A domain-adaptive deep learning-empowered integrative framework for condition monitoring of the underwater mooring system under unseen wave conditions

Abstract

Mooring system failures continue to be a leading cause of incidents involving floating platforms, underscoring the critical need for effective condition monitoring. Although deep learning methods have been widely adopted, their reliability is often compromised by the uncertain and unpredictable nature of the offshore environment. This study introduces a comprehensive framework that integrates an experimentally validated fluid-structure interaction (FSI) model with a domain-adaptive deep learning approach to detect mooring system damage under wave condition variability. First, a finite element (FE) model is developed using potential flow theory and Morison's equation to simulate the dynamic responses of floating platforms. The accuracy of the model in capturing fluid-structure interactions is validated through experimental wave tank tests. Second, a new domain-adaptive deep learning model is proposed, centered around a modified ResNet18 backbone and incorporating tailored Domain-Adversarial Neural Networks (DANN) with PAC-Bayesian regularization. This model is trained on a dataset generated from physical analyses to accurately predict mooring system damage under unseen wave conditions. On a separate target-domain evaluation set, the proposed model attains mean (Formula presented) of 0.8188/0.9960/0.8697 for mooring lines 1–3, surpassing a ResNet18 baseline (0.6189/0.9948/0.8227) with markedly lower variability (represented by standard deviation 0.0102/0.0008/0.0099 vs. 0.0605/0.0016/0.0237), under 10-fold cross-validation analysis. The results demonstrate that the proposed methodology significantly improves predictive accuracy and generalization in complex, stochastic offshore environments. This demonstrates the potential for practical condition monitoring of mooring systems in floating infrastructure.