Zhaoxu Meng of Clemson University was studying how cigarette smoking and other factors can contribute to deterioration of the tendon-bone connection when he started thinking about a broader engineering problem.
Nature had developed an effective way to join tendons, a soft material, with bones, a hard one. Could engineers learn from it and use the same approach elsewhere?
Meng, an associate professor of mechanical engineering, started digging deeper and now the results are going public.
“It’s been two years of effort, and we are very excited,” he said. “The big takeaway is that nature has already solved a problem engineers still struggle with: how to connect soft and hard materials without making the interface the place where failure is most likely to start.”
Meng and his first Ph.D. student, Zhangke Yang, now a postdoctoral fellow at Johns Hopkins University, reported their findings on Sept. 29 in the journal Advanced Intelligent Systems.
What they found could eventually help build better connections wherever soft and hard materials need to work together, including in prosthetics, orthopedic implants, energy-storage devices and “soft robots” made partly from flexible materials, Meng said.
The core challenge the researchers targeted is that stress tends to concentrate where soft and hard materials meet, making the interface more likely to fail, Meng said.
In the tendon-bone connection, the tissue gradually becomes more mineralized, creating a smooth transition from soft to hard.
That gradual change in mineral content, combined with changes in the tissue’s internal structure, helps spread out stress instead of concentrating it at one vulnerable spot, researchers found.
The researchers used machine learning to build on what nature had already figured out and identify new material designs based on what makes the tendon-bone connection so effective. The approach helped speed up their work.
Detailed computer simulations can take several minutes for a single design, but their machine-learning model, once trained, could evaluate a new design in about half a second, making it possible to explore far more possibilities, Meng said.
An important finding was that the optimized designs depended on more than how gradually the material changed from soft to hard, Meng said. They also depended on how the tiny collagen strands inside the material were arranged and oriented, he said.
“One important lesson is that strong interfaces may require more than a simple one-direction gradient,” Meng said. “Our model and AI optimization revealed coordinated patterns, including gradients across different directions and local collagen strand alignment near curved regions.
“In simple terms, nature does not just make the material gradually harder. It also carefully controls the internal architecture so forces can flow more smoothly across the connection.”
Meng said he is continuing to advance the research, looking to eventually move from computer simulations to real-world application.
“The exciting part is that this framework can work in tandem with advanced fabrication methods, such as 3D printing or spatially programmable material synthesis,” he said. “Instead of relying only on trial and error, we can use AI and simulation to narrow down promising material designs before making and testing them. That could make the process of developing stronger, more durable, and more functional materials much more efficient.”
The work was hailed as a creative approach to a long-standing challenge by Alexander Leonessa, chair of the Department of Mechanical Engineering, and Laine Mears, director of the School of Mechanical and Automotive Engineering.
“Research like this reflects Clemson’s growing strength at the intersection of artificial intelligence, advanced materials and engineering,” they said in a joint statement. “It also shows the impact of investing in talented faculty and students who are advancing discovery, developing new technologies and preparing for leadership in the fields that will shape the future.”
The title of the paper is “Machine Learning-Enabled Field Prediction and Inverse Design of Functionally Graded Materials Inspired by Tendon-Bone Enthesis.” Meng served as corresponding author.
Acknowledgment: The research referenced in the first paragraph of this story was supported through the NIH-funded South Carolina Translational Research Improving Musculoskeletal Health Center (SC TRIMH).
