Special Section on Data-Driven Mechanics and Digital Twins for Ocean Engineering

This Special Section issue focuses on the topic of Data-Driven Mechanics and Digital Twins for Ocean Engineering. Two categories of papers are included in this section that deals with (i) reduced-order modeling and data analytics and (ii) data-driven computing and digital twins. In the first category, Yin et al. presented the modal analysis of hydrodynamic forces in flow-induced vibrations using dynamic mode decomposition (DMD). Using snapshots of the flow field, spatio-temporal evolution characteristics of the wake patterns are analyzed. The dominant DMD modes with their corresponding frequencies are identified and used to reconstruct the flow fields. In another paper in this category, Janocha et al. presented a 3D large eddy simulation and data-driven analysis of the flow around a flexibly mounted cylinder via proper orthogonal decomposition (POD) analysis. The POD-based modal extractions are performed on slices in the wake to identify the coherent structure in the flow. Vortex shedding modes are analyzed and classified by examining three-dimensional wake flow structures. Such a body of work is useful for building reduced-order (surrogate) models that can be considered for multiquery analysis, design optimization, and feedback control. However, these POD/DMD studies are restricted to linear physics as well as to idealized canonical geometries. There is a need for further extension to large-scale marine and offshore structures (e.g., offshore wind turbines, marine risers, and pipelines). Moreover, projection-based POD/DMD techniques generally face difficulties to scale for highly nonlinear turbulent flow. Nonlinear model reduction and deep neural networks (e.g., convolutional autoencoders) are possible alternatives to be explored for advanced reduced-order modeling.In the second category, advancements in data-driven methods and machine learning toward the development of physics-based digital twins are sought. Essentially, this category focuses on the integration aspects of AI/ML and data analytics with sensor technology, the Internet of things, cloud computing, etc. Toward this aim, Mehlan et al. explored a novel virtual sensor concept for online load monitoring and life estimation of a wind turbine gearbox. By combining data from a condition monitoring system (CMS) and a supervisory control and data acquisition (SCADA) system, the authors demonstrate the efficacy of their framework for the estimation of remaining useful life and fatigue damage. This is an excellent demonstration of the digital twin framework relying on data, a virtual model, and decision support for structural health monitoring. State-of-the-art techniques such as the use of Kalman filters and a least-squares estimator are used for predicting loads. With regard to the physical modeling, the authors used an aero-hydro-elastic solver along with the multibody simulation. There is potential to improve the physical accuracy of the mechanics data by considering computational fluid dynamics and other coupled mechanics effects. Moreover, some of the linearized state predictions can be improved by nonlinear data-driven model reduction and deep learning techniques, as mentioned earlier. Finally, this special issue includes a technical brief by Panda and Warrior, which uses machine learning for the flow field over an axisymmetric body of revolution. State-of-the-art ML models namely random forest, artificial neural network (ANN), and convolutional neural network (CNN) algorithms are considered. The Reynolds stress transport model (RSTM)-based simulations are used to generate data for training and testing of ML-based surrogate models. The authors report promising results for predicting the flow field beyond the training range using the CNN- and ANN-based surrogate models. As pointed out by the authors, the work needs further investigation for a broad range of Reynolds numbers and geometry parameters to evaluate the generalization capabilities of the ML-based surrogate models.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.