MONITORING MOORING (MONIMOOR) LINES OF FLOATING STRUCTURES USING DEEP LEARNING-BASED APPROACHES

The renewable energy sector, specifically offshore wind energy, has grown rapidly in Europe in recent years, owing to the lower energy costs. A typical Floating Offshore Wind Turbine (FOWT) system is comprised of several coupled subsystems that should jointly ensure its integrity under nominal operating conditions as well as sustainability under extreme conditions or prolonged usage. From the reliability perspective, one of the most critical subsystems is the mooring system, which keeps the platform floating in a stable condition. Eventually, monitoring the mooring lines is important to ensure the safety and serviceability of FOWT throughout its service life. This article describes a comprehensive model for assisting businesses in planning real-time monitoring of the FOWT. The proposal combines an Auto-regressive model (AR) with a Convolutional neural network (CNN) in a near real-time approach for damage detection in FOWT. The CNN-based approach monitors and subsequently identifies anomalies in the AR model coefficients of the motion prediction model apriori trained for the FOWT platform under its undamaged condition. Accordingly, a model to predict the motion of a semi-submersible FOWT platform is prepared to employ undamaged time history response (single point displacements and rotations) and the optimal AR coefficients are identified under all sea states and damage conditions. The proposed deep learning-based CNN is further employed to attribute these coefficients to different damage/health states of the platform. The effectiveness of the proposed approach is validated through numerical simulations using NREL's open-source wind turbine simulation tool OpenFAST. In the numerical model, various scenarios are simulated in an attempt to replicate real damage scenarios under varying metocean conditions while taking into account the plausible failure mechanisms in the mooring lines. The strategy of combining AR and CNN in a novelty detection-based methodology performs admirably in damage identification and classification.