The present work is undertaken to optimize the geometry of the curved trapezoidal winglet mounted over each cooling tube to enhance the overall performance of the heat exchanger. Before conducting the optimization study, the most critical geometric entities have been identified, and this has helped in significantly reducing the computational overheads. It has been found that the three most critical design parameters are the arc radius (R), the angle subtended (θ) and the winglet's leading-edge height (h₁), and the same are chosen for optimization. To plan the experiment, we make use of the Latin hypercube sampling (LHS) scheme. The system responses viz. the Colburn factor (j) and the friction factor (f) ‒ for different combinations of the independent variables ‒ are computed using the numerical simulations together with the conjugate heat transfer approach. To get the best outcome from computations, both fluid and solid domains have been discretized using hexahedra. We make use of the Reynolds-averaged Navier-Stokes (RANS) approach ‒ with SST k–ω (a 2-equation) turbulence model as a closure to RANS ‒ for evaluating the dependent variables. The artificial neural network (ANN) is trained using the data from the DoE table to build the metamodel necessary for conducting the multiobjective optimization. The genetic algorithm (GA) is employed to generate the optimized data sets. The study reveals that the optimized HE variants outperform the baseline model not only at the on-design condition but also at the off-design conditions. The resulting Pareto front points reveal very interesting results, and these data would facilitate the designers to make choices over a wide range of the VG geometry of the heat exchanger.

进行当前工作以优化安装在每个冷却管上的弯曲梯形小翼的几何形状，以增强热交换器的整体性能。在进行优化研究之前，已经确定了最关键的几何实体，这有助于显着减少计算开销。已经发现，三个最关键的设计参数是圆弧半径（R），对角（θ）和小翼的前缘高度（h ₁），并且选择了相同的参数进行优化。为了计划实验，我们使用拉丁超立方体采样（LHS）方案。系统响应即。科尔本因数（j）和摩擦因数（f）‒是使用数值模拟与共轭传热方法一起计算出的自变量different的不同组合的‒。为了从计算中获得最佳结果，已使用六面体离散化了流体域和固体域。我们利用雷诺平均Navier-Stokes（RANS）方法S与SST k –ω（2方程）湍流模型作为RANS closure的闭环来评估因变量。使用来自DoE表的数据对人工神经网络（ANN）进行训练，以构建进行多目标优化所必需的元模型。遗传算法（GA）用于生成优化的数据集。研究表明，优化的HE变体不仅在设计时条件下，而且在设计外条件下都优于基准模型。所得的帕累托前沿显示出非常有趣的结果，这些数据将有助于设计人员在热交换器的VG几何形状的广泛范围内进行选择。