Data and code for assessing the incremental predictive value of Windsor probe testing beyond SonReb in an earthquake-damaged reinforced concrete building

Description

This dataset supports the study “Incremental predictive value of Windsor probe testing beyond SonReb in an earthquake-damaged reinforced concrete building”. The data comprise 20 paired column locations across nine floor groups in a single reinforced concrete building in Türkiye, anonymised as Building A. Variables include field indirect ultrasonic pulse velocity, mean Schmidt rebound number, mean exposed Windsor probe length, and corrected individual reference compressive strength derived from core tests. The dataset and accompanying analysis code evaluate whether Windsor probe measurements provide incremental predictive information beyond SonReb. Ordinary least squares and power-law models form the primary analyses; nested ridge regression provides a sensitivity analysis. Leave-one-location-out and leave-one-floor-out cross-validation assess prediction within the same building and do not constitute external validation. The files provide the analysis data and reproducibility materials. Windsor values are location-level means of three shots; individual shot lengths were not retained. Individual UPV transit times and transducer distances were not recorded. Reference strengths are already corrected and must not be corrected again. The findings are specific to this dataset and evaluation framework. They do not provide design strength values or establish structural safety.

Steps to reproduce

1. Download the supplementary data workbook and the reproducibility package, and extract the ZIP archive. 2. Follow the included execution instructions and install the specified dependencies. 3. Use the supplied final dataset, preserving the 20 location identifiers and nine floor groups. Reference compressive strengths are already corrected; do not apply additional corrections. 4. Run the supplied analysis code to reproduce the ordinary least squares, power-law and nested ridge evaluations under leave-one-location-out and leave-one-floor-out cross-validation. 5. Compare the regenerated predictions and model comparison metrics with the supplied outputs. Model fitting, standardisation, tuning and logarithmic back-transformation correction must remain within the relevant training folds. 6. Editing the Excel workbook does not automatically retrain the models. Any modified analysis requires rerunning the code with the intended input data.

Institutions

• Malatya Turgut Özal Üniversitesi

  Malatya, Battalgazi

Categories

Licence