Lattice structures, featuring intricate patterns and hierarchical designs, have great potential to revolutionize various industries ranging from aerospace to biomedical engineering due to their versatility and customizability. However, the complexity of these structures and the vast design space they encompass pose significant obstacles for engineers and scientists, as traditional design exploration and optimization methods often fall short when faced with the sheer number of possibilities in the lattice design realm.
Scientists and engineers at Lawrence Livermore National Laboratory (LLNL) hope to address these long-standing challenges by incorporating machine learning (ML) and artificial intelligence to accelerate the design of lattice structures with properties such as light weight and high strength, and optimize them with unprecedented speed and efficiency.
In a recently published study, Scientific ReportsLLNL researchers have blended ML-based approaches with traditional computational techniques to usher in a new era of lattice design. By harnessing the power of ML algorithms, researchers unlock the ability to predict the mechanical performance of lattices with millions of potential design options, optimize design variables, and speed up the computational design process.
“Leveraging a machine learning-based approach to our design workflow will enable us to accelerate the design process and truly take advantage of the design freedom offered by lattice structures and exploit their diverse mechanical properties,” said lead author Aldea GĂłngora, an LLNL engineer.
“This research advances the design field as it demonstrates a viable way to integrate an iterative ML-based approach into design workflows and highlights the critical role that ML and artificial intelligence (AI) play in accelerating the design process.”
At the heart of this new research is the development of ML-based surrogate models that serve as virtual prototypes to explore the mechanical behavior of lattice structures. Trained with a wealth of data incorporating different lattice families and geometric design variables, these surrogate models show excellent predictive capabilities and can provide valuable insights into design parameters and the role that geometry and structure have on mechanical performance with greater than 95% accuracy, Gongora said.
Moreover, by incorporating an ML-based approach in the design loop, the team demonstrated that it could speed up the optimal design by exploring less than 1% of the theoretical design space size, he said.
To efficiently navigate the vast world of lattice design possibilities, researchers have turned to approaches such as Bayesian optimization, a sophisticated form of active learning that streamlines the search process by intelligently selecting and sequentially evaluating designs, reducing the number of simulations needed to find a high-performance design by a factor of five and identifying high-performance lattice configurations with remarkable speed, the researchers say.
According to the researchers, this approach not only reduces the number of simulations needed to find new designs, but also minimizes the computational burden associated with exhaustive design searches.
The research team also employed Shapley Additive Explanation (SHAP) analysis — a method used to understand how different factors or variables contribute to a particular outcome or prediction within a model — to interpret the impact of individual design variables on performance. By analyzing the impact of each parameter on the overall mechanical behavior, the researchers say they can gain a deeper understanding of the complex relationships within the design space.
The researchers said their work sets a new standard for intelligent design systems, and that the blend of computational modeling, ML algorithms and advanced optimization techniques represents a breakthrough in engineering capabilities that could improve the performance of aerospace parts and revolutionize the field of advanced materials.
GĂłngora called the research “an important advancement in demonstrating the various ways that AI can play an essential and beneficial role in materials science and manufacturing,” and said the impact goes far beyond the realm of lattice structures.
While the paper focuses on mechanical design, the researchers say the approach could be applied to a variety of design challenges that rely on expensive simulations. Given LLNL's world-class expertise in additive manufacturing, GĂłngora said a variety of lattice structures could be physically manufactured, tested, and used in cross-disciplinary applications across the lab's mission areas.
“We envision our research being widely implemented in workflows that rely on expensive simulations,” GĂłngora said. “These ML-based alternative models could be important in multi-scale design problems that rely on one or more expensive simulators. Additionally, we also envision our research being used to accelerate parametric design optimization challenges, where a scientist, engineer, or designer must consider a vast number of design parameters spanning both structures and materials.”
“By accelerating the computational design process, we can intelligently narrow down interesting and novel designs for experimental testing, which creates many opportunities for scientists to use ML tools for their scientific research and design challenges.”
LLNL co-authors include Caleb Friedman, Deirdre Newton, Timothy Yee, Zachary Doorenbos, Brian Giera, Eric Duoss, Thomas Y.-J. Han, Kyle Sullivan and Jennifer Rodriguez.
More information:
Aldair E. Gongora et al. “Accelerating the design of lattice structures using machine learning” Scientific Reports (2024). DOI: 10.1038/s41598-024-63204-7
Courtesy of Lawrence Livermore National Laboratory
Citation: Researchers Use Machine Learning to Design Advanced Lattice Structures (August 22, 2024) Retrieved August 25, 2024 from https://techxplore.com/news/2024-08-unleash-machine-advanced-lattice.html
This document is subject to copyright. It may not be reproduced without written permission, except for fair dealing for the purposes of personal study or research. The content is provided for informational purposes only.