Research On Explicit Topology Optimization Considering Parallel Computing And Data Driven

Topology optimization is a fundamental and systematic method which has a powerful ability to create new design ideas.There are a wide range of various challenging structural design problems emerging during industrial production and scientific research.The increasing demand for more efficient,accurate and competitive advance structural optimization design method is required urgently.However,traditional pixel-based topology optimization methods have some insurmountable shortcomings,such as implicit description of structural boundary,too intricate to be made by existing manufacturing methods,immense amount of number of design variables,large computational demand and so on.Therefore,combining with cuttingedge technology of computer science to develop efficient and accurate topology optimization methods has been became one of the most promising development trends in the field of structural optimization in the future.For the purpose to integrate theory with practice and advanced algorithms should be combined with cutting-edge scientific computing technologies.Based on this,parallel computing and data driving are introduced into the topology optimization based on Moving Morphable Component(MMC)framework.In the present work,a large-scale parallel algorithm based on Moving Morphable Component framework(MMC)and Machine learning-Driven real time topology optimization baced on Moving Morphable Component(MMC)framework are proposed.The main specific research contents and results are divided into the following two sections:(1)The research on Parallel Computing based on Moving Morphable Component framework.With the increasing demand of analyzing and optimizing large-scale engineering model,programming based on scalable libraries PETSc(Portable,Extensible Toolkit for Scientific Computation)and C++ linear algebra template libraries Eigen with Object Oriented design to abstract topology optimization into several designed object,the main focus of the present work introduces a large-scale parallel algorithm based on Moving Morphable Component(MMC)framework.Several numerical examples reveal that the proposed method is particularly advantageous for increasing computational efficiency to Finite Element Analysis substantially and provides a high efficiency computing tool for three-dimensional higher resolutions topology optimization problem.(2)The research on Machine Learning-Driven Real-Time Moving Morphable Component framework.In the present work,with the help of characteristics of explicit boundary description based on Moving Morphable Component(MMC)framework,it is intended to consider Moving Morphable Component(MMC)topology optimization framework with machine learning techniques by using Supported vector regression(SVR)and K-nearestneighbors(KNN).The machine learning models are employed to map the external loading to the optimized design variables of components by building a mathematical functional relationship.What is worth mentioning is that as compared to existing topology representational approaches based on pixel points,the proposed approach in this paper can not merely slash the amount of training data from dozens of GBs into several MBs in learning process massively,but also take full advantage of dimensionality reduction of parameter space.Besides that,it has remarkable advantages to have engineering hunch during design process with respect to multifarious external loadings.The proposed paradigm demonstrates huge potential in realizing real-time topology optimization in a comprehensive way are verified by several numerical examples.