Deep-Learning-based damage detection of an experimentally tested full-scale wind turbine blade.

Reducing the operation and maintenance (O&M) costs of wind farms can foster adoption of environmentally sustainable and cost-effective energy sources. O&M costs in Wind Turbines can account up to 35% of the turbine’s total levelized cost per KWh, with a large portion attributed to that for wind turbine blades (WTB). In the present study, an experimental campaign for fatigue damage progression monitoring is presented. After subjecting the WTB to cyclic loading, an artificial transverse crack was introduced in the trailing edge (TE).The dynamic response of the blade to three different excitation schemes-white noise, white noise with the inclusion of a fundamental sinusoidal component and a frequency sweep-is obtained via the use of accelerometers, strain gauges and a real-time deflection monitoring system. These tests yielded generated 4 different health states, the initial healthy and 3 damaged states. The progressive damage is presented using frequency response functions (FRFs),while a simple deep-learning damage detection framework is developed. A conventional autoencoder (AE) is trained using the Frequency Response Functions (FRFs) corresponding to the blade’s healthy state; then, an unsupervised damage detection algorithm is formulated using the AE. © 2024, NDT. net GmbH and Co. KG. All rights reserved.