Digital Visual Testing and Machine Learning‐Based Prediction of Fatigue Strength in Circumferential Welded Joints of Offshore Monopiles

Abstract

Circumferential welded joints in offshore monopiles are critical structural elements subjected to cyclic loading. Thus, robust inspection and fatigue life prediction methods are required to ensure structural integrity and to optimize structural design. This study presents a digital visual testing (D‐VT) methodology for the assessment of weld quality in these joints, combined with machine learning (ML) models to predict their fatigue strength. High‐resolution optical surface geometry measurements are employed to characterize weld seam geometries and to detect surface‐level discontinuities in welds. Extracted features were used to train multiple ML algorithms, including artificial neural networks and extreme gradient boosting models, with input from both visual inspection data such as weld geometry but also loading conditions. The predictive models were validated against experimental fatigue test data. Results demonstrate that D‐VT, when integrated with ML‐based analysis, can effectively identify fatigue‐prone locations along weld seams and provide reliable estimates of fatigue strength. The methodology offers a robust, non‐destructive testing solution for weld quality assessment. It both saves time and resources and can be applied to various industries and weld types.