Prediction Of Wall Heat Flux Of Typical Shape Based On Deep Learning

Artificial intelligence,with its advantages of high efficiency and automation,has become an important means to supplement and improve existing technologies.In the field of heat flux calculation,Computational Fluid Dynamics(CFD)is currently the most commonly used method for obtaining heat flux due to its high accuracy and complete information acquisition.However,CFD computation is time-consuming and requires a high level of expertise from engineers,making it difficult to meet the demand for wall heat flux calculation in situations with a large number of different flight conditions and aircraft models.The rapid development of neural networks has provided a new way of predicting aerodynamic parameters.Based on deep neural networks,this thesis constructs a threedimensional wall heat flux prediction algorithm and a flow field aerodynamic parameter prediction algorithm.The main work is as follows:(1)SA-HFNet,a heat flux prediction network capable of automatically sensing changes in the shape of aircraft,has been designed.Existing intelligent heat flux prediction methods have insufficiently described three-dimensional external features and are not suitable for predicting heat flux when the shape of an aircraft changes.To address the issue of predicting wall heat flux for three-dimensional aircraft,this thesis proposes a local shape feature that describes the geometric characteristics of a single point,as well as a global shape feature that captures the overall shape of the aircraft to fully describe the threedimensional shape features of the aircraft.By predicting each point,SA-HFNet can accurately predict the wall heat flux for aircraft of different shapes.As far as we know,this is the first intelligent method that quickly obtains heat flux and adapts to both the overall and local shape changes of an aircraft.(2)Slice-FFNet,a three-dimensional aircraft flow field aerodynamic parameter intelligent prediction method based on deep learning,has been designed.Aerodynamic parameters such as pressure,velocity,and density of the flow field can be derived to obtain wall heat flux and intuitively represent flow field characteristics,which holds significant research value.Slice-FFNet segments and recombines slices to represent the shape of three-dimensional flow field,and predict pressure,velocity,and density of the surrounding flow field of a typical supersonic aircraft.The input of Slice-FFNet is designed based on the characteristics of supersonic inviscid flow.This thesis verifies the feasibility and effectiveness of Slice-FFNet through validation with a bluntcone-shaped aircraft.Aiming to address the difficulty of describing three-dimensional shapes in neural networks,this thesis proposes two different three-dimensional shape description methods for intelligent wall heat flux prediction and intelligent flow field aerodynamic parameter prediction,which can efficiently and accurately predict the corresponding physical quantities.The research on intelligent prediction methods for typical three-dimensional aircraft wall heat flux and flow field aerodynamic parameters can improve the efficiency of heat flux and flow field aerodynamic parameter prediction,providing data support for further supersonic aircraft shape design.