Lattice structures, characterized by their complex patterns and hierarchical designs, offer immense potential across various industries, including automotive, aerospace, and biomedical engineering.
With their outstanding high strength-to-weight ratio, customizability, and versatility, lattice structures enable the development of lightweight, durable components that can be precisely tailored to meet specific functional requirements.
However, the complexity of the structure and the vastness of the design space encompassed by lattice structures makes it challenging for traditional methods to thoroughly explore all possible configurations and pinpoint the optimal solution for the application. With each additional design variable, the possible configurations grow exponentially, making the design space intractable.
Lawrence Livermore National Laboratory (LLNL) engineers are looking to address these challenges by harnessing the power of machine learning (ML) and artificial intelligence (AI). Advanced computational tools powered by ML and AI have enabled LLNL researchers to accelerate and enhance the optimization of lattice structure designs significantly.
In a study published by Scientific Reports, LLNL researchers detailed how they used a combination of ML algorithms and traditional methods to optimize design variables, predict mechanical performance, and accelerate the design process for lattices with millions of potential configurations.
“By leveraging machine learning-based approaches in the design workflow, we can accelerate the design process to truly leverage the design freedom afforded by lattice structures and take advantage of their diverse mechanical properties,” said lead author and LLNL engineer Aldair Gongora.
“This work advances the field of design because it demonstrates a viable way of integrating iterative ML-based approaches in the design workflow and underscores the critical role ML and artificial intelligence (AI) can play in accelerating design processes.”
The LLNL researchers used ML to tackle two main challenges in designing lattice structures. First, they developed a model that helped them understand the impact of various design choices on the lattice’s mechanical performance. Second, they created a method to efficiently identify which designs are the most effective.
At the core of the research was the creation of ML-driven surrogate models that act as digital prototypes for investigating the mechanical properties of lattice structures. These models were trained on a vast dataset that included various lattice design variables.
The surrogate models were able to deliver valuable insights into design parameters and their impact on mechanical performance. According to Gongora, the accuracy of the surrogate models exceeded 95% and enabled the researchers to optimize lattice design by exploring only 1% of the design space size.
Using Bayesian optimization and Shapley additive explanation (SHAP) analysis, the researchers efficiently explored lattice design options, reducing both computational load and the number of simulations required to identify optimal designs. They claim that their custom active-learning approach to finding optimal lattice structures required 82% fewer simulations compared to traditional grid-based search methods.
The research has set a new benchmark for intelligent design systems using computational modeling and ML algorithms. It also highlights AI’s pivotal role in designing lattice structures for a variety of applications
Looking ahead, Gongora is hopeful that his research will have an impact that goes beyond the realm of lattice structures. He believes that the approach can be applied to various design challenges, which often rely on expensive simulations.
Related Items
LLNL Could Reverse Ocean Acidification
Researchers Use Machine Learning To Optimize High-Power Laser Experiments
Generative AI to Account for 1.5% of World’s Power Consumption by 2029
