Microstructures For Additive Manufacturing:Analysis,Design And Infilling Framework

As the essential part of smart manufacturing,additive manufacturing is one of the key technology driving global industrial improvement.In Smart Manufacturing 2025,MIIT(Ministry of Industry and Information Technology,China)emphasizes the importance of additive manufacturing and releases the improvement roadmap of additive manufacturing in the future.Additive manufacturing shows many advantages compared to traditional fabrication,such as it can be used to a mini batch fabrication with higher precision but lower cost.In addition,additive manufacturing techiques can be implement to producce those objects whose topology and geometry are extremely tiny and complicated.Superior to molding-based fabrication and CNC fabrication,the core competence of additive manufacturing is its capability to fabricate objects with complicated inner structures.Therefore,we can achieve excellent mechanical properties from well-designed and preciselymanufactured microstructures.However,this also poses new requirements and challenges for existing structural design work,such as slow simulation analysis speed in existing microstructure analysis design and application,difficulty in inverse design of microstructures,and inconsistent simulation results of non-conformal microstructure infilling with actual results.Research and exploration are needed in areas such as fast homogenization of microstructure material properties,inverse design of microstructures,and conformal boundary microstructure infilling.This thesis focuses on the field of additive manufacturing,mainly focusing on the design,modeling,simulation,and related geometric issues of microstructures.Specifically,this study explores the effective material property method for fast homogenization of irregular microstructures based on convolutional neural networks,and the microstructure design problem driven by macroscopic material properties based on generative models in deep learning.In terms of applications,this study also investigates a two-scale modeling,simulation,and optimization framework suitable for irregular microstructures.The innovations and contributions of this thesis mainly include the following aspects:1.PH-NET:a fast homogenization method for heterogeneous microstructuresIn this paper,we proposes a method of convolutional neural network for predicting the macroscopic homogeneous properties of microstructures,instead of using numerical homogenization methods that based on finite element analysis.Compared with the latter,PH-NET reduces the time complexity of predicting the macroscopic effective properties of microstructures by 2-3 orders of magnitude.This article also introduces a method for calculating the shape-material transformation of irregular microstructure to ensure that PH-NET can be trained without modifying the convolutional neural network architecture and has better generalization for microstructures with different boundaries and base materials.At the same time,this study innovatively introduces a loss function based on the minimum potential energy principle in the prior knowledge of homogenization,enabling PH-NET to be trained without labels and significantly reducing the cost of constructing the dataset.2.A property-driven microstructure inverse design methodThe paper proposes a property-driven microstructure design method that utilizes a generative deep neural network based on a variational autoencoder driven by material properties as inputs to generate new microstructures.Compared to previous methods,this paper considers simulation and manufacturing constraints of microstructures during the training process,adopts Occupancy networks as the backbone network for microstructure generation,and constructs an initialization layer based on Fourier series to ensure the periodicity of microstructures.In order to ensure the connectivity of microstructures:this paper introduces a mask layer to enable the generative network to learn microstructure completion in a self-supervised manner,and uses pointwise interpolation called shape function to achieve boundary compatibility between different microstructures.Additionally,the paper uses Gaussian mixture models to establish a multiple mapping between the material property space and the latent distribution of microstructures,significantly expanding the space of property-driven microstructure design.3.A boundary-conforming infilling framework for irregular microstructuresThis article proposes a modeling,simulation,and optimization framework based on two scales of irregular microstructures infilling.The framework uses a hexahedral structure for coarse mesh at the macroscopic scale and PH-NET to predict the macroscopic effective properties of the infilling microstructure at the microscopic scale.This method solves the simulation errors caused by boundary clipping in the two-scale framework based on regular meshes and the inability to adaptively adjust the mesh distribution.This article also solves the gradient calculation problem of PH-NET in predicting the properties of microstructures,eliminating errors caused by rigid body transformations in infilling microstructures.To reduce the prediction error of the properties of the two-scale microstructure,this article introduces a hexahedral mesh optimization method to constrain the deformation range of hexahedral units.The article introduces a coupled two-scale microstructure modeling framework,in which macroscopic mechanical behavior guides the design of elastic analysis while simultaneously performing internal structural strength optimization at the two scales.