Session: 04-01-05 Flexible Pipes V

Submission Number: 177170

Physics-Informed Neural Network Approach for Predicting Dynamic Responses and Fatigue in Flexible Risers 

Flexible risers are especially susceptible to fatigue due to mainly complex and stochastic wave loading. One of the most reliable approaches for fatigue evaluation is the time-domain fatigue assessment (TDFA), which computes damage by simulating the time history of loads and structural responses, directly determining stress cycles (by a stress cycle counting approach, such as Rainflow) and applying Palmgren–Miner damage accumulation rule. Leveraging finite element analysis, TDFA captures nonlinearities and transient effects, providing greater accuracy than other methods, such as frequency-domain methods, but at the expense of a significantly higher computational cost. Several studies have explored data-driven surrogate models for diminishing the computer burden of TDFA, with artificial neural networks (ANNs) showing a very strong potential.
This work aims to develop a neural network that predicts the dynamic responses at the top of a free-hanging flexible riser across all sea states for a given wave incidence direction and under various offset conditions. The proposed approach utilizes standard sea-state parameters (e.g., significant wave height, HS, and zero-crossing period, TZ) in conjunction with offset information to generate accurate axial force and bending moments time series. To enforce physical fidelity, governing-equation constraints and boundary conditions are embedded within the loss function following the Physics-Informed Neural Networks (PINNs) paradigm. Predictions are benchmarked against high-fidelity FEM simulations, and complete fatigue analyses of the tensile armors are performed using both computed responses (ANN and FEM).
Results demonstrate high predictive accuracy, robust generalization across sea states and offsets, and substantial reductions in computational cost. These findings indicate that the proposed ML-PINN framework can expedite riser fatigue evaluations while preserving engineering accuracy.

Presenting Author: Thiago Camargo Rodrigues Federal University of Rio de Janeiro

Presenting Author Biography: Thiago Rodrigues holds a Bachelor’s degree in Civil Engineering (2019) and a Master’s degree in Civil Engineering (2022), both from the Federal University of Rio de Janeiro (UFRJ). He is currently pursuing a Ph.D. in Civil Engineering at COPPE/UFRJ. His main areas of expertise comprise Structural and Offshore Systems Engineering, Machine Learning and scientific programming.

Authors:

Thiago Camargo Rodrigues Federal University of Rio de Janeiro
Thiago Ângelo Gonçalves De Lacerda Federal University of Rio de Janeiro
Luis Volnei Sudati Sagrilo Federal University of Rio de Janeiro
George Carneiro Campello Petrobras
Rodolfo Figueira De Souza Petrobras
Tiago Brun Coser Petrobras