State estimation and ML-aided fracture mechanics for fatigue assessment of offshore structures using sparse measurements

Fatigue-induced cracks in offshore jacket structures often initiate and propagate at submerged joints, where harsh marine exposure and restricted access hinder inspection and monitoring. In practice, offshore structural monitoring relies on a limited number of sensors installed at accessible topside locations, with many joints remaining uninstrumented. To address these limitations, a hybrid framework is proposed for fracture mechanics (FM)-based fatigue assessment using sparse measurements, which integrates a state estimation scheme with FM-based surrogate modelling. Within this framework, state estimation was performed via a Kalman filter-based estimator to reconstruct structural responses such as sectional force histories at uninstrumented joints, including submerged joints. The reconstruction is performed assuming that two topside acceleration measurements are available. The estimated sectional force histories, along with initial crack front characteristics, are used as inputs to a deep neural network surrogate to predict mixed-mode stress intensity factors (SIFs) governing crack growth. These predicted SIFs are subsequently used to assess crack growth and fatigue life. In the case study, sectional forces show close agreement with numerical reference results, achieving a time response assurance criterion of 0.99. Furthermore, the SIF predictions achieve a mean absolute error of 3.19 MPa mm 1/2 , while fatigue life differs by 7–8% from the reference results.