Open-Source Multi-Sensor Eddy Current Database for Non-Destructive Characterization of Carbon Fiber Structures

This dataset contains 738 multi-frequency eddy current scans of carbon fibre structures (CFRP) composite specimens, acquired with the EddyCus® Integration Kit (Fraunhofer IKTS, Dresden). All scans are provided in structured HDF5 format with full measurement metadata, material properties, and defect labels. The dataset is intended for non-destructive testing (NDT) research, machine learning benchmarking, sensor comparison studies, and fibre orientation analysis of composite materials. --- SENSORS Absolute sensors: S13131 — 7.3 MHz, 3.3 mm coil S13132 — 6.1 MHz, 5.8 mm coil S22329 — 18.1 MHz, 3.3 mm coil Differential (half-transmission) sensors: S14150 — 9.2 MHz, 5.8 mm coil, 6.3 mm spacing S14152 — 24.3 MHz, 3.3 mm coil, 3.5 mm spacing S15172 — 8.5 MHz, 2.4 mm coil, 2.4 mm spacing S17257 — 17.0 MHz, 7.0 mm coil, 7.0 mm spacing S22330 — 6.5 MHz, 2.4 mm coil, 2.4 mm spacing All sensors manufactured by Fraunhofer IKTS, Dresden. --- MATERIAL SYSTEMS The base textiles are primarily of three different types and all consist of high tenacity carbon fibers: (i) A commercially available unidirectional non-crimp fabric made off individual 50 K rovings, which are bonded togetherby a polyester stitch yarn (HP-U300/122C by HP-Textiles GmbH, Schapen, Germany). (ii) A commercially available unidirectional non-crimp fabric made off 50 K CF rovings, which were spread to form a more homogeneous fabric and bonded by an adhesive thread grid (Carbon fiber non-crimp fabric ST 50 g/m² by R\&G Faserverbundwerkstoffe GmbH, Waldenbuch, Germany). (iii) Carbon fiber non-crimp fabrics produced at the Institute of Textile Technology and High Performance MaterialTechnology (ITM) of TUD Dresden University of Technology. These were produced on the basis of 12 K fibers (HTS40 12K, Toho Tenax Europe GmbH, Wuppertal, Germany) on the institute's Malitronic Multiaxial Stitch Bonding Machine. --- DEFECT CATEGORIES Reference — no defect 84 scans Gap (missing rovings) 492 scans Layer with false orientation 80 scans PTFE film insert 24 scans Copper film insert 24 scans Copper coated roving inserted 24 scans Ondulation (fibre waviness) 6 scans Fuzzy Ball (fibre contamination) 4 scans --- HDF5 FILE STRUCTURE Each file contains four top-level groups: measurement_metadata/ — sensor type, datetime, temperature, frequency settings frequencies/ — per-channel: frequency (MHz), gain (dB), phase offset sample_properties/ — material, layup, thickness, defect type/size/depth spatial_data/ — x_mm, y_mm, z_mm, track and sample indices signal_data/fN/ — real, imaginary, complex_impedance (compound dtype) analysis_results/fN/ — magnitude, phase_degrees, phase_radians Data is compressed with gzip (level 6) + shuffle filter. Typical file size: 50–500 KB. --- SOFTWARE The open-source Python pipeline used to generate this dataset is available at:https://github.com/johannesmersch/eddycus-hdf5-converter (MIT License, v1.1.0) It provides: - EddyCus .txt → HDF5 conversion with full metadata - SQLite database for indexing and querying scans - Phase rotation for contrast optimisation - Fourier-based fibre orientation detection - Command-line interface and Python API Requirements: Python 3.9+, numpy, pandas, h5py, scipy, scikit-image, matplotlib --- RELATED PUBLICATION Mersch, J. et al. (in preparation). An Open-Source Multi-Sensor Eddy Current Database for Non-Destructive Characterization of Carbon Fiber Structures and Analysis Framework for Eddy Current Inspection of CFRP Composites. --- LICENSE Data: Creative Commons Attribution 4.0 International (CC BY 4.0)Software: MIT License