Parallel Computing Based Calculation Method For Molecular Weight Distribution Of Batch Free Radical Polymerization

The polymer property is greatly determined by the molecular weight distribution (MWD). The MWD calculation for free radical polymerization is essential to obtain quality polymer product. However, as the effective chain length of a typical polymer can be as large as 10s, the MWD calculation for a btach polymerization is always challenging. By using model decomposition, the dynamic model can be converted to a large-scale ordinary differential equations (ODEs). A method named sequential variable decoupling(SVD) method was proposed by our group to calculate the MWD effectively. To further speed up the calculation of MWD, we propose a parallel computing based method in this thesis. It is implemented both on a multi-core processor based parallel computing platform and a graphic processor unit (GPU) based parallel computing platform. The contributions of this thesis include the following three aspects.The parallel sequential variable decoupling (PSVD) method is proposed for MWD calculation. Both coarse-grained and fine-grained parallelism have been considered, which results to the three forms of PSVD method. The algorithm analysis was also presented to prove that the proposed method has the best efficiency in parallel computing.The PSVD method has been implemented on the multi-core processor based parallel computing platform. The coarse-grained parallelism form of PSVD method has been implemented according to the hardware and software features of the platform. The comparison of cost time with serial SVD method shows that, the PSVD method can greatly accelerate the calculation of MWD. Further experiments and result analysis show that the speed-up ratio of PSVD is linear to the number of cores of the platform under certain extend.The PSVD method has been implemted also on GPU based parallel computing platform. According to the two-level parallelism of the programming model and hardware structure of GPU, both coarse-grained and fine-grained parallelism have been implemented. Experimental results show that GPU based PSVD method can significantly speed up the calculation of MWD.