Autoregressive pde model rollouts improved by event triggered corrections
Stabilizing Autoregressive PDE Foundation Model Rollouts with Event-Triggered Context Healing
Computational Engineering, Finance, and Science
Summary
Predicting fluid flow using computer models often becomes less accurate the longer the model runs because small errors build up. The authors propose a method called Event-Triggered Context Healing (ETCH), which checks the model’s predictions against a few real measurements and fixes the predictions when they differ too much. This correction uses a special neural network to rebuild the flow context, greatly reducing errors while needing fixes only about 15% of the time. Their approach allows fast and more reliable predictions without having to correct every step.
What this means in practice
- •For computational fluid dynamics teams: Maintain accurate long-term velocity and pressure predictions by correcting model rollouts only when sensor data significantly deviates.
- •For industrial simulation engineers: Enable faster full-field fluid simulations that remain stable and accurate through event-triggered corrections using sparse physical feedback.
Authors
Jaewan Park, Jay Phil Yoo, Kazuma Kobayashi, Seid Koric, Syed Bahauddin Alam, Iwona Jasiuk, Souvik Chakraborty, Diab W. Abueidda
Abstract
Autoregressive PDE foundation models enable fast full-field forecasting but accumulate errors as predicted states are recursively reused. We introduce Event-Triggered Context Healing (ETCH), which compares forecasts with measurements from 12 cylinder-surface pressure taps. When their discrepancy exceeds a threshold, pressure-conditioned hourglass diffusion transformers reconstruct the velocity--pressure context for subsequent predictions. On 43 RealPDEBench Cylinder trajectories, ETCH reduces velocity and pressure errors from 48.05 and 63.71% to 6.29 and 7.25% for DPOT, and from 74.97 and 76.91% to 5.16 and 6.84% for Poseidon-T, using the same reconstructors and threshold. For DPOT, correcting 14.56% of windows maintains errors close to reconstructing every window and supports average real-time forecasting and correction on an A100 GPU. These results show that sparse physical feedback stabilizes long autoregressive rollouts across PDE foundation models while reducing reconstruction cost.