Discovery of Fatigue Strength Models via Feature Engineering and automated eXplainable Machine Learning applied to the welded Transverse Stiffener

TL;DR

This study addresses the challenge of predicting fatigue strength for welded transverse stiffeners beyond generalized design tables. It proposes a unified AutoML+XAI framework that combines domain-informed feature engineering with automated feature creation and SHAP-based explanations to produce accurate and interpretable predictions. Among three model hypotheses, the position-augmented $\mathcal{M}_2$ and the full-process $\mathcal{M}_3$ achieve similar generalization on the full fatigue-strength range ($R^2_{Test} \approx 0.78$, RMSE ≈ 30 MPa), while within the engineering-relevant 0–150 MPa domain $\mathcal{M}_2$ offers the best balance of accuracy and simplicity. Shapley-based analyses consistently identify stress ratio $R$, post-weld TIG dressing, and material yield strength $R_{eH}$ as key drivers, aligning with fatigue theory; the work demonstrates a practical, explainable AutoML pipeline suitable for integration into fatigue assessment workflows and digital twins.

Abstract

This research introduces a unified approach combining Automated Machine Learning (AutoML) with Explainable Artificial Intelligence (XAI) to predict fatigue strength in welded transverse stiffener details. It integrates expert-driven feature engineering with algorithmic feature creation to enhance accuracy and explainability. Based on the extensive fatigue test database regression models - gradient boosting, random forests, and neural networks - were trained using AutoML under three feature schemes: domain-informed, algorithmic, and combined. This allowed a systematic comparison of expert-based versus automated feature selection. Ensemble methods (e.g. CatBoost, LightGBM) delivered top performance. The domain-informed model $\mathcal M_2$ achieved the best balance: test RMSE $\approx$ 30.6 MPa and $R^2 \approx 0.780% over the full $Δσ_{c,50\%}$ range, and RMSE $\approx$ 13.4 MPa and $R^2 \approx 0.527% within the engineering-relevant 0 - 150 MPa domain. The denser-feature model ($\mathcal M_3$) showed minor gains during training but poorer generalization, while the simpler base-feature model ($\mathcal M_1$) performed comparably, confirming the robustness of minimalist designs. XAI methods (SHAP and feature importance) identified stress ratio $R$, stress range $Δσ_i$, yield strength $R_{eH}$, and post-weld treatment (TIG dressing vs. as-welded) as dominant predictors. Secondary geometric factors - plate width, throat thickness, stiffener height - also significantly affected fatigue life. This framework demonstrates that integrating AutoML with XAI yields accurate, interpretable, and robust fatigue strength models for welded steel structures. It bridges data-driven modeling with engineering validation, enabling AI-assisted design and assessment. Future work will explore probabilistic fatigue life modeling and integration into digital twin environments.

Figures (19)

• Figure 1: Designations of geometry parameters of transverse stiffener
• Figure 2: Research procedure.
• Figure 3: Overview of Missing Data in the Dataset
• Figure 4: Distributions of global geometric features related to plate and attachment dimensions.
• Figure 5: Distributions of mechanical strength characteristics of base and filler material.