DOE Final report for FAIR Data and Interpretable AI Framework for Architectured Metamaterials LC Brinson, Duke University, PI C Rudin, Duke University, co-I C Daraio, Caltech, co-I DOE DE-SC0021358, 9/23/2020 – 9/22/2025 1. Abstract This research program established a transformative framework for the discovery and design of mechanical metamaterials, which are architected structures engineered to control physical phenomena like sound and vibration in ways natural materials cannot. To overcome the traditional reliance on trial-and-error, the project developed an interpretable Artificial Intelligence (AI) framework that moves beyond "black box" models to reveal the specific geometric patterns—such as "unit-cell templates"—that govern a material’s performance. A major breakthrough was the development of a hierarchical design method, which allows a single material to block vibrations across multiple frequency ranges simultaneously by layering patterns at different scales without them interfering with one another. This was further expanded to include irregular, graph-based designs that use spanning tree algorithms to ensure structural connectivity while allowing for customized, direction-dependent properties like stiffness and acoustic impedance. Beyond design, the project addressed the practicalities of real-world production by developing uncertainty quantification techniques that account for manufacturing defects and material variability, reducing the need for expensive physical testing by orders of magnitude. To speed up the discovery process, the team implemented Gaussian Process Regression and other surrogate models that provide accurate performance predictions at a fraction of the traditional computational cost. The AI-generated designs were successfully validated through fabrication of physical samples and wave propagation experiments, confirming their ability to accurately guide or reflect waves as predicted. By contributing these tools and high-quality FAIR benchmark datasets to the wider scientific community, this work provides a scalable foundation for advancing technologies in aerospace vibration control, medical imaging, and noise reduction. 2. Brief technical description of findings and outcomes. The project established a comprehensive framework for the discovery, design, and characterization of mechanical metamaterials by integrating interpretable machine learning (ML) with physics-based simulations and experimental validation. The primary hypothesis driving the work was that hidden, discoverable patterns exist within material structures that can be distilled into simple, transferable rules to govern the relationship between geometry and bulk physical properties. To address this, the team proposed the Learning Refined Compositional Rules (LRCR) framework, a coarse-to-fine sampling method designed to bridge the gap between length scales while overcoming the exponential computational costs associated with high-resolution material design. Key Computational Approaches and Interpretable ML: A cornerstone of the project was the development and application of the Generalized Optimal Sparse Decision Trees (GOSDT) algorithm and the shape-frequency feature representation. Unlike traditional "black box" neural networks, which often performed poorly on these scientific datasets, GOSDT provided transparent if-then rules that achieved over 97% accuracy in predicting bandgaps—frequency ranges where vibrations are blocked—using simple geometric motifs like "unit-cell templates". The team extended this to the Unit Cell Template framework, which distinguishes between "fixed" regions necessary for performance and "free" regions that allow for design flexibility, such as weight reduction through "void" inclusions. High-Speed Surrogate Modeling for Dispersion Relations: To address the immense computational cost of traditional physics simulations, the team developed a specialized Gaussian Process Regression (GPR) surrogate model. This model acts as a high-speed approximation for computing dispersion relations—the mathematical map of how waves travel through a