Research On Intelligent Sensing Technology Of Scramjet Engine Based On Deep Learning

With the development of hypersonic propulsion technology,the critical parameter monitoring and state sensing requirements of scramjet engines are increasing,but there is no special monitoring system to fully monitor the parameters of scramjet engines.Besides,since the combustion efficiency of the scramjet engine is closely related to the flow combustion organization process in the combustion chamber,the parameter monitoring in the combustion chamber should pay more attention to its field distribution.Based on the above two points,this paper constructs a super-combustion ramjet intelligent sensing system to realize comprehensive parameter monitoring of the engine,and proposes several key technologies base d on deep learning to realize field distribution reconstruction and state perception.Further on these key technologies The application research was carried out,and the main contents are as follows:Based on the multi-level structure of the scramjet engine,we built the engine intelligent sensing system and used its coordinated matching characteristics to further construct the intelligent sensor network.Through the information transmission of each layer of the network,it realizes three important functions: engine operating parameter monitoring,engine state change perception and sensor failure processing.We have proposed a number of key technologies based on deep learning,introducing the multi-information fusion technology and integrating it into the en gine intelligent sensing system.The application research of the temperature field reconstruction algorithm of the combustion chamber wall based on deep learning is carried out.The algorithm was originally constructed by a deconvolution network.Under the simulated data set,high-precision wall temperature field reconstruction can be achieved,but the reconstruction result is too smooth and the details are not obvious.Furthermore,the reconstruction algorithm is used to construct the wall temperature field reconstruction algorithm.The results show that the reconstruction result of the algorithm can obtain more obvious details of the flow field shock structure at the cost of a small mean square error.The application research of the combustor section reconstruction algorithm based on TDLAS-GAN is carried out.The algorithm is based on TDLAS technology.The first solution is to first obtain the low-resolution temperature field through the algebraic reconstruction algorithm,and then use the super-resolution algorithm to enhance the data.In the second scheme,the integral value of the optical path is directly input,and the high-resolution temperature field is used to construct an antinetwork for the output to realize the reconstruction of the characteristic section temperature field.The training test was carried out in the 3D simulation data set.The results show that compared with the first scheme,the second scheme has lower requirements on the number of absorption spectra and the number of optical paths,and has higher precision,which can complete the task of reconstructing the characteristic section temperature field of the combustion chamber.The application research on the combustion state discrimination technology based on voice recognition is carried out.The voice recognition model takes the spectral map as input and uses the deep convolutional network(DCNN)structure,while the model 2 takes the Mel frequency cepstral coefficient as input and adopts the CLDNN network structure.Cross-validation in the underwater acoustic data set shows that compared with Model 1,Model 2 has higher classification accuracy and stronger generalization ability in the four classification tasks.Further refining th e task into an eight-category problem,the state label errors caused by experimental errors and insufficient data sets are amplified,which leads to a sharp decline in the generalization ability of the voice recognition model.Through the application research on underwater acoustic data,the application prospect of sound recognition technology in the determination of combustion state is proved.At the same time,higher requirements are put forward for sound data collection and category labeling.