Session: 02-14-03 Structural Analysis of Marine Structures 3

Submission Number: 180062

An Improved Multi-Fidelity Deep Neural Network Framework for Structural Reliability Analysis of Subsea Shear Ball Valves 

The subsea shear ball valve is a critical component of subsea test tree and light well intervention system, primarily used to sever coiled tubing, wirelines, and other structures in emergency scenarios, thereby ensuring the safety and reliability of deep-sea operations. However, the design and reliability analysis of these valves rely on costly experiments and complex numerical simulations. Traditional reliability analysis methods based on high-fidelity (HF) surrogate models require a large number of HF samples, resulting in high computational costs and low efficiency. While low-fidelity (LF) models are computationally cheaper, their accuracy is often insufficient.To address these challenges, this paper proposes an improved multi-fidelity deep neural network (MF-DNN) framework for structural reliability analysis. The framework integrates a large number of low-fidelity samples with a small amount of high-fidelity data, significantly reducing computational costs while maintaining accuracy. Notably, the framework incorporates the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) to optimize the model hyperparameters, substantially enhancing its efficiency.The effectiveness of the proposed framework is first validated through a high-dimensional nonlinear numerical case, demonstrating its higher accuracy and lower computational cost compared to traditional methods when handling complex and nonlinear problems. Subsequently, the framework is applied to an engineering case study of a subsea shear ball valve in combination with Subset Simulation, enabling efficient estimation of failure probabilities. Furthermore, a global sensitivity analysis is conducted to identify the influence of key design parameters on reliability.Compared to the conventional multi-fidelity Co-Kriging method, the improved framework demonstrates superior performance in both computational accuracy and cost control. This study presents an innovative and efficient strategy that achieves accurate failure probability assessment with low computational demand for complex engineering problems involving strong nonlinearity and rare failure modes, such as the subsea shear ball valve. The proposed approach offers a new and effective solution for structural reliability analysis of deep-sea equipment.

 

Presenting Author: Xintong Wang China University of Petroleum (East China)

Presenting Author Biography: Xintong Wang received the B.S. degree in mechanical design, manufacturing, and automation from Qingdao University of Technology, Qingdao, China, in 2022. He is currently pursuing the Ph.D. degree with the College of Mechanical and Electronic Engineering, China University of Petroleum (East China), Qingdao, China. His current research interests include structural reliability analysis and the design and manufacturing of riser-less light well intervention equipment.

Authors:

Xintong Wang China University of Petroleum (East China)
Baoping Cai China University of Petroleum (East China)
Chuntan Gao China University of Petroleum (East China)
Yinhang Zhang China University of Petroleum (East China)
Chenyushu Wang China University of Petroleum (East China)
Yuecheng Shen China University of Petroleum (East China)
Xinquan Jia China University of Petroleum (East China)