material. Notably, this GPR model is resolution-independent and representation-invariant, meaning it can learn from simple designs and accurately predict the performance of complex, finely detailed structures nearly eight times faster than traditional methods. Unlike standard AI, it also provides uncertainty estimates and is differentiable, making it a powerful tool for gradient-based design optimization. Hierarchical Design, Irregular Materials, and Uncertainty: The research achieved significant breakthroughs in multiscale engineering and manufacturing reliability: ●​ Hierarchical Scaling: The project demonstrated a novel multiscale technique where coarse-scale bandgap objectives remained robust even when fine-scale features were introduced. This "scale-separation effect" allowed for the design of materials capable of blocking vibrations across multiple, distinct frequency ranges simultaneously. ●​ Graph-Based Irregular Designs: To move beyond periodic grids, the team implemented a framework using graph algorithms and spanning trees. This ensured structural interconnectivity in irregular designs, facilitating customized anisotropy (direction-dependent properties) while guaranteeing manufacturability. ●​ Uncertainty Quantification (UQ): Recognizing the sensitivity of metamaterials to manufacturing defects, the team used Polynomial Chaos Expansion to develop surrogate models. This approach reduced the sampling requirements for characterizing material performance by two to three orders of magnitude compared to traditional brute-force testing. Experimental Validation and FAIR Data: The project results were rigorously validated through physical sample generation and laser Doppler vibrometry (LDV) experiments. Physical tests of hierarchical steel structures confirmed that numerically predicted bandgaps were accurately realized in real-world finite structures. Finally, to support the broader scientific community, the project developed the MetaMine portal and an extensible metamaterials ontology to ensure all generated data is Findable, Accessible, Interoperable, and Reusable (FAIR). Significant datasets were also contributed to the UCI Machine Learning Repository to serve as benchmarks for future AI research in materials science. 3. External presentations The PIs made a number of external presentations at conferences and universities presenting parts of this research. The most notable include: ●​ H Zhang, C Rudin, J Guilleminot, LC Brinson, Multi input, multi output neural operator learning for acoustic metamaterials, USNCCM, Chicago IL, 20-23 July 2025 ●​ LC Brinson, Gordon Research Conference on Multifunctional Materials and Structures, Ventura CA, AI for Metamaterials: Design, Manufacturing and Data, 28 Jan – 2 Feb 2024 ●​ LC Brinson, DOE ASCR Meeting, FAIR Data and Interpretable AI Framework for Architectured Metamaterials, Atlanta, GA, 5-7 Feb 2024 ●​ LC Brinson, SES annual conference, A robust framework to find the hidden patterns in acoustic and mechanical metamaterials, Minnesota, 8-11 October 2023. ●​ B Feng, A C Ogren, C Daraio, K L Bouman, "Visual Vibration Tomography: Estimating Interior Material Properties from Monocular Video", Computer Vision and Pattern Recognition 2022, Accepted, 2022. ●​ LC Brinson, M Bastawrous, Z Chen, C Rudin, ", Invited talk: Using AI to Unlock Nature’s Secrets to Design Mechanical Metamaterials,, May 8‐12, 2022.", Materials Research Society Meeting, Published, 2022. ●​ C Daraio, A Ogren, "Irregular architected materials with programmable properties, Plenary talk", 18th European Mechanics of Materials Conference (EMMC), Published, 2022. ●​ M Bastawrous, LC Brinson, Z Chen, C Rudin, "Designing 2D mechanical metamaterials with printability constraints using interpretable machine learning.,,", American Physical Society March Meeting, Published, 2022. ●​ A Ogren, C Daraio, "Gaussian Process Regression as a Surrogate Model for Dispersion Computations", The Minerals, Metals, and Materials Society (TMS), Published, 2022. ●​ LC Brinson, M Bastawrous, C Rudin, Z Chen, ", Mechanical Metamaterials: Structure, Properties and Data, keynote presentation", Duke Soft Matter Symposium, Published, 2021. ●​ LC Brinson, "Creating an Organized, Searchable Data Repository on Next Generation Materials", Bloch Symposium, presented as keynote, 2021. ●​ LC Brinson, EMMC (European Materials Modeling Council), Working toward Interoperability: Nanomine to Metamine and Ontologies, live presentation in virtual event, 2-4 March 2021. ●​ Additionally, all the conference papers listed in the section below were presented at the respective conferences listed. 