Skip to main content

Design of Emerging Material Systems

The heart of computational materials science and engineering lies in providing fundamental insights and understanding of materials behavior and properties across different scales, which further enables cost-effective design of materials with targeted properties. The significance of this topic area is highlighted by the Materials Genome Initiative (MGI) and the integrated computational materials engineering (ICME) paradigm where the central theme is inverse materials design by elucidating the link between processing, structure, properties and performance (aka PSPP links). Built on the PSPP links, our research covers three major components of design methods for emerging materials, i.e., representation, evaluation, and synthesis. For design of heterogeneous microstructural systems, we develop statistical frameworks and tools that are tailored to a wide range of material systems. Advanced machine learning (e.g., Gaussian random process and deep learning), Bayesian inference, dimension reduction, and many more techniques are employed to address challenges such as high dimensionality, lack of data, big data, mixed-variable metamodeling and optimization. For design of metamaterial systems, topology optimization and generative methods have been examined for designing multiscale and multifunctional structures. Our methods enable the integration of analyses and design decisions over multiple domains across manufacturing, structural mechanics, and design optimization.

The heart of computational materials science and engineering lies in providing fundamental insights and understanding of materials behavior and properties across different scales, which further enables cost-effective design of materials with targeted properties. The significance of this topic area is highlighted by the Materials Genome Initiative (MGI) and the integrated computational materials engineering (ICME) paradigm where the central theme is inverse materials design by elucidating the link between processing, structure, properties and performance (aka PSPP links). Built on the PSPP links, our research covers three major components of design methods for emerging materials, i.e., representation, evaluation, and synthesis. For design of heterogeneous microstructural systems, we develop statistical frameworks and tools that are tailored to a wide range of material systems. Advanced machine learning (e.g., Gaussian random process and deep learning), Bayesian inference, dimension reduction, and many more techniques are employed to address challenges such as high dimensionality, lack of data, big data, mixed-variable metamodeling and optimization. For design of metamaterial systems, topology optimization and generative methods have been examined for designing multiscale and multifunctional structures. Our methods enable the integration of analyses and design decisions over multiple domains across manufacturing, structural mechanics, and design optimization.

Jump to

Microstructure Characterization & Reconstruction (MCR)

Computational microstructure characterization and reconstruction (MCR) consists of statistical methods to quantitatively represent the microstructure (characterization), and to build an ensemble of statistically equivalent microstructure samples (reconstruction). Following the paradigm of “microstructure mediated design”, we have developed a wide range of MCR methods – correlation functions, physical descriptors, spectral density function, supervised learning, and deep learning techniques (e.g., transfer learning and generative adversarial networks). Stochasticity is a key feature across all our MCR methods. Leveraging the capabilities of MCR, we have explored PSPP linkages in several diverse material systems such as nanodielectrics for energy storage, rubber-based nanocomposites for vehicle tires, quasi-random photonic structures for light trapping, bulk heterojunction active layer in organic photovoltaics, etc.

                                                                                                                                 MCR Methods

Representative papers

Machine Learning of Materials Behavior

                                    Structural Equation Modeling

Dealing with high dimensional data is one of the major challenges in establishing microstructural PSPP relations. We have employed machine learning techniques for “feature selection” to identify key microstructure features and “feature extraction” to reduce the dimensionality of microstructure representation. For descriptor-based microstructure representation, we have employed principal component analysis (PCA), relief algorithm, and structural equation modeling (SEM) for machine learning. In our recent development, we transferred the convolutional layers of deep neural networks used for MCR to structure-property prediction, and employed adversarial networks to identify latent variables for feature extraction. In lack of data, Gaussian random process models have been used as statistical inference models for capturing PSPP relations.

Representative papers

Multiscale Uncertainty Quantification in ICME

Uncertainty is inevitably introduced in materials’ behaviors starting from the design and constituent selection stages, through the manufacturing processes, and finally during operation. For this reason, ever-growing research is being conducted to rigorously couple computational models with statistical uncertainty quantification (UQ) and uncertainty propagation (UP) methods to provide probabilistic predictions that are in line with the observed stochasticity in materials. UQ and UP are actively pursued in various fields of science and engineering. They are, however, seldom applied to multiscale simulations due to the significant computational costs and complexities.

Our goal is to devise a non-intrusive UQ and UP approach that characterizes the uncertainties via random fields (RFs) and is applicable to multiscale simulations where multiple uncertainty sources (including spatial microstructural variations) arising from different length-scales are coupled and spatially dependent. We are focusing on developing non-intrusive approaches without changing the formulations of multiscale computer models.

ICME Development of Carbon Fiber Reinforced Polymer Composites for Lightweight Vehicles

Representative papers

Data-driven Materials Design

To combat the experimental and computational costs of materials design, we are developing several data-driven methods. With parametric-based microstructure representations, e.g., the descriptor-based approach and the Spectral Density Function (SDF), we employ parametric optimization for microstructural design. Not only do those representations provide significant dimension reduction, they also offer physically meaningful PSPP mappings that are utilized to determine the optimal microstructure and improve manufacturability. Using deep learning methods such as Generative Adversarial Networks (GANs), we learn the complex morphological features and then leverage the latent variables to design the microstructure with the desired properties. For materials design with expensive computer simulations and the presence of both quantitative and qualitative design variables, we have developed a novel Latent Variable Gaussian Process (LVGP) approach that can be easily integrated with Bayesian Optimization, an effective data-driven design method, for adaptive global optimization.

                                                                Data-driven Design Explorations

Representative papers

Topology Optimization and Metamaterials Design

By determining the optimal material layout, topology optimization (TO) creates non-intuitive structures with properties superior to their conventionally-designed counterparts. Bolstered by advancements in additive manufacturing, multi-material and multiscale methods have emerged as powerful design tools for metamaterials, which are composed of micro- or mesoscale unit cells. These unit cells are tiled throughout the macroscale structure, leading to extraordinary mechanical, optical or acoustic properties. We have developed density-based, level-set, and Connected Morphable Components (CMC) TO methods for discrete, continuum and multiscale structures. Moreover, our robust shape and topology optimization (RSTO) methods account for the uncertainty inherent in manufacturing and real-world applications, i.e., variations in geometry, material or mechanical properties due to imprecise processes and random loads. Our work also addresses the challenges facing the integration of design and manufacturing, such as exploiting high dimensional design freedom while ensuring the manufacturability and connectivity of the designs.

Representative papers

Design for Additive Manufacturing

Additive manufacturing (AM) offers unprecedented design freedom but its multi-physics and multiscale nature, combined with exorbitantly expensive experiments, make it extremely challenging to ensure consistent and high quality parts. Although recent progress in AM simulations bode well for virtual testing, the high-fidelity models are still too computationally intensive for the validation of a single part, let alone iterative design and optimization. Our goal is to create data-driven computational methods and tools that quickly predict the highly location- and path-dependent properties within AM-built parts, and that capture the uncertainty in the processing conditions and materials. In particular, we are developing methods for dimension reduction and rapid prediction of the spatiotemporal evolution of the complex microstructure arising from cycles of extreme heating and cooling. Ultimately, our goal is to harness these efficient predictive models in a framework that integrates AM process and product design, assuring robustness and reliability from design to deployment.

Multiscale Topology Optimization for AM

Representative papers

List of Publications

Copyright 2024 Integrated DEsign Automation Laboratory (IDEAL) at Northwestern University. All Rights Reserved.