中文摘要：

乘波体是高超声速飞行器的主要组成部分,也是飞行器产生升力的主要部分.针对基于计算流体动力学（CFD）分析的乘波体优化设计问题,引入人工神经元网络响应面方法.选取一定数量的乘波体外形,进行气动性能分析后,利用乘波体的外形控制参数和气动参数做为训练样本对乘波体进行训练.利用这些训练样本对人工神经网络进行训练.在优化计算中以充分训练的神经网络替代CFD分析,发展了一种基于神经网络技术的乘波体优化设计方法.利用该方法在马赫数6、雷诺数7×10~6条件下,分别对乘波体进行了最大升阻比的单目标和综合考虑升阻比、容积及表面积的多目标优化.计算结果表明,采用神经网络响应面技术可在保证计算稳定性的条件下有效提高计算效率.

英文摘要：

Waveriders are supersonic or hypersonic lifting configurations. They are extensively utilized as the forebody part of hypersonic vehicles. As the core component to generate the lift and compress the incoming flow, a waverider should be designed for assuring the high performance of a vehicle. Various optimization works had been carried out to improve the aerodynamic performance： However, most of the optimization procedures are often time consuming and unstable when the computational fluid dynamic （CFD） analysis is employed for directly evaluating aerodynamic performance. To aim at this problem, an artificial neural networks （ANN） based response surface method was proposed. First of all, a number of waverider shapes are chose as the net-training samples, and the aerodynamic performance of each sample is evaluated by CFD analysis. Next, with respect to the training couple, the control parameters of each waverider and its aerodynamic coefficients are provided to a pre-constructed ANN. The weight of each connection in the ANN is adjusted until the error between the ANN output and the CFD result are acceptable for every training couple. Finally, the ANN is embedded in the optimization loop as the response surface of the time consuming CFD procedure. Two numerical cases in the design point of Mach 6 and Reynolds number 7×10^6 are carried out to validate the presented method, a single-objective optimization for maximize the lift-to-drag ratio （L/D）, and a multi-objective problem to improve the integrated performance of a waverider with the maximal L/D, the maximal cubage, and the minimal wet area. The numerical results show that the ANN based response surface method is stable with lower time consuming.