Research On A CPU-GPU Based Parallel Algorithm For Thermal Stress Model Of Continuous Casting Steel

The basis of the study of cracks of continuous casting steel is the thermal stress model.In the process of continuous casting,the defects which can be formed in cast material are the focus of attention,the problem of cracks is over 50 percent.The cracks are often caused by the billet subjected to stress exceeds its strength.In a condition of high temperature,the thermal stress model is usually established by numerical simulation method to analyze the internal stress state of the slab.The thermal stress model involves the calculation of the temperature and stress model.In the process of the numerical method of the Thermal stress model,Numerical solution is confronted with the huge problems of large amount of computational units and too long computing time.It is difficult to satisfy the demand of practical application of crack analysis and process optimization,and it is urgent to improve computational speed.With the improvement of the GPU hardware and CUDA parallel computing platform for highperformance parallel computing,it provides the conditions for the acceleration of problems of large-scale computation.Therefore,it is significance to study the thermal stress model of parallel algorithm based on CPU and GPU.The main contents in this paper are following:(1)Establishing 2D thermal stress model based on finite volume and finite element method.A heat transfer and solidification model based on the finite volume is established.Based on the heat transfer model,the stress model is established.The finite element method is used to discretize the model.(2)Designing and implementing a parallel algorithm for thermal stress model based on CPU and GPU.Numerical solution of the thermal stress model.The heat transfer and solidification model of billet is solved by parallel explicit algorithm.According to analysis of the stress model and characteristic of the finite element,the parallel processing method of element stiffness matrix and global stiffness matrix is implemented.The parallel parts of direct iterative method and Conjugate Gradient are applied.(3)The results and optimization of program are analyzed.For an example of billet with grid of 32 × 32 and time step of 0.01 seconds,the model is further accelerated by optimizing the memory access and the thread configuration of kernel function in the GPU.The results of the thermal stress model of parallel algorithm based on CUDA are consistent with the results of thermal stress model of serial calculation based on C language.And acceleration ratio for parallel computing is 3.034.Analyzing the effect of different speed of Continuous casting and quantity of water of secondary cooling,the stress of billet of weak cooling strength is less.