Finite-element endpoint dataset for SDCM–geogrid-reinforced high-speed railway embankments under hydro-climatic scenarios
Description
This dataset contains 810 records of hydro-mechanical endpoint responses for high-speed railway embankments reinforced with stiffened deep cement mixing (SDCM) columns and a geogrid. The records are reported outputs from a PLAXIS 2D finite-element campaign and cover 162 configurations, each evaluated under five prescribed hydro-climatic states (C0–C4). The configurations combine two soft-soil thicknesses and three levels each of SDCM column diameter, embankment height, equivalent uniformly distributed operating load and column length. Each record contains eight predictors and two responses. The predictors describe the five configuration variables, event groundwater-level change, event duration and the retained stiffness ratio of the affected upper fill. The responses are the signed event settlement increment, ΔS_e (mm), and the spatial maximum absolute excess pore pressure at the event endpoint, P_abs,e (kPa). Settlement is positive downward, and negative increments indicate relative upward displacement between the pre-event and event endpoints. P_abs,e does not retain pressure sign and is not a maximum over the full event time history. The release provides the endpoint table in CSV and Excel formats, a data dictionary, scenario definitions, matched-configuration identifiers, source-row mapping and validation records. All 810 records and their stored numerical values are retained. The three environmental predictors form five joint sensitivity scenarios, without site calibration or assigned occurrence probabilities. C0 is a one-day reference rather than a duration-matched control for the other states. For evaluation on unseen configurations, all five records belonging to one configuration should remain in the same data partition. The dataset supports analysis and surrogate-model benchmarking within the sampled configuration and scenario space. Native finite-element projects, complete input and extraction scripts, response histories and recovery outputs are not included. Exact response extraction locations are not specified in the source workbook. The package therefore supports reproducible analysis of the endpoint table, but does not by itself permit independent regeneration or verification of the underlying simulations.
Files
Steps to reproduce
1. Download the files, keeping the supplied directory structure. Read README.md, metadata/data_dictionary.csv and metadata/scenario_definitions.csv before importing the data. 2. Import data/endpoint_data.csv as a UTF-8, comma-delimited table with a header row and decimal points. The Data worksheet in data/SDCM_Geogrid_Endpoint_Dataset.xlsx contains the same records. 3. Use the eight predictor columns, in this order: Hs_m, Dp_m, He_m, q_kPa, Lp_m, delta_gwl_event_m, event_duration_days, upper_fill_R_E. The two response columns are delta_S_event_mm and P_abs_endpoint_kPa. Exclude record_id, design_id and state_id from the numerical feature matrix. 4. Check the package with Python 3 by running python code/validate_dataset.py from the package root. The script uses the Python standard library and verifies file checksums, the 810-record structure, the complete 162-configuration design, scenario coverage, finite numeric values and CSV–Excel agreement. Results of the curation audit are supplied in provenance/validation_report.json. 5. For grouped benchmark evaluation, partition by design_id and retain its five states together. Estimate any scaling, transformations and model hyperparameters using only the appropriate training folds. No manuscript-specific split assignments or trained models are included in this data release. 6. Consult the accompanying article for the reported modeling procedure and benchmark methods. Regenerating the finite-element responses would require the original model setup, input specifications and extraction procedure, which are not supplied in this package.