Presentations

Abstract

"The structural integrity of wind turbine blades (WTBs) is governed by adhesive bonded composite joints, particularly along trailing-edge connections where complex mixed-mode (peel and shear) loading conditions prevail. Extending the service life of ageing wind turbine blades requires reliable identification of early-stage damage in such critical joints, as this directly influences maintenance planning and repair strategies. In this context, early detection enables informed lifetime-extension decisions. Under continuous severe loading, these joints are susceptible to progressive damage mechanisms such as cohesive failure and interfacial debonding. Such damage often initiates locally and remains undetected until significant stiffness or strength degradation occurs, limiting timely maintenance. As the timing and quality of maintenance actions directly influence service life, early detection, quantification of damage progression, and accurate diagnosis are essential for informed maintenance and lifetime management. Moreover, failure thresholds derived from early damage detection can serve as key inputs for modelling stiffness degradation and estimating remaining useful life. This study presents a hybrid experimental–numerical framework for early damage detection in composite bonded joints representative of wind turbine blade structures. Glass fiber reinforced polymer (GFRP) adhesively bonded joint specimens, analogous to trailing-edge connections, are investigated under mixed-mode loading conditions. The specimens consist of two GFRP adherends with a symmetric laminate configuration bonded using epoxy adhesive. Static tests are conducted using a guided loading fixture that enables combined peel and shear loading. Full-field strain measurements are obtained using digital image correlation (DIC), providing insight into strain localisation and damage evolution. Multiple adhesive thicknesses are examined, with several specimens tested for each configuration to capture variability in joint behaviour. Finite element models are developed in Abaqus/CAE to represent the bonded specimens, with cohesive zone formulations used to capture damage initiation and propagation within the adhesive layer. The models are validated against experimental force–displacement responses and selected strain fields and subsequently used to generate strain data for further analysis. In the data preparation stage, experimental strain fields are processed to obtain time histories. Strain data from experiments and numerical simulations are combined to form a hybrid dataset. The novelty lies in enriching experimental DIC measurements with finite element–generated strain fields, increasing data availability and improving representation of damage evolution. Deep learning models are applied to this dataset, including a feature-based multi-layer perceptron (MLP) using engineered strain descriptors and a sequence-based long short-term memory (LSTM) network that learns from strain time-series data. The MLP serves as a baseline for capturing nonlinear relationships, while the LSTM models temporal dependencies and progressive damage evolution, enabling comparison of feature-based and sequence-based approaches.Both models classify the structural state into no damage, mild damage, and severe damage. Preliminary results demonstrate the capability to identify early damage initiation and track its progression. The LSTM shows improved sensitivity to temporal strain evolution compared to the MLP, enabling earlier detection of damage onset. While this work focuses on early damage detection and classification, it provides a foundation for future prognostic development; specifically, remaining useful life (RUL) prediction will be addressed in subsequent work. The extracted damage indicators will be used to model damage evolution and support timely maintenance and targeted repair of bonded joints before critical conditions, contributing to lifetime-extension strategies for WTBs."