Multi-Objective Optimization of Primary Loop System for Nuclear Reactor Coolant Pump Test Bench

Abstract

To address the issues of pressure loss and nonuniform flow velocity distribution in the primary loop system of a nuclear reactor coolant pump (RCP) test bench, an integrated optimization framework is proposed that combines computational fluid dynamics (CFD), Kriging surrogate modeling, explainable machine learning, and an improved multiobjective genetic algorithm. CFD simulations coupled with optimal Latin hypercube sampling are employed to construct the experimental design, based on which Kriging surrogate models are established. The SHAP (SHapley Additive exPlanations) method is introduced to reveal the underlying influence mechanisms of design parameters on flow performance. Subsequently, an improved Non-dominated Sorting Genetic Algorithm II (NSGA-II) is applied to conduct multi-objective optimization of key pipeline dimensions. The results demonstrate that the optimized primary loop system achieves a 3.51% reduction in pressure loss and an 8.59% improvement in flow velocity uniformity, thereby significantly enhancing the overall flow field quality. This research provides an explainable and verifiable optimization pathway for the efficient design and engineering decision-making of nuclear reactor coolant pump test benches, and has been successfully implemented in industrial practice.