AI Advances Microstructure Analysis for Stronger Composites
Researchers at the University of British Columbia’s Okanagan campus have developed a streamlined method for examining the intricate architecture of fibre-based and multiscale composite materials, aiming to improve their strength and reliability under demanding loads. Leveraging materials informatics and machine learning, the team has introduced a data-driven approach to evaluate advanced woven fabric composites widely used in aerospace, construction, automotive, and sports engineering.
Composite materials with interwoven structures offer a lightweight yet robust alternative to simpler, unidirectional designs. Their complex geometries enhance performance but also present significant challenges for analysis. Dr. Abas Milani, Professor in UBC Okanagan’s School of Engineering and founding Director of the Materials and Manufacturing Research Institute, emphasized the importance of linking geometry to microstructural properties. “For example, if we want the wings of an aircraft to resist specific high shear forces, building a composite material with a particular microstructure will help us achieve that,” he said.
The research team, which includes doctoral student Tina Olfatbakhsh, applied machine learning to connect high-resolution images of fabric composite structures directly to their mechanical properties. Traditional experimental and numerical modeling techniques, while effective, are often slow, costly, and computationally intensive. Olfatbakhsh noted, “They also often assume the material geometry to be perfect, although, in the actual manufacturing process, textile composites can have many different internal complexities like waviness, voids and even fibre misalignment. This complicates matters significantly.”
To address these challenges, the team employed advanced X-ray computed tomography to capture detailed three-dimensional images of composite specimens without damaging them. This imaging reveals subtle microstructural variations that can significantly influence performance. By feeding these images into machine learning algorithms, the researchers can predict mechanical properties without exhaustive physical testing. The resulting data is stored in a growing materials database, enabling scientists worldwide to share insights and avoid redundant experiments.
Olfatbakhsh, who manages the Composite Research Network’s (CRN) Okanagan Node, described the database as a tool for targeted material selection. When specific performance characteristics are required, engineers can consult the database to identify optimal microstructural configurations. CRN itself is a collaboration between academia and industry, supporting composite innovation across Canada and internationally.
“As manufacturers develop more innovative composite materials that are formulated at the micro-scale, our testing needs to keep pace so we can ensure the integrity and strength of these new microstructures,” said Dr. Milani, principal researcher at CRN’s Okanagan Node. The use of non-destructive X-ray imaging allows for precise internal analysis without compromising the specimen, an essential capability for rare or costly materials.
The approach integrates seamlessly with existing manufacturing processes, making it practical for industrial adoption. “By streamlining the analysis using machine learning techniques, we are making great strides towards a framework for smart, data-driven design and optimization of woven fabric composites,” Olfatbakhsh explained. She highlighted the potential impact on sectors such as aerospace and transportation, where material performance directly influences safety and efficiency.
The study, published in *Composites Science and Technology* and funded by the Natural Sciences and Engineering Research Council of Canada, represents a significant step toward intelligent composite design. By combining high-resolution imaging with predictive analytics, the method bridges the gap between complex microstructural realities and the performance demands of modern engineering applications.
