Research On Temperature And Concentration Reconstruction Based On Deep Learning And Absorption Spectroscopy Tomography

Combustion is the main way to obtain energy in contemporary society,and in order to improve the combustion efficiency and reduce the pollution generated by combustion,it becomes crucial to grasp the combustion process in real time.With the development of machine learning and the proposed Tunable Diode Laser Absorption Spectroscopy Tomography(TDLAT),there is an increasing demand for the efficiency,accuracy,and universality of reconstruction algorithms.In order to perform combustion diagnosis more accurately,this paper revolves around temperature and concentration distribution reconstruction algorithms based on deep learning and laser absorption spectroscopy,and the main contents of the paper are as follows:Firstly,a temperature imaging scheme based on Swin Transformer is designed in this paper.In this paper,the entire combustion field is taken as the region of interest,and a 2D image of the temperature of the entire combustion field is reconstructed based on a limited number of laser absorption measurements.The scheme uses a multi-headed self-attention mechanism across windows as a way to capture the rich near-and far-range features in the laser path absorption values,to complete the interaction between information and to enhance the expressiveness of key features,thus achieving the goal of improving the accuracy of temperature reconstruction in the combustion field.Secondly,a progressive temperature imaging scheme based on convolution and selfattention mechanism is designed in this paper.The scheme reconstructs two images of temperature distribution with different resolutions for the whole combustion field.The network in the scheme uses convolution and self-attention modules to extract features and fuse the extracted information separately,which solves the problem of missing global sensory field faced by using convolution layer alone and also solves the problem of insufficient local representation that may be faced by using self-attention module alone,further improving the accuracy of combustion field temperature reconstruction.Finally,for the demand of simultaneous imaging of temperature and concentration in practical combustion diagnosis,a Swin U-Net-based simultaneous imaging scheme for temperature and concentration is designed in this paper.The scheme reconstructs the twodimensional distribution images of temperature and concentration in the combustion field simultaneously.The network in the scheme acquires rich contextual features through feature maps at different scales,and enhances the image reconstruction process by jumping connections between deep and shallow levels,thus effectively improving the accuracy of simultaneous temperature and concentration reconstruction in combustion fields.