enables multi-domain analysis for broader applications. The insights gained underscore the developed model’s utility in EV inverter systems and its potential to optimize thermal stability and component lifespan in a range of industries, including automotive and aerospace. This study’s findings emphasize the importance of accurate reference data and highlight areas for future research, including model refinement to enhance granularity and extending applications to other semiconductor devices like IGBTs and SiC-based transistors. Ultimately, the proposed thermal model offers a robust approach to real-time control strategies, supporting next-generation thermal management in advanced electronics.