Research On Combustion Field Absorption Spectroscopy Tomography Method Based On Deep Learning

Combustion is used in a wide range of applications,on the one hand the main way of obtaining energy in industry.On the other hand,it is the main source of energy for transportation.It is an important research direction to test the combustion process and to grasp the combustion efficiency in real time.With the development of artificial intelligence and TDLAT there is an increasing demand for speed,accuracy and applicability of reconstruction algorithms.To address the shortcomings of existing TDLAT algorithms in terms of imaging accuracy and imaging efficiency,this paper focuses on a hierarchical TDLAT temperature distribution reconstruction method based on deep learning,as follows.Firstly,most of the existing TDLAT schemes use the absorption values of the laser beam in the whole measured space to reconstruct the temperature distribution in a local Ro I in the central region of the combustion field,which can lead to the problem of biased reconstruction results.In this paper,a spatially hierarchical discretization of the combustion field is proposed,and then a hierarchical temperature laminar imaging scheme based on a residual network is designed.The scheme enables a complete reconstruction of the temperature image of the whole combustion field based on a limited number of spectral absorption measurements,and optimises the allocation of computational resources to the imaging resolution of different spatial regions of the combustion field,focusing on achieving high spatial resolution imaging of the temperature distribution within the Ro I.Secondly,a novel quality-hierarchical temperature imaging network for TDLAS tomography is developed based on stacked Long Short Term Memory.From limited lineof-sight TDLAS measurements,this network with different number of layers reconstructs the temperature distribution image of the whole combustion field in two qualities,i.e.b coarse-quality image and fine-quality image.The coarse-quality reconstruction enables realtime dynamic monitoring of the combustion field at a low computational cost,while the fine-quality reconstruction provides more detailed information on the temperature distribution of the combustion field and meets the needs of high-precision off-line analysis and diagnosis of the combustion process.Finally,this paper designs a spatially hierarchical TDLAT flame temperature imaging scheme based on Convolutional Neural Networks and Transformer.The scheme introduces a spatially hierarchical discrete model and designs a two-stage reconstruction based on Convolutional Neural Networks and Transformer respectively,i.e.the high-efficiency reconstruction stage and the high-accuracy reconstruction stage.Among them,the highefficiency reconstruction stage focuses on achieving high spatial resolution reconstruction within the Ro I at a lower computational cost;the high-accuracy reconstruction stage makes full use of the spatial correlation of the background hierarchical discrete model to achieve high spatial resolution reconstruction of the complete burning area.