Research On FCT Reconstruction Technology Based On Convolutional Neural Network

Flame emission spectroscopy tomography(FCT)is a technology that combines flame emission spectroscopy and tomography for non-contact flame field diagnosis.This method mainly measures the radiation intensity of various free radicals generated during the flame combustion process.It can directly detect and receive information with an industrial camera.Its experimental device is relatively simple,easy to install and implement,and has a wide range of non-contact and transient states.The advantages of full-field measurement have gradually become one of the research hotspots in the field of combustion diagnosis.Iterative algorithms are the basic reconstruction algorithms for various CT problems,especially in the reconstruction of incomplete projection data,and they are the most commonly used algorithms in FCT reconstruction.However,the iterative algorithm has the problem of poor general adaptability of the iterative parameters during the reconstruction process.For different CT reconstruction data,the iterative parameters are different,and the iterative time is long,which makes it difficult to achieve real-time monitoring of the flame field.In view of the above problems,this paper mainly studies the application of Convolutional Neural Networks(CNN)in FCT reconstruction,and explores a new rapid reconstruction technology of flame combustion field based on deep learning.The main work of this paper is as follows:Firstly,in view of the shortcomings of the iterative algorithm in FCT reconstruction,this paper designs a CT reconstruction method based on CNN combustion field.Multi-directional projection data was obtained as the input of the network through Radon transform,and the reconstructed field was used as the output data.After the CNN was constructed and the training was completed,the reconstruction of the two-dimensional Gaussian field was successfully achieved,which proved the feasibility of the method.Secondly,the method was extended to 3D CT reconstruction,and the CNN model was built and trained to successfully implement 3D FCT reconstruction.The influence of the constructed CNN network structure on the reconstruction results is discussed.Based on the foregoing work,FCT reconstruction experiments of real combustion fields were performed.After building the CNN model and completing the training,the fast reconstruction of the three-dimensional structure of the candle flame was successfully achieved.Compared with the original iterative algorithm,the time required for the CNN model to predict the three-dimensional structure of the flame is extremely short.Through the research in this paper,a new way combined with deep learning is explored for the reconstruction technology of flame emission spectrum CT.Using the research in this paper will effectively improve the reconstruction speed of FCT,and it is expected to realize the real-time monitoring and measurement of the combustion flame field.