4. Personnel Involvement. The following personnel were involved in the work over the course of the project: Post docs: ●​ Mary Bastorous, Duke University ●​ Rayehe Karimi-Mahabadi, Duke University ●​ Marc Palmeric, Duke University Grad students: ●​ Zhi Chen, Duke University ●​ Yingfan Wang, Duke University ●​ Chudi Zhong, Duke University ●​ Han Zhang, Duke University ●​ Alex Ogren, Caltech ●​ Zhicheng Guo, Duke University ●​ Jiachang Liu, Duke University ●​ Jon Donnelly, Duke University ●​ Rui Zhang, Duke University Undergraduates: Donggyun Kong, Duke University 5. Unexpended funds. $4425.14 6. Listing of all published articles acknowledging support by the award. Published Journal Articles ●​ Bastawrous, M. V., Chen, Z., Ogren, A. C., Daraio, C., Rudin, C., & Brinson, L. C. (2025). "A multiscale design method using interpretable machine learning for phononic materials with closely interacting scales." Computer Methods in Applied Mechanics and Engineering, 440, 117833; doi.org/10.1016/j.cma.2025.117833. ●​ Zhang, H., Karimi Mahabadi, R., Rudin, C., Guilleminot, J., & Brinson, L. C. (2024). "Uncertainty quantification of acoustic metamaterial bandgaps with stochastic material properties and geometric defects." Computers and Structures, 305, 107511; 10.1016/j.compstruc.2024.107511. ●​ Karimi Mahabadi, R., Chen, Z., Ogren, A. C., Zhang, H., Daraio, C., Rudin, C., & Brinson, L. C. (2025). "Graph-based design of irregular metamaterials." International Journal of Mechanical Sciences, 295, 110203; 10.1016/j.ijmecsci.2025.110203. ●​ Chen, Z., Ogren, A., Daraio, C., Brinson, L. C., & Rudin, C. (2022). "How to see hidden patterns in metamaterials with interpretable machine learning." Extreme Mechanics Letters, 57, 101895; doi.org/10.1016/j.eml.2022.101895. Conference Papers and Proceedings ●​ McTavish, H., Zhong, C., Achermann, R., Karimalis, I., Chen, J., Rudin, C., & Seltzer, M. (2022). "Fast Sparse Decision Tree Optimization via Reference Ensembles." Proceedings of the AAAI Conference on Artificial Intelligence 2022. ●​ Feng, B., Ogren, A. C., Daraio, C., & Bouman, K. L. (2022). "Visual Vibration Tomography: Estimating Interior Material Properties from Monocular Video." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2022. ●​ “Exploring the Whole Rashomon Set of Sparse Decision Trees”, Advances in neural information processing systems, 2022, proceedings.neurips.cc, https://arxiv.org/abs/2209.08040 ●​ “Exploring and Interacting with the Set of Good Sparse Generalized Additive Models”, Advances in neural information processing systems, 2023, proceedings.neurips.cc, https://arxiv.org/abs/2303.16047 ●​ “FasterRisk: Fast and Accurate Interpretable Risk Scores, Advances in Neural Information Processing Systems”, 2022, proceedings.neurips.cc, https://arxiv.org/abs/2210.05846 ●​ “Fast Sparse Decision Tree Optimization via Reference Ensembles”, Proceedings of the AAAI conference on artificial intelligence, 2022, ojs.aaai.org, https://doi.org/10.48550/arXiv.2112.00798 ●​ “Interpretable Image Classification with Adaptive Prototype-based Vision Transformers – ProtoViT”, Advances in Neural Information Processing Systems, 2024, proceedings.neurips.cc, https://arxiv.org/abs/2410.20722 ●​ “FastSurvival: Hidden Computational Blessings in Training Cox Proportional Hazards Models”, Advances in Neural Information Processing Systems, 2024, proceedings.neurips.cc, https://arxiv.org/abs/2410.19081 ●​ “Interpretable Generalized Additive Models for Datasets with Missing Values”, Advances in Neural Information Processing Systems, 2024, proceedings.neurips.cc, https://arxiv.org/abs/2412.02646