Measured 21-component geometric error maps and reconstruction code for three CNC machining centres (STAN S500, Kondia A10, HAAS VF-3)

Supplementary data and code for the study "Reconstruction of smooth CNC volumetric error fields is uncertainty-limited: a traceable three-machine study and the functional case for a tuning-free tensor-product Bézier model." The dataset contains the complete measured 21-component geometric error maps of three structurally different three-axis CNC vertical machining centres — STAN S500 GRAFIT (Russia), Kondia A-10 (Spain) and HAAS VF-3 (USA) — acquired with a Renishaw XL-80 laser interferometer, together with Leica LTD800 laser-tracker validation points measured on a large reference cube of the STAN machine. It also provides the complete Python code that assembles the measured component errors into a dense three-component volumetric error field through the ISO 230-1 rigid-body kinematic model (50 mm functional-point offset) and reconstructs the field from sparse nodes by six methods: trilinear interpolation, inverse-distance weighting, thin-plate radial basis functions, Gaussian-process regression, tensor-product cubic interpolation and a tensor-product Bézier model. Running the provided scripts reproduces the central quantitative results of the paper — the held-out reconstruction residuals, the bootstrap 95% confidence intervals and the paired Wilcoxon p-values reported in Table 3 and Figures 5–6 — for all three machines and the node-density sweep. All residuals lie far below the U = 34 µm (k = 2) measurement screening envelope. Contents: data/ — three measured error maps (STAN_S500_GRAFIT.xlsx, Kondia_A10.xlsx, HAAS_VF3.xlsx) and tracker data (tracker_BigCube.xlsx); code/ — core model and baselines (core.py), data loaders (load_stan.py, load_haas_kondia.py) and the benchmark script (reconstruct.py); plus README.md and requirements.txt.