Corrosion estimation in floating offshore wind turbine mooring with particle filter coupled with DeepONet predictor

Mooring lines of floating offshore wind turbines (FOWTs) are continuously subjected to harsh environmental loading, making them highly susceptible to fatigue damage, while corrosion further accelerates degradation and may cause premature failure. Conventional monitoring methods rely on underwater sensors, which are costly to deploy and maintain, and more prone to failure than sensors mounted above the floating platform. Alternatively, platform motion responses can be exploited to infer underlying corrosion states; however, existing approaches often depend on complex finite element models that are computationally prohibitive for real-time applications. This study introduces a novel Bayesian filtering framework that integrates a data-efficient Deep Operator Network (DeepONet) surrogate for corrosion estimation in FOWT mooring systems. Instead of direct finite element simulations, DeepONet parameterizes mooring health through physically interpretable corrosion descriptors (diameter loss) and maps them to platform motion responses. Unlike conventional black-box neural networks, DeepONet learns nonlinear operators acting on time-series inputs, enabling the capture of complex spatiotemporal dependencies with reduced training data requirements. Within this framework, DeepONet serves as a high-fidelity surrogate for real-time inference inside a particle filter, which assimilates observed platform motions to recursively estimate latent corrosion parameters. DeepONet is trained on synthetic time-series data generated from OpenFAST simulations of the NREL 5MW OC4 platform with systematic corrosion scenarios. The proposed PF-DeepONet framework enables robust, rapid, and computationally efficient inverse estimation of mooring line health under realistic offshore excitations, offering a cost-effective alternative to traditional sensor-based monitoring.