摘要
为了提高内转式进气道的设计效率,实现对内收缩基准流场的快速预测,采用准均匀B样条方法实现内收缩基准流场的参数化设计,提出了基于深度学习残差神经网络架构的流场预测模型。结合峰值信噪比、结构相似性指数等图像质量评估方法,对预测流场进行定量评价,并从中提取壁面特性分布、激波形态等关键流场特性,以实现基于基准流场几何参数快速获取流场云图和特性参数分布的目标。研究结果表明,所构建的流场快速预测模型精度较高,其整体平均峰值信噪比为42.51 dB,平均结构相似性指数为0.9973,且能有效地从预测结果中提取流场的关键特性与参数分布,为内收缩基准流场的快速设计与优化提供有力支持。
To enhance the design efficiency of inward-turning inlet and enable rapid prediction of internal contraction basic flowfield,a parametric design of internal contraction basic flowfield was implemented using quasi-uniform B-spline methods,and a flow field prediction model based on deep learning residual neural network architecture was proposed.The predicted flowfields were quantitatively evaluated using image quality assessment methods including PSNR(peak signal-to-noise ratio)and SSIM(structural similarity index),from which key flow field characteristics such as wall property distributions and shock wave shape were extracted to achieve the goal of rapidly obtaining flow field contours and characteristic parameter distributions based on basic flowfield geometric parameters.Research result shows that the constructed flow field rapid prediction model is characterized by high accuracy,with an overall average PSNR of 42.51 dB and an average SSIM of 0.9973.Key characteristics and parameter distributions are effectively extracted from the prediction results,providing strong support for the rapid design and optimization of the internal contraction basic flowfield.
作者
杨孔强
熊冰
范晓樯
王翼
唐啸
YANG Kongqiang;XIONG Bing;FAN Xiaoqiang;WANG Yi;TANG Xiao(Advanced Propulsion Technology Laboratory,National University of Defense Technology,Changsha 410073,China)
出处
《国防科技大学学报》
北大核心
2026年第1期28-39,共12页
Journal of National University of Defense Technology
基金
国家自然科学基金资助项目(12102470,12372298)
中国科协青年人才托举工程资助项目(YESS20230689)。
关键词
高超声速
内收缩
基准流场
参数方法
流场预测
残差神经网络
hypersonic
internal compression
basic flowfield
parametric approach
flow field prediction
residual neural network
