The Research On Phase Field Model Simulation Based On GPU-CUDA And Deep Learning

Using the phase field model to simulate the growth and evolution of dendrites over time can help us to deeply explore the microscopic nature of dendrite evolution and improve related research theories.In addition,with the development of machine learning and deep learning,applying machine learning and deep learning to the calculation and prediction of phase field models can further explore the internal mechanism of the model.The main work of this paper is as follows:(1)In this paper,a PF-LBM model including flow field,temperature field,solute field and phase field is established by coupling Phase Field Model(PF)and lattice Boltzmann Model(LBM).Using the model to simulate the growth and evolution process of equiaxed dendrites under natural convection conditions,and explore the growth law and related influencing factors of dendrites under different conditions in the presence of natural convection.Based on the model,this paper simulates the dendrite growth under natural convection.The growth process of dendrites with different preferential growth angles under the condition of natural convection,dendrites with the same undercooling and different inclination angles,and equiaxed dendrites with the same inclination angle and different undercooling are studied and analyzed respectively.The external factors and internal mechanisms affecting the length of the main dendrite arm length and the tip velocities of dendrite are analyzed.(2)In order to solve the problem of low efficiency of the traditional CPU serial computing PF-LBM model,this study uses the CPU+GPU heterogeneous system and adopts the GPU-CUDA programming mode to design and implement the simulation process in parallel.Combined with the characteristics of the model,the different calculation modules of the simulation work are divided,and the parallel model is further optimized based on the GPU-CUDA programming mode.The parallel optimized model has achieved obvious acceleration effects.And compared with the calculation of the PF-LBM model using only the CPU,the calculation speed of the PF-LBM model based on GPU-CUDA reaches more than 24 times.(3)This paper combines machine learning and deep learning methods with phase field models to predict quasi-phase equilibria.The paper first uses the least squares method to obtain the required data and then applies eight machine learning methods and five deep learning methods to train the quasi-phase equilibrium prediction models.After obtaining different models,this paper compares the reliability of the established models by using the test data and uses two evaluation criteria to analyze the performance of these models.This work find that the performance of the established deep learning models is generally better than that of the machine learning models,and the Multilayer Perceptron based quasi-phase equilibrium prediction model achieves the best performance.Meanwhile,the Convolutional Neural Network based model also achieves competitive results.The experimental results show that the model proposed in this paper can predict the quasi-phase equilibrium of the KKS phase-field model accurately,which proves that it is feasible to combine machine learning and deep learning methods with phase-field model simulation.