首页|基于多种优化方法的轴流风扇叶型气动优化

基于多种优化方法的轴流风扇叶型气动优化

扫码查看
为了解决轴流风扇叶型W65优化设计的问题,选取高斯过程回归、人工神经网络和序列二次规划优化方法对叶型进行优化.首先,通过类别形状函数变换方法表示叶型,并生成一定区间内的叶型样本集,使用B样条曲线进行光滑化处理,通过CFD模拟仿真方法获得光滑叶型升阻比数据.然后,分别采用高斯过程回归方法、人工神经网络方法和序列二次规划方法,对带有面积约束的多攻角升阻比目标函数进行寻优优化.其中,前两种优化方法分别结合遗传算法和梯度下降法,序列二次规划方法未结合其他优化方法.研究工况为攻角0°~8°,马赫数为0.5.将优化后的叶型通过CFD方法进行验证,结果表明:通过3种方法获得的优化叶型综合升阻比分别提高了8.41%,8.49%和2.08%,优化方法中预估的相对误差分别为0.25%,-0.39%和6.31%,高斯过程回归方法和人工神经网络方法优化误差较小,而序列二次规划方法的优化误差较大.
Aerodynamic Optimization of Axial Fan Blade Airfoil based on Multiple Optimization Methods
In order to solve the problem of optimizing the design of axial fan blade airfoil W65,Gaussian process regression,artificial neural network and sequential quadratic programming were selected to opti-mize the blade airfoil.Firstly,the class function/shape function transformation(CST)method was used to represent the blade airfoil and generate a sample set of blade airfoils within a certain interval.B-spline curves were used for smoothing treatment,and the smooth blade airfoil lift-drag ratio data were obtained through computational fluid dynamics(CFD)simulation.Then,the Gaussian process regression,artifi-cial neural network and sequential quadratic programming methods were respectively used to optimize the objective function of multi angle of attack lift-drag ratios with area constraints.The first two optimization methods were combined with genetic algorithm and gradient descent method,while the sequential quadratic programming method did not combine other optimization methods.Under the working condition of the angle of attack varying from 0° to 8°,the Mach number was 0.5.The optimized blade airfoil was validated using CFD method.The results show that the comprehensive lift-drag ratio of the optimized blade airfoil obtained through three methods increase by 8.41%,8.49%and 2.08%,respectively.The estimated relative errors in the optimization methods are 0.25%,-0.39%,and 6.31%,respectively.The Gaussian process re-gression method and artificial neural network method have smaller optimization errors,while the sequential quadratic programming method has larger optimization errors.

airfoil optimizationmachine learningGaussian process regressionartificial neural net-worksequential quadratic programming

陈晨铭、郭雪岩、李春

展开 >

上海理工大学能源与动力工程学院,上海 200093

叶型优化 机器学习 高斯过程回归 人工神经网络 序列二次规划

2024

热能动力工程
中国 哈尔滨 第七0三研究所

热能动力工程

CSTPCD北大核心
影响因子:0.345
ISSN:1001-2060
年,卷(期):2024.39(9)