Abstract Numerical simulation models for laser powder bed fusion of metals (PBF-LB/M) vary in complexity and fidelity, ranging from high-fidelity models that capture melt pool dynamics to simplified models suited for part-scale temperature predictions and process optimizations. Validation against experimental data is essential to build confidence in their predictive capabilities. However, for in-situ thermographic measurements, a direct comparison is challenging due to the data’s size and the multi-scale nature of the process. Similarities must be analyzed at different spatial and temporal scales based on the model’s fidelity and its intended application. For example, agreement between a thermal simulation and measurement in a steady-state scenario does not guarantee accuracy during transient phases. Statistical similarity measures provide a quantitative means to assess model-measurement agreement, highlighting regions of high and low similarity. In this work, we validate a thermal simulation model, discretized using the space-time finite element method, against thermographic camera measurements using various similarity measures and evaluate their applicability to PBF-LB/M. We also propose a multi-scale similarity assessment approach tailored to model fidelity and application.

Abstract The increased adoption of polymer matrix composites (PMCs) in failure-critical applications is impeded by the challenges in developing reliable datasets for predictive models linking salient attributes of PMC microstructures to their damage resilience properties. We present a comprehensive set of computational protocols for producing high-value simulation datasets that can be used for building the desired machine-learnt models. These new protocols combine (i) a novel generative approach to produce ensembles of distinct statistical volume elements (SVEs) targeted to specified combinations of fiber volume fractions and the degree and directionality of fiber clustering, and (ii) consistent protocols for the construction of extreme value distributions describing microscale damage drivers from finite element-predicted stress fields. It is demonstrated that the proposed protocols can produce a large dataset comprised of distinct SVEs in a computationally efficient manner, and the produced dataset is openly shared with the broader research community to serve as a benchmark for future studies.

Abstract Multiscale crystal plasticity modeling of metal forming offers potential for effective design processes that consider microstructural evolution during significant plastic deformation. However, conventional multiscale methods for forming are expensive due to complex loading conditions and high-cost microscale models, making them challenging to apply in practice. These methods can be significantly accelerated through use of low-cost machine learning surrogate models for the microscale response. While these techniques have been demonstrated for simple load cases, they have not yet been demonstrated for manufacturing applications with full-field texture results. In this work, we develop efficient multiscale simulation workflows for manufacturing through single-point incremental forming using recurrent neural networks for constitutive response and texture evolution of a crystal plasticity model. This approach achieves up to a 63.6x increase in speed compared to conventional techniques and is demonstrated for an aluminum alloy with two unique forming paths. These workflows yield consistent trends between forming force and thickness variation, texture results in agreement with ground truth models, and permit extraction of the full-field texture evolution over the entire formed part. This enables efficient multiscale trade studies and optimization of local microstructures in industrial forming applications.

Abstract High-performance and lightweight materials design is a pressing need in aerospace applications (e.g., aircraft ailerons, flaps, and rudders). However, the unique functionality requirements in strength, weight, and resistance to environmental factors, such as temperature fluctuations and corrosion, challenge traditional structure design methods such as topology optimization. While sandwich panel composites with lattice cores are widely used in aerospace components and modern additive manufacturing techniques open new possibilities for sandwich core structure design with requirement functionalities, the delicate design brings computational challenges for both optimization and manufacturing. This paper presents an inverse design framework for sandwich structure optimization with implicitly represented architected cellular materials to address these issues. Specifically, cellular materials are implicitly represented (described by implicit functions) as building blocks in the core structure design. A multi-objective topology optimization problem is formulated to maximize the core structure’s mechanical and thermal performances. Lastly, the nature of function representation for ease-of-additive manufacturing computations is illustrated with a direct slicing algorithm without generating memory-expensive standard tessellation language (STL) files. The proposed design framework is validated in two practical aerospace design case studies, and experimental results demonstrate the effectiveness of the proposed optimization algorithm and STL-free scheme for additive manufacturing.

Abstract We present a data-efficient neural-network model for predicting linear-elastic properties of spinodoid metamaterials from their mesoscale structure. Our machine-learning model requires 75 data points for training, greatly improving data efficiency over previous models that required thousands of training samples. We achieve this by leveraging concepts from geometric learning. Specifically, we exploit physical properties, such as positive semi-definiteness of the elasticity tensor, as well as structural invariances and equivariances of the problem, for example with respect to coordinate axes permutations. The neural network model is designed to exactly fulfill these constraints; it does not have to learn them from data. The resulting model enables data- and compute-efficient inverse design of spinodoid metamaterials. In inverse design, the goal is to find a material mesostructure that leads to desired mechanical properties on the macroscale. Exactly fulfilling physical and structural constraints, the present neural network model remains differentiable. This allows using fast gradient-based optimizers for inverse design. We demonstrate this by inversely designing spinodoid metamaterials that achieve desired linear elastic target properties in three dimensions. Inverse design is treated as a constrained optimization problem over the parameters describing the metamaterial. The results confirm that the present approach requires significantly less training data than previous machine-learning approaches and allows incorporating multiple objectives in the inverse design process. Since the structure of the design space is independent of the target material properties, we hope that such data-efficient models will be useful also for inverse design of spinodoids beyond linear elasticity.