Crashbox surrogate study — OpenRadioss data and code

Description

Data and code for "Data requirement and physics-constraint diagnosis for neural surrogates of axially crushed thin-walled tubes". * decks/ input of the 24 production and 2 mesh-study analyses: <case>.k (keyword-format starter input, read by the OpenRadioss starter) and <case>_0001.rad (engine input) * logs/ starter and engine output of each analysis; how each run ended is tabulated in results/tables/p345_co_termination_census_v1.csv * time_history/ interface impulse and energy histories (T01), from which every mean crushing force is computed * results/ case-level metrics, learning-curve, exponent and constraint-campaign tables * code/analysis/ extraction (p366), verification gates and figure scripts * code/ml/ surrogate and physics-constraint training pipeline How the analyses ended ---------------------- Every analysis runs to a velocity-dependent end time chosen so that the stroke exceeds the longest measurement window (133.07 mm at t = 4 mm). In 16 of the 24 production analyses the engine stops earlier with "ERROR: NODAL VELOCITY IS TOO HIGH FOR INTERFACE": late in the crush, when folds stack against each other and against the plates, individual contact nodes move faster than the interface check allows within one time step, and the engine stops rather than risk a missed contact. Every such stop occurs after the impactor has passed the end of the longest window of that thickness (smallest margin 19.1 mm, results/tables/p345_co_termination_census_v1.csv). All metrics use the histories only up to the end of their window, and in explicit time integration the state at a given time depends only on earlier steps, so the late stop does not affect any reported value. Every file is listed with its SHA-256 in MANIFEST.sha256. Reproduce the figures: see code/analysis/run_CO5.bat (paths at the top of each script).

Steps to reproduce

1. Install Python 3.10 with numpy, scipy, pandas and matplotlib. 2. Put code/analysis and code/ml on the path, and set the root path at the top of each script to the folder where this dataset is unpacked. 3. Extract the crushing metrics from the time histories: python p366_cn_mcf_extract_v1p3.py 4. Reproduce the campaigns: p370 (constraint campaign), p371 (strength sweep), p372 (channel experiment), p378 (component decomposition), p91_learning_curve.py (learning curves). 5. Verify and draw: run code/analysis/run_CO5.bat, which runs the gates (p163, p168, p173) and then the figure scripts (r1, r2, r3, r4, p340). Each gate checks the data against the values quoted in the paper and stops before writing if a check fails. 6. Re-running the finite element analyses requires OpenRadioss (Windows 64-bit, build of 28 July 2026): starter on decks/<case>/<case>.k, then engine on decks/<case>/<case>_0001.rad.

Institutions

• Pusan National University

  Busan, Busan

Categories

Licence