Research On Data Driven Intelligent Reconstruction Algorithm Of Scramjet Combustion Flow Field

The flow and combustion of scramjet engines are extremely complex.In milliseconds,various scientific issues will be involved,such as the interaction between shock waves and boundary layers,the mixing of fuel and oxidants,ignition and flame propagation,and drastic changes in combustion modes.To work efficiently and stably in complex flow and combustion environments,it is necessary to monitor and diagnose the engine.On the one hand,The monitoring and diagnosis technology for scramjet engines mainly relies on a single sensor to collect discrete and low-density raw information,which limits the accuracy of monitoring and diagnosis.On the other hand,the current hypersonic aircraft cannot carry heavy optical measurement instruments and cannot obtain combustion flow information during flight.Therefore,there is an urgent need for an intelligent combustion monitoring system capable of high-precision diagnosis to accurately reconstruct the combustion flow field of the scramjet in real time which can provide technical support for the safe and reliable operation of the scramjet during flight.Based on the above two engineering issues,this thesis has built a variety of deep learning models to achieve data-driven intelligent reconstruction of scramjet combustion flow field based on multi-source spatiotemporal data.Furthermore,a detailed analysis is conducted on the results of these models’reconstruction.The main research contents are as follows:(1)Research on the reconstruction method of hydrogen fuel self ignition flow field in combustors based on deep learning has been carried out.Through ground pulse combustion wind tunnel experiments,the schlieren flow field and upper and lower wall pressure data of the combustor were synchronously obtained.A deep learning model for intelligent reconstruction of the flow field based on the wall pressure in the combustor was proposed.The reconstruction accuracy of the combustor flow field was compared among three convolutional neural network models,verifying that the multi-branch fusion convolutional neural network(MBFCNN)model proposed in this paper achieved the highest reconstruction accuracy of the combustor flow field.In the test datasets with different equivalence ratios,the average peak signal-to-noise ratio(PSNR)reached 20.205 and the average correlation coefficient(R)reached 0.935.Finally,the impact of different number of wall pressure sensors on the accuracy of flow field reconstruction was analyzed.(2)Research on the super-resolution reconstruction method of the flow field in the combustor of a scramjet engine has been carried out.The combustor flow field obtained based on the reconstruction of wall pressure has the phenomenon of low resolution,blurred and excessively smooth.Therefore,two models of flow field super resolution dense network(FSRDN)and flow field super resolution generation countermeasure network(FSRGAN)were constructed,and compared with the traditional Bicubic interpolation interpolation method.Three methods have improved the resolution of flow field images by 4~2 times,and FSRGAN has achieved better results,optimizing the flow field image perception index(PI)to a smaller value,and clarifying complex combustion flow field structures such as shock wave structures and turbulent pulsations in confined spaces.(3)Research on deep learning based flame reconstruction methods for scramjet combustiors has been carried out.Based on the pressure data of the upper and lower walls of the combustor and the hydrogen injection pressure,a reparameterized convolutional neural network(Rep CNN)model was constructed to achieve rapid and high-precision reconstruction of the flame.Multiple evaluation indicators were comprehensively considered to analyze the flame reconstruction results,and the reconstruction results of the MBFCNN model were compared and studied.Under the condition of maintaining good reconstruction accuracy,the reasoning speed of the Rep CNN model was improved by 39.9%,reaching a speed of 2.83m/s per frame.The comparison of multiple indicators of the combustion flow field reconstructed by the self-designed deep learning model indicates that the method used in this paper provides a new approach for accurate prediction and rapid diagnosis of the combustion flow field in scramjet engines.