XGBoost–Bayesian Optimization Method for Multi-Blade Centrifugal Fan

To overcome the trade-off between airflow rate and efficiency in a multi-blade centrifugal fan for automotive air-conditioning systems, as well as the low optimization efficiency of conventional simulation- or experiment-based approaches resulting from the large number of impeller design parameters, a surrogate-model-based optimization framework coupled with an intelligent optimization algorithm is proposed for impeller design. Latin hypercube sampling was employed to generate 120 samples within the design space, and an XGBoost model was integrated with a Bayesian optimization strategy to perform the optimization process. Validation results indicate that the prediction error of the surrogate model is less than 3%. Under the final optimized impeller configuration, the airflow rate is increased by 22% at the design operating condition, while the total pressure efficiency is improved by approximately 20%, satisfying the prescribed design requirements.