Efficient Fatigue Modeling: Applying Operator Networks for Stress Intensity Factor Prediction and Analysis

Tushar Gautam, Nathan Crosby, Sara Restrepo-Velasquez, Juan D Ocampo, Harry Millwater, Jacob Hochhalter, Mike Kirby, Shandian Zhe

Keywords: Stress Intensity Factors, Scientific ML, Operator Network, Crack Growth Simulation, Fatigue Modeling, Solid Mechanics, AI for Science

TL;DR: Simulating the crack growth and predicting material failure probability using accurate stress intensity factor predictions from neural operators.

Abstract:

Fatigue modeling is essential for material-related applications, including design, engineering, manufacturing, and maintenance. Central to fatigue modeling is the computation and analysis of stress intensity factors (SIFs), which model the crack-driving force and are influenced by factors such as geometry, load, crack shape, and crack size. Traditional methods are based on finite element analysis, which is computationally expensive. A common engineering practice is manually constructing handbook (surrogate) solutions, though these are limited when dealing with complex scenarios, such as intricate geometries. In this work, we reformulate SIF computation as an operator learning problem, leveraging recent advancements in data-driven operator networks to enable efficient and accurate predictions. Our results show that, when trained on a relatively small finite element dataset, operator networks --- such as Deep Operator Networks (DeepONet) and Fourier Neural Operators (FNO) --- achieve less than 5% relative error, significantly outperforming popular handbook solutions. We further demonstrate how these predictions can be integrated into crack growth simulations and used to calculate the probability of failure in small aircraft applications.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 5052