The peridynamic (PD) method has been widely utilized in the numerical modelling of ice–structure interactions due to its capability to naturally capture material failure and fracture evolution. PD formulations can be categorized into bond-based and state-based models. While the bond-based model offers computational simplicity, it inherently restricts Poisson’s ratio to 1/4 in three-dimensional (3D) simulations and 1/3 in two-dimensional (2D) simulations. In contrast, the state-based model allows for arbitrary Poisson’s ratios, which is essential for accurately modelling ice mechanics, as Poisson’s ratio of ice commonly exceeds 0.33 and can reach up to 0.42. This study employs the PD method coupled with the discrete energy release rate criterion to analyze the influence of Poisson’s ratio on ice failure mechanisms and ice-induced loads in two typical scenarios: cylindrical impact on an ice disc and ice–structure interaction with a propeller blade. The benchmark compression test yielded computed Poisson’s ratios of 0.215, 0.258, 0.333, and 0.401 for prescribed values of 0.2, 0.25, 0.33, and 0.40, corresponding to relative errors of 7.5%, 3.2%, 1.02%, and 0.38%, respectively. And the fracture simulations indicate that variations in Poisson’s ratio affects ice fracture patterns and load distributions, underscoring the limitations of the bond-based PD model in accurately representing ice–structure interactions. These findings highlight the necessity of adopting state-based PD formulations for improved numerical predictions of ice-induced mechanical responses.

Offshore wind turbines operate under harsh and time-varying marine environments, leading to significant distribution shifts between source and target domains. These challenges are further intensified by limited fault samples, imbalanced class distributions, and difficult maintenance access in maritime scenarios. To address these issues, this paper proposes a multi-source domain adaptation method, termed the class-wise dynamic anchor weighting fusion framework (CDAWF), for offshore wind turbine drivetrain fault diagnosis. CDAWF dynamically assigns weights to different source domains according to their feature-level similarity to the target domain, enabling adaptive feature fusion across operating conditions. Convolutional neural networks are used to extract domain-specific representations, while adversarial training is introduced to reduce cross-domain discrepancies. The proposed method is validated on a multi-condition bearing fault dataset generated from high-fidelity simulations of a 5 MW offshore wind turbine drivetrain. Experimental results show that CDAWF effectively improves diagnostic robustness under domain shift, limited labeled data, and class imbalance. Compared with benchmark methods, CDAWF achieves superior classification performance and serves as a useful reference for cross-condition fault diagnosis of offshore wind turbine drivetrain systems.