Effect of Constituent Materials on Composite Performance: Exploring Design Strategies via Machine Learning

Machine learning-enabled forward prediction and inverse design of 4D-printed active plates

Shape transformations of active composites (ACs) depend on the spatial distribution of constituent materials. Voxel-level complex material distributions can be encoded by 3D printing, offering enormous freedom for possible shape-change 4D-printed ACs. However, efficiently designing the material distribution to achieve desired 3D shape changes is significantly challenging yet greatly needed. Here, we present an approach that combines machine learning (ML) with both gradient-descent (GD) and evolutionary algorithm (EA) to design AC plates with 3D shape changes. A residual network ML model is developed for the forward shape prediction. A global-subdomain design strategy with ML-GD and ML-EA is then used for the inverse material-distribution design. For a variety of numerically generated target shapes, both ML-GD and ML-EA demonstrate high efficiency. By further combining ML-EA with a normal distance-based loss function, optimized designs are achieved for multiple irregular target shapes. Our approach thus provides a highly efficient tool for the design of 4D-printed active composites.

An interpretable deep learning approach for designing nanoporous silicon nitride membranes with tunable mechanical properties

The high permeability and strong selectivity of nanoporous silicon nitride (NPN) membranes make them attractive in a broad range of applications. Despite their growing use, the strength of NPN membranes needs to be improved for further extending their biomedical applications. In this work, we implement a deep learning framework to design NPN membranes with improved or prescribed strength values. We examine the predictions of our framework using physics-based simulations. Our results confirm that the proposed framework is not only able to predict the strength of NPN membranes with a wide range of microstructures, but also can design NPN membranes with prescribed or improved strength. Our simulations further demonstrate that the microstructural heterogeneity that our framework suggests for the optimized design, lowers the stress concentration around the pores and leads to the strength improvement of NPN membranes as compared to conventional membranes with homogenous microstructures.

Nature-inspired architected materials using unsupervised deep learning

Nature-inspired material design is driven by superior properties found in natural architected materials and enabled by recent developments in additive manufacturing and machine learning. Existing approaches to push design beyond biomimicry typically use supervised deep learning algorithms to predict and optimize properties based on experimental or simulation data. However, these methods constrain generated material designs to abstracted labels and to “black box” outputs that are only indirectly manipulable. Here we report an alternative approach using an unsupervised generative adversarial network (GAN) model. Training the model on unlabeled data constructs a latent space free of human intervention, which can then be explored through seeding, image encoding, and vector arithmetic to control specific parameters of de novo generated material designs and to push them beyond training data distributions for broad applicability. We illustrate this end-to-end with new materials inspired by leaf microstructures, showing how biological 2D structures can be used to develop novel architected materials in 2 and 3 dimensions. We further utilize a genetic algorithm to optimize generated microstructures for mechanical properties, operating directly on the latent space. This approach allows for transfer of information across manifestations using the latent space as mediator, opening new avenues for exploration of nature-inspired materials.

Shen and colleagues reported an unsupervised generative adversarial network (GAN) to identify patterns in leaves associated with superior mechanical properties and use 3D printing to build architected materials inspired by the patterns. In the future, this approach may be applied more broadly to natural materials to enable efficient algorithmic construction of structures with customized properties and form factors.

Machine learning approaches for biomolecular, biophysical, and biomaterials research

A fluent conversation with a virtual assistant, person-tailored news feeds, and deep-fake images created within seconds-all those things that have been unthinkable for a long time are now a part of our everyday lives. What these examples have in common is that they are realized by different means of machine learning (ML), a technology that has fundamentally changed many aspects of the modern world. The possibility to process enormous amount of data in multi-hierarchical, digital constructs has paved the way not only for creating intelligent systems but also for obtaining surprising new insight into many scientific problems. However, in the different areas of biosciences, which typically rely heavily on the collection of time-consuming experimental data, applying ML methods is a bit more challenging: Here, difficulties can arise from small datasets and the inherent, broad variability, and complexity associated with studying biological objects and phenomena. In this Review, we give an overview of commonly used ML algorithms (which are often referred to as "machines") and learning strategies as well as their applications in different bio-disciplines such as molecular biology, drug development, biophysics, and biomaterials science. We highlight how selected research questions from those fields were successfully translated into machine readable formats, discuss typical problems that can arise in this context, and provide an overview of how to resolve those encountered difficulties.

