|
|||||
 |
|
|
|
|
|
|
|
Multiobjective
design of bone scaffolds based on active learning
Xin Wu, Fan Zhang, Yaozong Huang, Bintao Wang and Yufei Xinye |
|
Vol. 21 (2026), No. 1, 29–48
|
|
DOI: 10.2140/jomms.2026.21.29
|
Abstract |
|
The bone scaffold needs to balance mechanical properties and biological properties, but traditional methods struggle to achieve efficient collaborative optimization. This study is based on a data-driven framework integrating active learning with generative adversarial networks and finite element simulations for multiobjective design of modified face-centered cubic bone scaffolds. We address the critical challenge of simultaneously optimizing mechanical properties (elastic modulus, yield strength) and partial biological performance indicators (porosity and specific surface area) in bone scaffold design. Through a novel combination of convolutional neural networks for property prediction and variational autoencoders for structural generation, our approach explores a significantly expanded design space compared to conventional methods. The framework successfully identified Pareto-optimal solutions that achieve: (1) 11.7% increase in yield strength with only 0.55% porosity reduction and 4.1% specific surface area enhancement (Yj solution), (2) 4.78% porosity improvement with 28.5% specific surface area increase despite 19.2% strength decrease (Pj solution), and most remarkably (3) concurrent 10.1% strength enhancement and 2.24% porosity increase while maintaining near-baseline specific surface area (Cj solution) in 75% porosity systems. These results demonstrate a 69% variation range in specific surface area (3.23–5.47 m/kg), revealing that these biological indicators can be precisely tuned without necessarily compromising mechanical integrity. Experimental validation confirmed the accuracy of finite element simulations (deviation ) and demonstrated the method’s capability to precisely match target elastic modulus while overcoming inherent material limitations. This study can be used for the design of other bone scaffold structures and different bone tissues, and can efficiently explore breakthrough structures with collaborative optimization of mechanical properties and key biological indicators. |
Keywords
active learning, bone scaffolds, MFCC scaffolds,
multiobjective design, elastic modulus
|
Milestones
Received: 6 July 2025
Revised: 22 December 2025
Accepted: 5 January 2026
Published: 9 March 2026
|
Authors |
| Xin Wu | |
| Laboratory of Intelligent Control
and Robotics Shanghai University of Engineering Science Shanghai China |
|
| Fan Zhang | |
| Laboratory of Intelligent Control
and Robotics Shanghai University of Engineering Science Shanghai China |
|
| Yaozong Huang | |
| Laboratory of Intelligent Control
and Robotics Shanghai University of Engineering Science Shanghai China |
|
| Bintao Wang | |
| College of Innovation and
Entrepreneurship Shanghai University of Engineering Science Shanghai China |
|
| Yufei Xinye | |
| Laboratory of Intelligent Control
and Robotics Shanghai University of Engineering Science Shanghai China |
|
| © 2026 MSP (Mathematical Sciences Publishers). |
