Artificial neural network prediction and multi-objective genetic algorithm optimization of the microchannel heat sink with trapezoidal ribs

• Published in: Thermal science and engineering progress Vol. 50; p. 102546
• Main Authors: Lu, Kaijie, Wang, Chunju, He, Haidong, Fan, Xueliang, Chen, Feng, Qi, Fei
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.05.2024
• Subjects: Artificial neural networks; Heat transfer enhancement; Multi-objective optimization; Sobol sensitivity analysis; Trapezoidal rib microchannel heat sink; Heat transfer enhancement; Sobol sensitivity analysis; Multi-objective optimization; Trapezoidal rib microchannel heat sink; Artificial neural networks
• ISSN: 2451-9049
• DOI: 10.1016/j.tsep.2024.102546

•The performance of 500 microchannel heat sinks with different trapezoidal rib structures is simulated.•The effects of the design parameters are assessed by single-factor analysis and Sobol sensitivity analysis.•Neural network prediction combined with a multi-objective genetic algorithm is used to optimize the design parameters.•The TOPSIS method based on entropy weights is used to compute the best compromise solution. Trapezoidal rib microchannel heat sinks (TR-MCHS) are widely used for efficient heat dissipation in electronic integrated devices. To enhance the heat transfer and hydraulic performance of TR-MCHS, an optimization method using neural network prediction combined with multi-objective genetic algorithm NSGA-II is proposed. First, the effects of trapezoidal rib geometry parameters and inlet velocity (V) on the Nusselt number (Nu¯) and pressure drop (ΔP) of TR-MCHS are investigated using numerical simulation and Sobol sensitivity analysis. The results show that the trapezoidal rib height (ht) and inlet V have the most significant effect on the TR-MCHS, and the interaction effect between the parameters can be neglected. Then, optimization is performed with Nu¯ and ΔP as objectives. On the obtained Pareto solution set, Nu¯ can be improved by 7.3 % compared to the unoptimized TR-MCHS without consuming additional pumping power. In addition, if the same Nu¯ is obtained as the unoptimized TR-MCHS, the ΔP can be reduced to 223 Pa. On this basis, the TOPSIS method based on entropy weight is developed to solve the optimal compromise solution. The best compromise solution has the Nu¯ of 50.9 and the ΔP of only 184.6 Pa, which is 87.9 Pa lower than that of the unoptimized TR-MCHS. These results have important implications for the design of TR-MCHS for electronics cooling.