Predicting the Non-Deterministic Response of a Micro-Scale Mechanical Model Using Generative Adversarial Networks

Recent improvements in micro-scale material descriptions allow to build increasingly refined multiscale models in geomechanics. This often comes at the expense of computational cost which can eventually become prohibitive. Among other characteristics, the non-determinism of a micro-scale response makes its replacement by a surrogate particularly challenging. Machine Learning (ML) is a promising technique to substitute physics-based models, nevertheless existing ML algorithms for the prediction of material response do not integrate non-determinism in the learning process. Is it possible to use the numerical output of the latest micro-scale descriptions to train a ML algorithm that will then provide a response at a much lower computational cost? A series of ML algorithms with different levels of depth and supervision are trained using a data-driven approach. Gaussian Process Regression (GPR), Self-Organizing Maps (SOM) and Generative Adversarial Networks (GANs) are tested and the latter retained because of its superior results. A modified GANs with lower network depth showed good performance in the generation of failure probability maps, with good reproduction of the non-deterministic micro-scale response. The trained generator can be incorporated into existing multiscale models allowing to, at least partially, bypass the costly micro-scale computations.

Learning hidden elasticity with deep neural networks

Elastography is an imaging technique to reconstruct elasticity distributions of heterogeneous objects. Since cancerous tissues are stiffer than healthy ones, for decades, elastography has been applied to medical imaging for noninvasive cancer diagnosis. Although the conventional strain-based elastography has been deployed on ultrasound diagnostic-imaging devices, the results are prone to inaccuracies. Model-based elastography, which reconstructs elasticity distributions by solving an inverse problem in elasticity, may provide more accurate results but is often unreliable in practice due to the ill-posed nature of the inverse problem. We introduce ElastNet, a de novo elastography method combining the theory of elasticity with a deep-learning approach. With prior knowledge from the laws of physics, ElastNet can escape the performance ceiling imposed by labeled data. ElastNet uses backpropagation to learn the hidden elasticity of objects, resulting in rapid and accurate predictions. We show that ElastNet is robust when dealing with noisy or missing measurements. Moreover, it can learn probable elasticity distributions for areas even without measurements and generate elasticity images of arbitrary resolution. When both strain and elasticity distributions are given, the hidden physics in elasticity-the conditions for equilibrium-can be learned by ElastNet.

Machine Learning-Based Detection of Graphene Defects with Atomic Precision

Defects in graphene can profoundly impact its extraordinary properties, ultimately influencing the performances of graphene-based nanodevices. Methods to detect defects with atomic resolution in graphene can be technically demanding and involve complex sample preparations. An alternative approach is to observe the thermal vibration properties of the graphene sheet, which reflects defect information but in an implicit fashion. Machine learning, an emerging data-driven approach that offers solutions to learning hidden patterns from complex data, has been extensively applied in material design and discovery problems. In this paper, we propose a machine learning-based approach to detect graphene defects by discovering the hidden correlation between defect locations and thermal vibration features. Two prediction strategies are developed: an atom-based method which constructs data by atom indices, and a domain-based method which constructs data by domain discretization. Results show that while the atom-based method is capable of detecting a single-atom vacancy, the domain-based method can detect an unknown number of multiple vacancies up to atomic precision. Both methods can achieve approximately a 90% prediction accuracy on the reserved data for testing, indicating a promising extrapolation into unseen future graphene configurations. The proposed strategy offers promising solutions for the non-destructive evaluation of nanomaterials and accelerates new material discoveries.

