| The thermal physical property parameters of materials are used to characterize the heat transfer and heat carrying capacity of materials.which is an important basis for material selection and thermal process analysis.The transient plane heat source method(TPS)is one of the important general testing methods for thermal conductivity and thermal diffusion coefficient and is widely used for testing the thermal physical parameters of various materials such as fluids,solids,powders,thin plates,and thin films.The thermal conductivity and thermal diffusion coefficient of the test material can be quickly obtained for TPS by least-squares iterative fitting based on the instantaneous average temperature rise collected by the sensor.But the analytical model used for data analysis ignores the influence of sensor geometry,heat capacity,and contact thermal resistance on measurement accuracy.In this paper,in order to improve the test accuracy and inversion efficiency,the inverse thermal conductivity optimization model of TPS was established by using the thermal conductivity differential equation as the constraint conditions.combined with parallel Bayesian optimization.The main work of this paper is as follows:(1)To improve effectively the calculation accuracy of the numerical model,a thermal conductivity model of TPS was established,which took the thickness,heat capacity and contact thermal resistance of the sensor into consideration.The unstructured meshing was used in thermal conductivity model and the finite volume method was used to discretize the thermal conductivity differential equation.The comparison of the model with the simulation results of commercial CFD software and the concentric ring analytical model shows that the relative errors of the instantaneous average temperature of the heater and the analytical solution are controlled within ±0.1%and ±5%,respectively,which verifies the correctness of the established numerical heat transfer model of the transient planar heat source method.(2)In order to improve the efficiency of inversion.Bayesian optimization based on Gaussian process is used to dynamically build and update probabilistic agent models,and a parallel Bayesian optimization algorithm based on a multi-objective hybrid strategy is proposed.To improve the utilization of computational resources,a parallel Bayesian optimization algorithm based on a multi-objective hybrid strategy was proposed.The algorithm was made up of two separate multi-objective parallel strategies that chose parallel solutions iteratively using a probabilistic function that broadens the range of sample locations.The Pareto frontier representing the best trade-off of the collection function was found by a multi-objective optimization algorithm under each parallel strategy.When the batch evaluation was feasible,the optimization was accelerated by uniformly collecting multiple points in the Pareto frontier.Finally,the proposed algorithm is tested against three other more advanced parallel optimization algorithms in six standard functions as well as single-parameter inversion examples,and the good test results verify the accuracy and efficiency of the developed parallel algorithm.(3)The identification problem for thermal conductivity,specific heat capacity,and contact heat resistance was changed into a minimization optimization problem,and a multi-parameter collaborative optimization inversion framework was established by fusing a parallel Bayesian optimization algorithm based on a multi-objective hybrid strategy with a numerical thermal conductivity model of the TPS.Through sensitivity analysis,it was found that the contact thermal resistance is highly correlated with the thermal conductivity and specific heat capacity over a long period of time,and by studying the influencing factors of the parameter inversion,it was found that the choice of the time range has a significant influence on the multi-parameter inversion.The maximum relative error of specific heat capacity was around 10%.The initial number of samples and the number of parallel points have a small impact on the inversion error,and the increase in the number of parallel points can significantly accelerate the convergence speed of the algorithm. |
