使用条件生成对抗网络生成预定导热率多孔介质

Predetermined thermal conductivity porous medium generated by conditional generation adversarial network

多孔介质在工程领域中的应用非常广泛,其中有效导热率和孔隙率为多孔介质材料非常重要的性质,得到一个符合需要的有效导热率和孔隙率的多孔介质材料具有重要意义.本文使用四参数随机生成方法制作了训练数据集,搭建了一个条件生成对抗网络(CGAN),使用预定的有效导热率和孔隙率作为输入,生成一个满足输入条件的多孔介质结构.特别地,由于多孔介质的孔隙结构分布对材料的有效导热率影响巨大,提出局部结构损失函数参与网络训练,使得网络更好地学习到孔隙分布与导热率之前的关系.通过使用格子Boltzmann方法验证神经网络生成的多孔介质结构的有效导热率,结果表明该方法能够快速且准确地生成预定参数的多孔介质结构.

Porous media are extensively used in the engineering field.The effective thermal conductivity and porosity are very important properties of porous medium materials.It is of great significance to obtain a porous medium material that meets the needs of effective thermal conductivity and porosity.In this paper,a four-parameter random generation method is used to produce a training data set,a conditional generation adversarial network(CGAN)is built,and a predetermined effective thermal conductivity and porosity are used as inputs to generate a porous medium structure that meets the input conditions.In particular,since the pore structure distribution of porous medium has a great influence on the effective thermal conductivity of the material,a local structure loss function is proposed to participate in the network training,so that the network can better learn the relationship between the pore distribution and the thermal conductivity.By using the lattice Boltzmann method to verify the effective thermal conductivity of the porous medium structure generated by the neural network,the results show that the method can quickly and accurately generate the porous medium structure with predetermined parameters.