Research On Deep Learning Based Combustion Field Temperature Tomography

The heat energy generated by combustion is the main source of industrial energy.In order to improve the conversion efficiency between chemical energy and heat energy,it is very important to monitor the combustion field and master the dynamic change of gas parameters.Tunable Diode Laser Absorption Tomography(TDLAT)can reconstruct twodimensional distribution images of gas parameters in combustion field based on multi-path projection data.The temperature distribution of combustion field has strong correlation with combustion efficiency and pollutant generation.In order to improve the quality of temperature image reconstruction,this paper discusses the temperature reconstruction method of non-uniform transient combustion field based on deep learning and TDLAT technology.The main work is as follows:Firstly,in order to improve the accuracy of existing temperature reconstruction algorithms based on deep learning,a temperature tomography model based on crossfeature attention mechanism is designed.By introducing convolution operator to improve the problem of insufficient extraction of local features from the self-attention mechanism,a cascade form of cross-feature attention mechanism is adopted to carry out information exchange between the extracted local features and global features.The flame dynamics simulation tool was used to construct the data set,and the simulation and real combustion field experiments were carried out for the proposed model.Compared with existing algorithms based on deep learning,this scheme improves the accuracy of temperature reconstruction.Secondly,in order to further improve the quality of temperature image reconstruction by utilizing multi-scale correlation features of measurement data from local to global,this paper introduces Vision Transformer(Vi T)and multi-scale feature fusion into TDLAT domain.A temperature tomography network based on fusion of cascade Vi T and multiscale features is proposed.The network uses cascaded Vits to extract multi-scale features of measured data step by step from local to global,and constructs multi-scale feature fusion architecture to achieve high precision reconstruction of temperature distribution.Simulation and real combustion field experiments show that the network can further improve the quality of temperature image reconstruction.Finally,a multi-output temperature imaging scheme is designed to achieve high quality reconstruction of temperature images of the central combustion area and the entire sensing area.In order to quickly monitor the dynamic change of flame temperature,the combustion zone temperature reconstruction is realized with low calculation cost.In order to reconstruct the thermal radiation phenomenon formed by the interaction between the flame boundary and the cold air,the temperature of the sensing area is reconstructed with high precision.Compared with the existing algorithm,which needs to run twice to output two temperature images respectively,this scheme not only realizes the simultaneous reconstruction of two images,but also outperforms the existing algorithm in reconstruction accuracy.