Related Experiment Video
Updated: Jun 19, 2026
06:14
Multiscale Structures Aggregated by Imprinted Nanofibers for Functional Surfaces
Published on: September 11, 2018
6.5K
Machine learning-based optimal design of fibrillar adhesives
Mohammad Shojaeifard1, Matteo Ferraresso1, Alessandro Lucantonio2
1Mechanical Engineering Department, University of British Columbia, Vancouver, BC V6T1Z4, Canada.
Journal of the Royal Society, Interface
|February 25, 2025
Summary
We developed a machine learning tool to optimize fibril compliance for stronger adhesion. This approach enhances the design of bio-inspired adhesives and micro-architected materials for improved fracture resistance.
Area of Science:
- Biomimetics and Materials Science
- Robotics and Engineering
Background:
- Fibrillar adhesion in nature (beetles, spiders, geckos) uses micro/nanofibrils for enhanced adhesion via 'contact splitting'.
- This principle inspires engineering applications in robotics, transportation, and medicine, with functional grading of fibril properties showing potential for improved adhesion.
- Previous designs focused on simplified geometries, and no prior work has optimized fibril-array scale using machine learning.
Purpose of the Study:
- To introduce a machine learning (ML) based tool for optimizing fibril compliance distribution to maximize adhesive strength.
- To demonstrate the tool's capability in recovering known design results for simple geometries and generating novel solutions for complex configurations.
- To accelerate the design process and reduce errors in creating high-performance fibrillar adhesives.
Main Methods:
- Utilized a two-neural network (NN) system: a predictor NN to estimate adhesive strength from compliance distributions and a designer NN for gradient-based optimization.
- Applied ML to optimize the distribution of fibril compliance across an array.
- Validated the approach on both simple and complex geometric configurations.
Main Results:
- The ML tool successfully recovered previous design outcomes for simplified geometries.
- Novel and optimized compliance distributions were identified for complex configurations, leading to enhanced adhesive strength.
- The method significantly reduced test error and accelerated the optimization process compared to traditional approaches.
Conclusions:
- The proposed ML-based tool offers a high-performance solution for designing advanced fibrillar adhesives.
- This approach enables the creation of optimized micro-architected materials for applications requiring superior fracture resistance through equal load sharing.
- Machine learning provides an efficient pathway for tackling complex design challenges in bio-inspired adhesion.
