Multiobjective Robust Optimisation Design for Nanocrystalline High‐Frequency Transformer Based on Multi‐Physical Field Considering Core Uncertainty

ABSTRACT High‐frequency transformers (HFTs) are critical in modern power electronics, especially for application scenarios requiring compact size, high efficiency and superior thermal stability. However, their performance is often constrained by manufacturing tolerance and material property variations. This paper presents a robust optimisation design ( Rob. D ) method for nanocrystalline HFTs ( nano HFTs) that considers the material uncertainty, focusing on optimising power density, leakage inductance and thermal stability simultaneously. The support vector regression (SVR) is applied to the surrogate model to replace the computationally expensive finite element analysis (FEA) during the extensive uncertainty evaluations required in MCA, whereas the nondominated sorting genetic algorithm III (NSGA‐III) leverages the surrogate model to efficiently navigate the trade‐offs between competing objectives under variability. The Pareto‐optimal solutions achieve a 1.26°C decrease in hotspot temperature of nano HFT. The validation via simulations and a 20 kVA prototype confirms that Rob. D reduces the standard deviations of the core uncertainty to 2.7 × 10 −3 °C, substantiating the framework's efficacy in balancing competing objectives under uncertainty. Compared to conventional deterministic optimisation design ( Det. D ) approaches, the proposed method demonstrates enhanced robustness by treating core dimensions and magnetic properties as statistically distributed variables. This enables optimisation of both mean performance and standard deviation of objectives, ensuring resilience against the actual manufacturing dispersions.