NURBS-Based Surrogate Model for Fast and High-Fidelity Predictions of Pressure Drop and Pin-Fins Temperature in Traction Inverter

The growing use of wide-band-gap semiconductors in power electronics increases the demand for efficient yet low-cost thermal management. Pin-fin heat sinks offer a practical solution due to their robustness and ease of fabrication, but their geometric optimization via Computational Fluid Dynamics (CFD) models remains computationally demanding, particularly in multi-variable contexts. This work presents a novel surrogate modeling approach based on Non-Uniform Rational Basis Splines (NURBS) entities for predicting thermal and hydraulic performance in pin-fin cooling systems for traction inverters. Two CFD modeling strategies were explored: one allowing variable pin count, and another preserving topological consistency by adjusting the domain width. While the former introduced geometric discontinuities that degraded metamodel accuracy, the latter enabled the NURBS-based surrogate model to achieve high predictive performance with a few seconds evaluation times. Compared to Kriging surrogate models and Artificial Neural Networks (ANNs), the NURBS approach delivers superior computational efficiency, making it particularly well-suited for real-time applications, sensitivity analyses, and gradient-based optimization.