Experimental data and supporting materials for concrete permittivity characterization using a large-aperture OECP and MLP inversion
Description
This dataset supports the study “Real Relative Permittivity of Concrete Using a Large-Aperture Open-Ended Coaxial Probe and Simulation-Trained Multilayer Perceptron Inversion.” The study developed a dielectric characterization framework using a 100 mm open-ended coaxial probe (OECP) compatible with standard 100 × 200 mm concrete cylinders. Complex reflection coefficients were measured over 0.7–1.3 GHz and converted into real relative permittivity using a simulation-trained multilayer perceptron (MLP) inversion approach. The repository contains three principal categories of research materials. First, the raw experimental data cover Exp. 1–3. Exp. 1 examined changes in mass and estimated real relative permittivity in three normal-strength concrete cylinders during accelerated drying. Exp. 2 examined nine high-strength concrete cylinders from different casting batches under ambient curing conditions and at multiple measurement ages. Exp. 3 compared the OECP-based estimate with a ground-penetrating radar result obtained from one prismatic concrete block as an electromagnetic consistency cross-check. Second, the processed-data files contain the numerical values underlying the figures presented in the associated article, including specimen-level measurements, normalized mass, estimated real relative permittivity, repeated acquisitions, summary statistics, and data used to construct the plots. Each specimen-level OECP estimate was based on three consecutive acquisitions performed without disturbing the probe–specimen contact. These repetitions characterize short-term acquisition stability and do not represent contact-repositioning or inter-operator reproducibility. Partial data available at 6 h in Exp. 1 are retained in the raw dataset for completeness but were excluded from the specimen-level permittivity comparisons because they did not have the same complete sampling structure as the other drying times. Third, the repository provides the MATLAB code used for the MLP component of the study, including the construction of frequency-local input features and the training of regression models for dielectric-property estimation. The code should be interpreted together with the methodological description and parameter definitions provided in the associated article. The GPR result in Exp. 3 was obtained from a single concrete block and should therefore be interpreted as a one-block electromagnetic cross-check rather than independent ground-truth validation.
Files
Institutions
- Korea Institute of Civil Engineering and Building TechnologyGyeonggi-do, Goyang-si