Data-efficient multi-objective structural optimization and design space exploration of refrigerant-based cold plate for laser device

Abstract

Fiber lasers are widely used in advanced manufacturing and aerospace due to their high beam quality, compactness, and operational stability. However, as laser power density and integration level continue to increase, thermal management becomes a key factor limiting performance and service life. This study proposes a data-efficient design framework for a refrigerant-based cold plate applied to fiber-laser cooling. A multi-objective optimization strategy is developed by combining a multi-fidelity surrogate with Bayesian optimization, where pressure drop and mean absolute temperature deviation are treated as conflicting objectives. In addition, Shapley additive explanations analysis is introduced to interpret the influence of structural parameters on thermo-hydraulic performance. The results show that the multi-fidelity surrogate model improves predictive accuracy by 29%–43% compared with single-fidelity surrogate model while significantly reducing the number of expensive high-fidelity simulations. The optimized design achieves a substantial reduction in flow resistance with only a limited thermal-performance penalty. Experimental results further show that, under a heat load of 2800 W, the optimized cold plate reduces flow resistance by 42.5%, while the average surface temperature in the critical region increases by only 1.7 °C relative to the reference design. The proposed framework provides an effective and interpretable solution for the structural design of compact, high-performance refrigerant-cooled cold plates for high-power laser devices.