The global installed capacity of floating offshore wind turbines is projected to increase by at least 100 times over the next decades. Station-keeping of floating offshore renewable energy devices is achieved through the use of mooring systems. Mooring systems are exposed to a variety of environmental and operational conditions that cause corrosion, abrasion, and fatigue. Regular physical in-service inspections of mooring systems are the golden standard for monitoring their health status. This approach is often expensive, inefficient, and unsafe, and for this reason, researchers are focusing on developing tools for digital solutions for real-time monitoring. Floating offshore renewable energy devices are usually equipped with a wide range of sensors, some low-cost, low/zero maintenance, and easily deployable (, accelerometers on the tower), contrary to others (, direct tension mooring line measurements), producing real-time data streams. In this paper, we propose exploiting the data coming from the first type of sensors for mooring systems health status nowcasting. In particular, we will first rely on state-of-the-art supervised shallow and deep learning models for predicting the health status of the different sections of the mooring lines. Then, since these supervised models require types and amount of data that are seldom available, we will propose new shallow and deep weekly supervised models that require a very small amount of data regarding worn mooring lines. Results will show that these last models can potentially have practical applicability and impact for real-time monitoring of mooring systems in the near future. In order to support our statements, we will make use of data generated with a state-of-the-art digital twin of the mooring system, OrcaFlex, for a floating offshore wind turbine reproducing the physical mechanism of the mooring degradation under different loads and environmental conditions. Results will show errors around 1% in the simplest scenario and errors around 4% in the most challenging one, confirming the potentiality of the proposed approaches.

中文翻译：

---

浮式海上风力发电机系泊线段健康状态临近预报：从监督浅层学习到弱监督深度学习

未来几十年，全球浮动式海上风力涡轮机装机容量预计将增加至少 100 倍。浮动海上可再生能源设备的位置保持是通过使用系泊系统来实现的。系泊系统暴露在各种环境和操作条件下，导致腐蚀、磨损和疲劳。定期对系泊系统进行实际使用中检查是监测其健康状况的黄金标准。这种方法通常昂贵、低效且不安全，因此，研究人员正在专注于开发用于实时监控的数字解决方案的工具。漂浮式海上可再生能源设备通常配备各种传感器，其中一些传感器成本低、低/零维护且易于部署（塔上的加速度计），与其他设备相反（直接张力系泊线测量），产生真实的-时间数据流。在本文中，我们建议利用来自第一类传感器的数据来进行系泊系统健康状态即时预报。特别是，我们将首先依靠最先进的监督浅层和深度学习模型来预测系泊线不同部分的健康状况。然后，由于这些监督模型需要很少可用的数据类型和数量，我们将提出新的浅层和深层每周监督模型，这些模型需要非常少量的有关磨损系泊线的数据。结果将表明，最后这些模型可能在不久的将来对系泊系统的实时监控具有实际适用性和影响。为了支持我们的陈述，我们将利用最先进的系泊系统 OrcaFlex 数字孪生生成的数据，为浮动式海上风力涡轮机再现不同负载下系泊退化的物理机制，环境条件。结果显示，在最简单的场景中误差约为 1%，在最具挑战性的场景中误差约为 4%，从而证实了所提出方法的潜力。