Nano-topology optimization for materials design with atom-by-atom control

Atoms are the building blocks of matter that make up the world. To create new materials to meet some of civilization's greatest needs, it is crucial to develop a technology to design materials on the atomic and molecular scales. However, there is currently no computational approach capable of designing materials atom-by-atom. In this study, we consider the possibility of direct manipulation of individual atoms to design materials at the nanoscale using a proposed method coined "Nano-Topology Optimization". Here, we apply the proposed method to design nanostructured materials to maximize elastic properties. Results show that the performance of our optimized designs not only surpasses that of the gyroid and other triply periodic minimal surface structures, but also exceeds the theoretical maximum (Hashin-Shtrikman upper bound). The significance of the proposed method lies in a platform that allows computers to design novel materials atom-by-atom without the need of a predetermined design.

A Bayesian experimental autonomous researcher for mechanical design

While additive manufacturing (AM) has facilitated the production of complex structures, it has also highlighted the immense challenge inherent in identifying the optimum AM structure for a given application. Numerical methods are important tools for optimization, but experiment remains the gold standard for studying nonlinear, but critical, mechanical properties such as toughness. To address the vastness of AM design space and the need for experiment, we develop a Bayesian experimental autonomous researcher (BEAR) that combines Bayesian optimization and high-throughput automated experimentation. In addition to rapidly performing experiments, the BEAR leverages iterative experimentation by selecting experiments based on all available results. Using the BEAR, we explore the toughness of a parametric family of structures and observe an almost 60-fold reduction in the number of experiments needed to identify high-performing structures relative to a grid-based search. These results show the value of machine learning in experimental fields where data are sparse.

Generative Deep Neural Networks for Inverse Materials Design Using Backpropagation and Active Learning

In recent years, machine learning (ML) techniques are seen to be promising tools to discover and design novel materials. However, the lack of robust inverse design approaches to identify promising candidate materials without exploring the entire design space causes a fundamental bottleneck. A general-purpose inverse design approach is presented using generative inverse design networks. This ML-based inverse design approach uses backpropagation to calculate the analytical gradients of an objective function with respect to design variables. This inverse design approach is capable of overcoming local minima traps by using backpropagation to provide rapid calculations of gradient information and running millions of optimizations with different initial values. Furthermore, an active learning strategy is adopted in the inverse design approach to improve the performance of candidate materials and reduce the amount of training data needed to do so. Compared to passive learning, the active learning strategy is capable of generating better designs and reducing the amount of training data by at least an order-of-magnitude in the case study on composite materials. The inverse design approach is compared with conventional gradient-based topology optimization and gradient-free genetic algorithms and the pros and cons of each method are discussed when applied to materials discovery and design problems.

Recovery from mechanical degradation of graphene by defect enlargement

The extraordinary properties of graphene have made it an elite candidate for a broad range of emerging applications since its discovery. However, the introduction of structural defects during graphene production often compromises the theoretically predicted performance of graphene-based technologies to a great extent. In this study, a counterintuitive defect enlargement strategy to recover from defect-induced mechanical degradation is explored, of which the realization may lead to an enhanced operating efficiency and manufacturing feasibility. Our molecular-dynamics simulation results show that the enlargement of a preexisting defect to an elliptical shape can potentially recover from the mechanical degradation that the very defect has caused. For a defective graphene sheet having a failure strain of 48% of the pristine graphene sheet, enlarging the defect can enhance the failure strain up to 80% of the pristine graphene sheet. The mechanism of degradation recovery lies in a reduced change in curvature during deformation, which is further solidified by theoretical quantification and stress-field analysis. This theory can also predict and pinpoint the location of the initiation of the fracture-where the curvature changes most significantly during the deformation. In addition, the influence of an elliptical defect on the mechanical properties of a graphene sheet is systematically studied, which is not well understood today. Finally, the degradation recovery potential of defect of various sizes is examined, showing that the initial defect that can create the highest degree of geometric asymmetry has the best potential for degradation recovery. This study investigates the recovery from defect-induced mechanical degradation and the influence of elliptical defects on the mechanical properties of a graphene sheet, which widens our understanding of the possibility of fine-tuning mechanical properties via defect engineering and has the potential to improve materials for emerging technologies such as supercapacitor devices.