Open Access

Data for the Paper: Audio Signal-Based Defect Detection for Wind Turbine Rotor Blades Using an Autoencoder

Files

test_data.zip (424.44 MB)

validation_data.zip (70 MB)

unaugmented_data.zip (50.93 MB)

noise_data.zip (1.96 MB)

raw_measurement_data.zip (560.26 MB)

Full item page

Date

2026-05-26

Type

Sound

Authors

Meding, Finn

Feldmann, Robert

Description

This folder contains all the data used for the paper: R. Feldmann, F. Meding, and T. Melz, "Audio Signal-Based Defect Detection for Wind Turbine Rotor Blades Using an Autoencoder," in Proceedings of ISMA2026 - International Conference on Noise and Vibration Engineering, Leuven, Belgium, 2026. This dataset contains: acoustic measurement data for the training of neural networks to detect defects in structures like wind turbine rotor blades. Recorded in 2026 in the semi-anechoic chamber at the Technical University of Darmstadt (SAM), the data comprises individual acoustic emission signals (.wav) obtained via local acoustic resonance spectroscopy across a 48-point grid on eight glass fibre reinforced polymer (GRP) test plates (360 mm * 280 mm). While plates 1–4 represent fully intact structures subsequently subjected to impact damage measurements, plates 5–8 feature specifically introduced delamination defects. The dataset includes: both raw emissions and a pre-split structure (70% training, 15% validation, 15% test) tailored for neural network training. Augmented subsets feature environmental background noise (drone, traffic, birds, wind, wind turbine) augmented at Signal-to-Noise Ratios (SNR) from 9 dB to 3 dB. A standardized file-naming convention directly embeds structural parameters, mounting types, spatial coordinates, noise types, and SNR values. Details and the use of the dataset can be found in the readme.