SIFBench: An Extensive Benchmark for Fatigue Analysis

Tushar Gautam, Robert M. Kirby, Jacob Hochhalter, Shandian Zhe · 2025

Read SIFBench: An Extensive Benchmark for Fatigue Analysis by Tushar Gautam, Robert M. Kirby, Jacob Hochhalter, Shandian Zhe (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-01173.txt.

Introduction

Tushar Gautam and co-authors launch in 2025 SIFBench, a large open data set for fatigue crack work. Crack growth is a top cause of failure in planes, bridges, cars, and power plants. A key value that guides crack spread must be known to judge safe life.

The data set holds more than five million crack and part shapes from high-grade computer models across thirty seven scenes. A shared Python tool lets users fetch and shape the data with ease. The authors test four common model types and share code for training and scoring. The goal is to lower the bar for new work in this field.

Catalog record

CREATORTushar Gautam, Robert M. Kirby, Jacob Hochhalter, Shandian Zhe
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE46 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-01173

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

SIFBench: An Extensive Benchmark for Fatigue Analysis Tushar Gautam, Robert M. Kirby, Jacob Hochhalter, Shandian Zhe Abstract Fatigue-induced crack growth is a leading cause of structural failure across critical industries such as aerospace, civil engineering, automotive, and energy. Accurate prediction of stress intensity factors (SIFs) — the key parameters governing crack propagation in linear elastic fracture mechanics — is essential for assessing fatigue life and ensuring structural integrity. While machine learning (ML) has shown great promise in SIF prediction, its advancement has been severely limited by the lack of rich, transparent, well-organized, and high-quality datasets. To address this gap, we introduce SIFBench, an open-source, large-scale benchmark database designed to support ML-based SIF prediction. SIFBench contains over 5 million different crack and component geometries derived from high-fidelity finite element simulations across 37 distinct scenarios, and provides a unified Python interface for seamless data access and customization. We report baseline results using a range of popular ML models — including random forests, support vector machines, feedforward neural networks, and Fourier neural operators — alongside comprehensive evaluation metrics and template code for model training, validation, and assessment. By offering a standardized and scalable resource, SIFBench substantially lowers the entry barrier and fosters the development and application of ML methods in damage tolerance design and predictive maintenance. 1 Motivation Fatigue is a leading cause of failure in engineering structures across numerous critical industries, such as aerospace ( Aliabadi et al., 1987 ) , civil engineering ( Albrecht and Yamada, 1977 ) , automotive ( Dehning et al., 2017 ) , and energy ( Callister Jr and Rethwisch, 2020 ) . Components subjected to repeated or cyclic loading can develop cracks that propagate over time, even when the applied stresses are below the material’s yield strength. This process, known as fatigue crack growth , can lead to sudden and catastrophic failure without prior warning. Ensuring the safety, reliability, and longevity of such structures hinges critically on the ability to accurately predict and manage the initiation and propagation of fatigue cracks. The central parameter for predicting fatigue behavior is the stress intensity factor (SIF), a fundamental quantity in linear elastic fracture mechanics that describes the stress and strain fields near a crack tip ( Newman, 2000 ) . The SIF quantifies the severity of the stress concentration and is directly linked to the crack driving force ( Irwin, 1957 ) . It also determines the fracture mode — Mode I (opening), Mode II (in-plane shear), or Mode III (out-of-plane shear) — and, in the context of fatigue, serves as the primary variable governing crack growth under cyclic loading, as described by Paris’ Law: \frac{da}{dN}=C(\Delta K)^{m} where K denotes the SIF. Accurate determination of the SIF for a given crack geometry and loading condition is therefore indispensable for predicting a component’s remaining fatigue life, conducting damage tolerance assessments, and designing against fracture. While analytical solutions for SIFs exist for simple geometries and loading conditions, real-world engineering structures often feature complex shapes, multiple cracks, and intricate boundary conditions. To handle such complexities, numerical methods, particularly the Finite Element Method (FEM), are widely used to compute SIFs ( Dixon and Pook, 1969 ) with high accuracy. However, high-fidelity FEM simulations are often computationally expensive and time-consuming, making them impractical for large-scale structures or applications requiring rapid evaluation, such as design optimization or real-time structural health monitoring. Although experimental techniques can offer valuable insights, they are typically costly and limited in applicability due to constraints related to equipment, testing environments, and the range of accessible parameters. As a result, engineers and researchers frequently rely on handbook solutions manually crafted by domain experts. These solutions serve as efficient surrogate models, offering reasonably accurate estimates of SIFs. For instance, the classical Raju-Newman ( Newman Jr and Raju, 1981 ; Raju and Newman, 1979 ) and Fawaz and Andersson, 2004 equations, expressed as polynomial functions, are capable of predicting SIFs for simple crack geometries such as semi-elliptical surface cracks and quarter-elliptical corner cracks near holes. Similarly, Pommier et al., 1999 provided a set of engineering formulas for Mode I surface cracks, while Wang and Lambert, 1995 constructed local weight functions to compute SIFs for semi-elliptical surface cracks in finite-thickness plates. However, these handbook solutions are confined to simple geometries and loading conditions. These limitations highlight the pressing need for more flexible, general, and automated methods that can accommodate a broader range of geometric variations and complex loading scenarios — without requiring tedious and nontrivial manual derivations ( Gautam et al., 2025 ) . Machine learning (ML) has recently emerged as a promising approach for SIF prediction. ML models, such as deep neural networks, offer great flexibility to handle complex geometries, boundary conditions, and loading scenarios through automated training, while also demonstrating improved accuracy ( Zhang et al., 2023 ) . Notable recent contributions include the work of Xia et al., 2022 , who employed a convolutional neural network (CNN) to predict Mode I SIFs in coal rocks; and Xu et al., 2022 , who explored Gaussian process (GP) regression ( Williams and Rasmussen, 2006 ) , tree-based models, and feed-forward neural networks for probabilistic failure risk assessment in aeroengine disks. Their hybrid model improved SIF prediction accuracy by 5%–35% compared to the widely adopted universal weight function ( Glinka and Shen, 1991 ) . In addition, Zhang et al., 2023 used a multi-layer feed-forward neural network to predict mixed-mode SIFs in composite materials. However, ML approaches are data-driven and their success depends critically on the availability of rich, high-quality training data ( Merrell et al., 2024 ) . Given analytical solutions are limited and experimental data is scarce and costly to collect, the training datasets are mainly generated using high-fidelity FEM ( Galić et al., 2018 ) . However, generating accurate FEM data for SIFs is far from trivial ( Liu et al., 2004 ) . It demands domain expertise in meshing, particularly near crack tips where stress singularities arise. Accurately capturing the stress field in these regions often requires extremely fine meshes or specialized elements, e.g. , quarter-point singular elements. The computation can be highly costly and time-consuming ( Bathe, 2006 ) , especially when dealing with complex 3D geometries or conducting parametric studies varying crack dimensions and locations. Furthermore, extracting SIFs from FEM results involves sophisticated post-processing techniques, e.g. , ( Courtin et al., 2005 ; Fu et al., 2012 ; Hou et al., 2022 ) , which adds another layer of complexity and potential for error. Collectively, the high cost, required expertise, and time-consuming nature of this process create a substantial barrier to advancing ML methods in this domain. To address these challenges, we introduce: • SIFBench — an open-source, large-scale benchmark database designed to support the development and evaluation of ML methods for SIF prediction and fatigue analysis. The database contains an extensive collection of high-fidelity FEM-derived SIF solutions, covering approximately 5 million unique cracks, geometries, and loading conditions across 37 distinct scenarios. • A unified and convenient Python interface — enabling users to easily access and customize the datasets tailored to specific use cases. • Baseline ML models –— including random forests, support vector regression, feed-forward neural networks, and Fourier neural operator — applied to the datasets to

Read the complete text