Study On Temperature Distribution Reconstruction Method

In real-world combustion and heat transfer equipments,obtaining accurate temperature distribution information plays a crucial role in ensuring the safety operation of equipmens,reducing pollutant emissions and improving the operation efficiency of system.This thesis mainly studies temperature distribution reconstruction methods,and the research contents are as follows:The basic principles of temperature distribution measurement methods are analyzed and discussed in detail,and their advantages and disadvantages are summarized.A new reconstruction method for the inverse problem with the focus on the acoustic tomography is proposed to improve the precision of measurement and robustness.The method consists of two stages.In the first stage,the measurement area is divided into a coarse discrete meshe,which reduces the number of unknown variables and alleviates the ill-posed nature of the temperature distribution reconstruction problem.A new objective function is constructed to transform the inverse problem in the acoustic temperature distribution measurement into an optimization problem,in which the Tikhonov regularization method is introduced to stabilize the numerical solution,the L1-norm is deployed as the data fidelity to alleviate the influences of the outliers contained in the measurement data on the final estimation result,the weighted nuclear norm and the weighted L1-norm are deployed as the regularizers to emphasize the low rank and smoothness of a soltuion.The split Bregman iteration method is developed to efficiently solve the proposed cost function to get the temperature distribution on the coarse mesh.In the second stage,the measurement area is further divided into a finer discrete grid.The extreme learning machine method is used to predict the temperature distribution information on the refined mesh and capture the details of temperature distribution.Numerical experiment results confirm that the proposed method is feasible,and the reconstruction accuracy outperforms popular reconstruction methods,which makes a breakthrough in the numerical solution of the inverse problem with the focus on the acoustic temperature measurement.A low-dimensioal representation based reconstruction method is put forward to accurately reconstruct the temperature distribution from the finite number of local temperature measurement data,which significantly reduces the number of measurement sensors and decreases the complexity and cost of measurement.In the proposed mwthod,the nonnegative matrix factorization method is empolyed to extract the low-dimensional representation basis matrix,which can ensure that nonnegative characteristic of the data matrix and has a clear physical meaning.With the consideration of the inaccurate properties about the approximate mode and the measurement data,a new cost function is constructed to convert the estimation problem of the low-dimensional representation coefficient into an optimization problem,and a new numerical method is developed to efficient solve it.Numerical simulation results validate that the proposed reconstruction method is feasible and effective,and can accurately reconstruct the temperature distribution under a low sampling ratio.Different from the tomography-based temperature measurement methods,the proposed method is not subject to the scale of measured objects and does not require the closure of the sensor array.Unlike conventional numerical simulation methods,the new method does not need to solve complex control equations,and does not require initial conditions,boundary conditions,physical conditions,etc.Different from conventional local measurement methods that only obtain the local temperature distribution information,the new method can accurately reconstruct the temperature distribution from the limited number of scattered observations.A new reconstruction method is proposed to reconstruct three-dimensional temperature distribution from the finite measurement information,which can significantly reduce the complexity and cost of measurement.Different from conventional measurement methods,the new method uses Gaussian process regression(GPR)method to establish the mapping model between the measurement point and the temperature distribution,and does not need to provide the physical three-dimensional temperature distribution measurement model.Numerical simulation results show that the new method has good robustness,does not solve complex control equations,and can accurately reconstruct three-dimensional temperature distribution from a relatively low sampling ratio.Moreover,the GPR method has favorable numerical performances,and has good adaptability to complex problems with small sample and high dimensionalities,and the estimation result has the probability significance.Furthmore,the GPR method has the distinct advantages,including excellent generalization ability,easy numerical implementation,adaptive choice of super-parameters,etc.A three-dimensional diesel combustion experiment platform is builded,the GPR method is deployed to successfully predict the three dimensional temperature distribution according to the limited number of the temperature measurement data from high-temperature thermocouples.The feasibility and effectiveness of the GPR method on the three-dimensional temperature distribution measurement are validated through the comparison of the predicted results and the measured data.The findings of the study provides a new way for the temperature distribution measurement,which will contribute to the energy saving and emission reduction strategy advocated in China.