Research On Temperature Field Reconstruction And Image Denoising Of Flame Based On Deep Learning

Combustion is the most important energy source in human activities,thus it is very important to study combustion for its utilization and development.Combustion diagnostic is a vital discipline of quantitative research on the temperature and concentration distribution of components in the combustion field,which plays an indispensable role in the combustion control and stability analysis in key areas such as aerospace.Among many diagnostic technologies,optical diagnostic technology has been widely investigated and applied due to its non-invasive nature,especially the combination of optical measurement and tomographic technology,such as laser-induced nonlinear tomographic absorption spectroscopy(NTAS)and computed tomography of chemiluminescence(CTC)without additional light source.NTAS measure the absorbance integral along multiple laser paths in the combustion field,and the distribution of temperature field and species concentration field is reconstructed by simulated annealing algorithm;CTC is to obtain the 2D chemiluminescence projection images of flame from different angles,and calculate the 3D structure of flame by algebraic reconstruction algorithm.However,the simulated annealing algorithm cannot meet the on-line requirements due to its high computational cost;the 2D chemiluminescence projection images of flame usually preprocessed only by a simple threshold method.Based on the two shortcomings,this paper explores the feasibility of deep learning algorithms to solve the aforementioned problems.Recently,deep learning has been widely used in the fields of computer vision and natural language processing.The deep learning algorithm is also proved to be effective in inversion problems and image denoising.In this paper,we use deep learning algorithms to construct neural networks,which are applied to the inversion problem of NTAS solving the problem of low computational efficiency of simulated annealing.Meanwhile,we build a deep denoiser to deal with the 2D projection of CTC and improve the accuracy of 3D reconstruction.In the end,we construct a domainadversarial reconstruction network,solving the problem of reconstructing flame 3D structure with few labeled samples.