Hybrid DeepONet-Transolver Framework for PET Bottle Buckling Surrogates

Technology
In development
University
Scouted & summarized by Halo AI

A high-performance AI-physics surrogate framework for rapid PET bottle buckling analysis. By combining DeepONet and Transolver, it replaces computationally intensive finite element analysis with fast, accurate predictions of spatial displacements and temporal reaction forces, accelerating packaging design and multi-physics exploration.

Overview

The Hybrid DeepONet-Transolver framework is an advanced AI-physics surrogate model designed to predict the non-linear buckling behavior of PET bottles under top-load compression. By integrating deep operator networks (DeepONet) with attention-based transformer solvers (Transolver), this solution offers a rapid, highly scalable alternative to traditional, computationally expensive finite element analysis (FEA). It enables engineers to accelerate packaging design exploration by simultaneously outputting static spatial displacement fields and dynamic time-evolving reaction force histories, effectively bypassing the long simulation times associated with conventional numerical methods.

Technical specifications

This framework employs a dual-component architecture tailored for complex, non-parametric geometric domains:

  • Transolver Component: Uses an attention-based mechanism to process non-Euclidean geometries. By representing mesh data as point clouds, it maps local geometry into a latent space to discover consistent physical behaviors across irregularly shaped surfaces, enabling accurate prediction of nodal displacement fields.
  • DeepONet Component: Models the temporal evolution of reaction forces. The branch-trunk network architecture ingests geometry-specific latent representations to infer the full reaction force trajectory as a function of both bottle shape and simulation progress.
  • Performance: The model achieves mean relative L2 errors of 2.5–13% for displacement fields and approximately 2.4% for dynamic reaction force curves, maintaining high accuracy even across diverse bottle geometries.
Technology readiness level

This technology is currently at a prototype demonstration stage (TRL 4-6). It has been validated through research studies using datasets of over 250 unique bottle designs, demonstrating accurate prediction of physical phenomena such as buckling behavior. The underlying methodologies, including DeepONet, are...

Generated Jul 8, 2026. Please verify key information.

About Brown University

Brown University is a private Ivy League research university in Providence with broad disciplinary breadth and a culture of open, collaborative inquiry. Industry engagement is enabled by shared core facilities and co-located labs, structured mechanisms for corporate-sponsored research, and an urban innovation district within walking distance of campus. Clinical translation is supported through the medical school’s network of affiliated hospitals across the region. Research is backed by competitive federal funding from NIH, NSF, DoD, and other agencies, alongside state and industry partnerships. A dedicated technology transfer and corporate engagement office supports IP strategy, licensing, sponsored research agreements, and startup formation.

Halo home
Partner smarter. Move faster.
Get new partnering requests
delivered to your inbox.