Data-Driven Damage Quantification of Floating Offshore Wind Turbine Platforms Based on Multi-Scale Encoder–Decoder with Self-Attention Mechanism

Abstract

A Multi-Scale Convolutional Neural Network with Self Attention-based Auto Encoder–Decoder (MSCSA-AED), is a novel high-performance framework, presented here for the quantification of damage on a multibody floating offshore wind turbine (FOWT) structure. The model is equipped with similarity measurement to enhance its capability to accurately quantify damage effects from different scales of coded features using raw platform responses and without human intervention. Case studies using different damage magnitudes on tendons of a 10 MW multibody FOWT were used to examine the accuracy and reliability of the proposed model. The results showed that addition of Square Euclidean (SE) distance enhanced the MSCSA-AED model’s capability to suitably estimate the damage in structures operating in complex environments using only raw responses. Comparison of the model’s performance with other variants (DCN-AED and MSCNN-AED) used in the industry to extract the coded features from FOWT responses further demonstrated the superiority of MSCSA-AED in complex operating conditions, especially in low magnitude damage quantification, which is the hardest to quantify.