Deep learning-based prediction of time-resolved adhesive forces in viscoelastic Hertzian contacts

2026-07-22 · Source: stat.ML updates on arXiv.org · Field: Science & Research — Mathematics & Computational Sciences, Engineering & Applied Sciences · Depth: Expert, extended

Summary

A deep learning model accurately predicts time-resolved adhesive forces in viscoelastic Hertzian contacts, addressing the computational cost of traditional numerical simulations. The study trained a scalar-conditioned, stateful, sequence-to-sequence deep learning model, specifically an LSTM architecture with concatenated Tabor conditioning (M1-concat), on a dataset of 12,450 BEM-generated trajectories. These trajectories span loading and unloading rates across four orders of magnitude, varied dwell times, and Tabor parameters from 0.2 to 3.2. The M1-concat model achieved a held-out mean-squared error of 5.0\times 10^{-4}, with median pull-off-force and hysteresis errors of approximately 2.2% and 1.1% respectively. Crucially, it provides a complete force trajectory prediction with a median inference time of 0.16 s, offering a speedup of about three orders of magnitude over numerical methods, making it suitable for real-time applications in soft robotics and material design.

Key takeaway

For Robotics Engineers or ML Engineers working on soft contact systems who need real-time force predictions for dynamic control or material design, this LSTM-based surrogate offers a rapid and accurate alternative to computationally expensive numerical simulations. You should consider integrating this model for applications requiring fast, repeated evaluations, such as closed-loop control of soft robotic end-effectors or iterative material property optimization, to significantly reduce development and operational times.

Key insights

Deep sequence models can rapidly predict full viscoelastic adhesive force trajectories, overcoming slow numerical simulations.

Principles

• Physics-guided input features significantly reduce the learning burden.
• Architectural simplicity can outperform complexity for this problem scale.
• Adaptive fading memory (LSTM) is well-suited for viscoelastic adhesion.

Method

A scalar-conditioned, stateful sequence-to-sequence deep learning model, specifically an LSTM with concatenated Tabor conditioning, predicts force trajectories from displacement history using a fixed-measurement-step (FMS) representation.

In practice

• Use for real-time control in soft robotic grippers.
• Accelerate iterative material design evaluations.
• Enable rapid exploration in optimization loops.

Topics

• Deep Learning
• Viscoelastic Adhesion
• Hertzian Contact
• Sequence-to-Sequence Models
• Soft Robotics
• Surrogate Models

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• Deep learning-based prediction of time-resolved adhesive forces in viscoelastic Hertzian contacts — A deep learning model accurately predicts time-resolved adhesive forces in viscoelastic contacts, enabling real-time soft robotics applications.
• Google hints at New Topological Flat Transformer — A new transformer architecture based on continuous flow dynamics promises enhanced reasoning and memory efficiency.
• Physics-Guided Dimension Reduction for Simulation-Free Operator Learning of Stiff Differential--Algebraic Systems — An extended Newton implicit layer enables simulation-free, accurate operator learning for stiff DAEs by enforcing constraints and reducing dimensions.
• Deep Learning Method for Stationary Distribution of Reflected Brownian Motion — Deep learning can accurately approximate high-dimensional Reflected Brownian Motion Laplace transforms for tail probability estimation.
• A Hybrid Conditional Diffusion-DeepONet Framework for High-Fidelity Stress Prediction in Hyperelastic Materials — Decoupling stress morphology and magnitude via a hybrid cDDPM–DeepONet framework significantly improves stress field prediction.
• Deep Learning Method for Stationary Distribution of Reflected Brownian Motion — The deep learning method accurately computes Laplace transforms for high-dimensional Reflected Brownian Motions, overcoming analytical intractability.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.