As modern ships grow larger, monitoring structural integrity becomes increasingly critical, particularly high-frequency hull-girder vibrations that accelerate structural fatigue. Since hull monitoring systems are expensive and difficult to maintain, this research explores using machine learning models based on ship motion sensors for prediction of the vertical bending moment (VBM), analysing in-service data of a 2800 TEU container ship. The tested models include LightGBM, Random Forest, XGBoost, Extra Trees, with LightGBM emerging as the best-performing and fastest framework, reinforcing its status as a state-of-the-art choice for tabular data regression tasks. Two spectral methodologies are developed: a frequency energy densities approach and a statistical-feature approach. Both consistently demonstrated that the optimal input for predicting VBM is a combination of heave, roll, and pitch motions, limited to cutoff frequency of 0.3 Hz for global wave-frequency loads, and bow acceleration, extended to 2.0 Hz to capture high-frequency hull vibrations. The report provides ship operators with a practical, cost-effective alternative to conventional strain measurement systems, predicting key VBM parameters such as the standard deviation and zero-crossing frequency, enabling fatigue analysis, critical for ensuring the structural integrity and operational safety of modern vessels. • Vertical Bending Moment (VBM) prediction using several machine learning frameworks. • Model inputs are based on in service-data from ship motion sensors. • LightGBM offers the best predictive accuracy and computational speed. • Better estimates using bow acceleration for capturing high-frequency vibrations. • Accurate estimates of VBM parameters overall, enabling direct fatigue estimations.
Mekki et al. (Tue,) studied this question.