REAL-ADHERE · Real-time Adhesion Trajectories & Inverse Design via Physics-Enhanced Machine Learning

Viscoelastic adhesive interfaces with tunable properties are central to emerging technologies in robotics, biomedicine, and smart materials. However, their design and control are severely constrained by the slow computational cost of classical contact mechanics and empirical models, which cannot deliver real-time predictions across complex, viscoelastic, and non-smooth surfaces. This project introduces a Physics-Enhanced Machine Learning (PEML) framework for the real-time design, prediction, and inverse optimization of tunable adhesive interfaces. The approach integrates high-fidelity simulations with advanced learning architectures, spanning Neural Controlled Differential Equations (NCDEs) and Neural Operators, while embedding physical principles such as viscoelastic constitutive laws, energy balances, and adhesion models. This hybrid strategy ensures physically consistent and interpretable results while leveraging acceleration of ML inference for design. Furthermore, the project advances beyond forward modeling by employing generative design for the inverse problem, allowing the tailored engineering of adhesive surfaces under physical and manufacturing constraints. Prototype fabrication and validation will bridge the gap between simulation and experiment, ensuring demonstrable feasibility. The expected outcomes include: (i) the first real-time, physics-consistent tool for adhesive interface dynamics, (ii) design strategies generalizable to non-smooth and multi-asperity surfaces, and (iii) transformative applications in adaptive gripping, biomedical devices, and switchable adhesion systems. The project thus addresses urgent scientific and technological challenges by combining fundamental physics, cutting-edge ML, and experimental validation to pave the way for sustainable and high-impact adhesion technologies.