Investigation On Acceleration Method For Hydrocarbon Combustion Simulation

For high-speed propulsion systems operating in a large range of speed,a broad range of trajectory,and a variety of combustion modes,such as the RBCC(Rocket-Based CombinedCycle)engine,the large variation of incoming flow and complex working conditions will lead to the extremely complex turbulent combustion.Limited by the high cost of ground test and flight test,the research and engine design through numerical simulation has been developed and applied greatly.However,due to the complexity and high computational cost of the detailed chemical reaction model of hydrocarbon fuel,it is still difficult to achieve high-precision,largescale combustion simulation of a full-scale engine.Therefore,it is of great practical significance to develop a numerical simulation acceleration method suitable for supersonic and other complex conditions.Under this background,based on the traditional method and the deep learning method,the applicability and characteristics of different combustion numerical simulation methods under complex simulation conditions are studied.The main research contents and conclusions are as follows:(1)The application and development of Dynamic adaptive chemistry(DAC)method are carried out.Firstly,the applicability of DAC in supersonic combustion simulation are studied.Based on this,the dynamic mechanism table,MRU algorithm and DP algorithm were proposed to accelerate the real-time simplified process of chemical reaction mechanism,namely the DAC-ST method.Furthermore,the DAC-ST and ISAT are integrated to form the DAC-DT method.The accuracy and efficiency were further studied by using the Sandia Flame D and the DLR.The results show that the DAC can obtain more simplified and optimized chemical reaction mechanism than the previous simplified chemical reaction mechanism.In terms of efficiency,for the Flame D cases using the 53-species mechanism of methane,the efficiency of the chemical integration can be increased by about 3 to 4 times,and for the DLR cases using the 9-species hydrogen mechanism,an acceleration of about 1.5 times can be obtained.The DAC-DT further accelerates the chemical integration.About 7.8 times of acceleration can be obtained for the Flame D,and 3.9 times for the DLR.(2)An ANN(Artificial neural network)-based chemical reaction mechanism construction method is developed.The improved LHS method was used to generate universal,hypothesisfree samples,and then the samples were transferred to the two-layer neural network combined with SOM and BPNN.The method was further verified by constructing a SOM-BPNN-based chemical reaction of methane.The results show that the calculation accuracy of the SOMBPNN-based methane chemical reaction is comparable to the original chemical reaction mechanism.In terms of efficiency,the integration time of the chemical reaction can be reduced by about two orders of magnitude.(3)Based on physical-informed deep learning Network(PINN),the solution method of NS equation based on PINN is exploratory studied.Incompressible N-S equation,compressible N-S equation and chemical reaction are respectively studied.Compared with the traditional numerical discrete method,PINN method has the same accuracy level.In terms of computational efficiency,the PINN method requires high computational cost in the training stage,while can achieve more than two orders of magnitude acceleration once the learning is completed.This study lays a solid foundation for the development of a new numerical simulation framework based on PINN.(4)By using the new method developed above,the validation and research of a kerosenefueled RBCC engine combined with concave cavity and struts are further carried out.In terms of accuracy performance,it is found that the three methods can obtain accurate numerical simulation results.As for computational efficiency,due to the complex combustion state and strong dynamics,the DAC-ST method and DAC-DT method have serious problems of unbalanced calculation amount between cores.Meanwhile,the percentage of retrieve of ISAT is insufficient,hence,1.26 times and 1.9 times of chemical integral calculation can be obtained in the end.However,the ANN-based chemical reaction can obtain the maximum efficiency gain of 19 times of chemical reaction mechanism calculation.Meanwhile,there is no imbalance of computation between cores,indicating that it has natural parallelism.