A hybrid machine learning and physics-based model to predict the fatigue life of free-spanning pipelines

Predicting the dynamic response of a free-spanning subsea pipeline is complicated because of the intricate interactions dependent on the seabed behaviour, pipeline deformation and hydrodynamics. Traditionally, deterministic modelling techniques based on analytical, numerical and/or empirical solutions have been employed for predicting the response of subsea pipelines and inferring their fatigue damage. However, uncertainties in external loading conditions due to the stochastic nature of hydrodynamic forces pose challenges, necessitating innovative approaches for robust estimation. In this paper, we propose a novel methodology, termed Gaussian-Fourier latent force modelling (GFLFM) for stochastic pipeline fatigue analysis, which integrates Gaussian processes (GP), random Fourier features (RFF), latent force modelling (LFM), and fatigue damage assessment. This framework combines physics-based partial differential equations (PDEs) describing the system dynamics with GP regression to infer latent input forces; it is cast to address the limitations of deterministic approaches. GFLFM facilitates the accurate prediction of subsea pipeline responses while quantifying uncertainties. The proposed GFLFM framework offers a promising avenue for modelling and predicting the behaviour of free-spanning subsea pipelines. Integrating machine learning techniques with physics-based models enables more accurate predictions while accounting for uncertainties, thereby enhancing the reliability and sustainability of offshore infrastructure design and maintenance strategies.