Advanced machine learning and sensitivity analysis framework for fixed riser systems

Offshore riser systems are fundamental components in subsea oil and gas production, yet their design and performance assessment remain challenging due to nonlinear dynamics and uncertain environmental conditions. This study presents a comprehensive framework that integrates high-fidelity finite element simulations with advanced machine learning (ML) techniques to evaluate the nonlinear behavior of fixed risers under coupled hydrodynamic, current, and wind loads. A three-dimensional finite element model developed in ABAQUS/Standard captures the structural response, quantified through a von Mises stress-based performance function. To improve computational efficiency, three surrogate models—Deep Neural Network (DNN), Gaussian Process Regression (GPR), and Support Vector Regression (SVR)—were trained on data generated via a hybrid sampling strategy that combines copula-based methods with adaptive sampling. Additionally, a Polynomial Chaos-Kriging (PCK) model was employed for robust uncertainty quantification, while Sobol sensitivity analysis was conducted to identify dominant environmental and structural parameters. Results show that the DNN model achieved the highest predictive accuracy (R² = 0.987, MSE = 0.021) compared to GPR and SVR, with current velocity and buoy dimensions emerging as the most influential parameters. The adaptive sampling strategy reduced data requirements by approximately 40% while maintaining accuracy, and the PCK model provided reliable uncertainty estimates with narrower prediction bounds compared to pure GPR. The proposed framework demonstrates the potential of combining physics-based simulations with ML surrogates to enable efficient performance evaluation, enhance reliability assessment, and support resilient design optimization of offshore riser systems under uncertainty.