Material nonlinear analysis of truss structure based on automatic differentiation
Based on deep learning techniques,a solution is proposed for solving the material nonlinear problems of truss structures based on automatic differentiation technology.The displacement of each member is taken as the optimization variables,the member displacements are initialized with a random algorithm and the loads are divided into several sections.In each load section,the strains of the members are first calculated from the displacement parameters through the geometric relationships,and then the stresses of each member are determined using the constitutive relationships of material.The loss function is constructed based on the equilibrium equation of nodal force.To minimize the loss function,the automatic differentiation technique is employed to construct the computational diagram of the loss function with respect to nodal displacements,which realize efficient calculation of the gradient of the loss function.Finally,using the gradient descent method,the member displacements are iteratively optimized until the convergence criterion is met.Taking truss structures as the research object and using a linear reinforcement model,the material nonlinear problems of different trusses are solved.The computation results are compared with those of the finite element analysis,and the feasibility of the proposed method in the analysis of elastoplastic problems and its high accuracy is demonstrated.
deep learningautomatic differentiationgradient descent methodtruss structurematerial non-linearity
万方数据
维普
