基于响应面法的再生冷却通道尺寸传热优化

Thermal Optimization for Regenerative-Cooled Channel Dimension Based on Response Surface Methodology

关键词:  再生冷却  传热分析  Box-Behnken试验设计  响应面  优化

Key words:  Regenerative-cooling  Thermal analysis  Box-Behnken design  Response surface  Optimization

摘要:

为了降低液体火箭发动机推力室壁温和冷却剂压力损失，对再生冷却通道尺寸参数进行优化设计。以再生冷却通道高度、宽度、数目和推力室内壁厚为设计变量，推力室平均壁温、最高壁温和冷却剂压力损失为目标函数，采用Box-Behnken试验设计方法获取样本点，根据样本点建立再生冷却通道计算模型，利用传热分析程序针对不同方案得到目标函数关于设计变量的二阶响应面模型，分别用梯度投影、积极集法和遗传算法进行优化计算，同时利用逐步回归法和样本点更新技术提高模型精度。计算结果表明，建立的响应面模型能以较小的计算成本准确地反映设计变量和目标函数的关系；存在一个最佳的通道高宽比和通道数目使得冷却通道传热特性最优；对于两种不同优化方案，优化设计后的目标函数最多比初始设计降低13.5%和23.5%；使用遗传算法优化后得到的目标函数值最低。

Abstract:

To lower the wall temperature of liquid rocket thruster chamber and reduce the cooling channel pressure loss，the optimization design was carried out for the parameters of the regenerative-cooled channel. The height，width，channel number of the regenerative-cooled channel and the inner wall thickness of the thrust chamber were determined as design variables; the average and the highest wall temperature of thruster chamber and the coolant pressure loss were chosen as objective functions; Box-Behnken design was adopted to get sample points that can be used to establish regenerative cooling channel model and then the objective function of the second order response surface of design variable was obtained by analyzing heat transfer program and making various schemes. The optimal results were calculated by gradient projection method，active set method and genetic algorithm. Meanwhile，the model accuracy was improved by using stepwise regression and renewing sample points at each step optimization. Calculation results show that the response surface model established in this paper precisely reflect the relationship between design variables and objective functions at lower calculation cost; there are a best aspect ratio and the number of channel that can make the optimal heat transfer characteristics in cooling channel. For two different kinds of design schemes，the objective function values at most can be reduced by 13.5% and 23.5% as compared with the initial design; the values of objective functions calculated by genetic algorithm are lower than those of gradient projection method and active set method.