ARCHIVES
VOL. 1, ISSUE 2 (2025)
Machine learning models for early detection of structural fatigue in pipelines
Authors
Ifeanyi Osagiede
Abstract
Pipeline fatigue
fractures represent a critical failure mechanism in oil and gas infrastructure,
resulting in significant economic losses and environmental hazards. Early
detection of fatigue-induced damage is essential for preventive maintenance
strategies. This study develops and validates machine learning models to
predict structural fatigue progression in welded steel pipelines using
non-destructive evaluation data. A dataset comprising 847 pipeline segments was
analyzed, including ultrasonic thickness measurements, magnetic particle
inspection records, pressure cycling history, and material composition data.
Three classification algorithms were evaluated: Random Forest, Gradient
Boosting Machines (GBM), and Support Vector Machines (SVM). Model performance
was assessed using 5-fold cross-validation with sensitivity, specificity, and
area under the receiver operating characteristic curve (AUC-ROC) as primary
metrics. The Random Forest model achieved the highest performance with 94.2%
sensitivity (95% CI: 91.8–96.1%), 96.1% specificity, and AUC-ROC of 0.981.
Feature importance analysis identified wall thickness reduction, cyclic stress
amplitude, and weld defect size as the most predictive variables. Integration
of these models with existing inspection protocols reduced false negative
fatigue predictions by 87% compared to conventional visual assessment methods.
The developed models demonstrate significant potential for improving pipeline
safety through early fatigue detection, enabling targeted maintenance interventions
before critical failure points. Implementation of this framework could prevent
approximately 73% of fatigue-related pipeline failures and generate substantial
cost savings through optimized maintenance scheduling. Future research should
focus on real-time sensor integration and model validation across diverse
pipeline materials and operational environments.
Download
Pages:17-24
How to cite this article:
Ifeanyi Osagiede "Machine learning models for early detection of structural fatigue in pipelines". World Journal of Engineering and Technology, Vol 1, Issue 2, 2025, Pages 17-24
Download Author Certificate
Please enter the email address corresponding to this article submission to download your certificate.