gomezmfg/digital-machining-database

Digital Machining Database (mirror)

Simulated milling data for developing and benchmarking machine-learning methods: chatter detection, stability prediction, force and vibration modeling, and more.

The original dataset is on Kaggle: tonylschmitz/digital-machining-database (version 17). Please cite and link to the Kaggle page.

This repository is an unmodified mirror of the Kaggle files (.mat, .wav and the documentation), kept so they can be downloaded file by file and loaded directly by tools such as the Machining Database Explorer. File names, folder layout and contents are identical to Kaggle. The .h5 copies on Kaggle hold the same data and are not mirrored here.

All 155 datasets are here: 185 folders, 49,806 files, about 60 GB, checked file by file against the Kaggle download.

License

CC BY-ND 4.0, the same license as on Kaggle. You may share the data with attribution; modified versions may not be redistributed. Mirrored with the authors' permission.

What is in it

155 datasets, each one set of fixed simulation inputs (tool and workpiece dynamics, cutting-force model, tool geometry, cut geometry, noise level):

Datasets	Contents
1–60	Unique input sets
61–90	Same inputs as 1–30, plus 1% Gaussian noise
91–120	Same inputs as 1–30, plus 5% Gaussian noise
121–125	Variable tooth spacing (0/93/178/270°), 5% noise
126–155	Same inputs as 1–30, longer simulation (500 revolutions), resultant force plus .wav sound

Each dataset folder holds:

• stability_boundaryN.mat: analytical stability boundary, omega_vector (rpm) and blim_vector (m).
• timeN_K.mat: one simulated cut per spindle speed and axial depth (K = grid point). Datasets 1–125 store omega, b, time and 14 signals (tool and workpiece displacement, velocity and acceleration in x and y, plus Forcex, Forcey). Datasets 126–155 store omega, b, timenew, ForceRnew.
• ForceRN_K.wav (126–155 only): the resultant force as 16-bit mono audio, peak-normalized.

All inputs for every dataset are listed in docs/Dataset descriptions.xlsx.

Layout

data/
  Dataset N mat/   stability_boundaryN.mat, timeN_K.mat
  Dataset N wav/   ForceRN_K.wav            (N = 126..155)
docs/              original PDFs, the inputs spreadsheet, stability_map_dataset_139.m

Loading in Python

from huggingface_hub import hf_hub_download
import scipy.io as sio

path = hf_hub_download("gomezmfg/digital-machining-database", "data/Dataset 1 mat/time1_34.mat", repo_type="dataset")
m = sio.loadmat(path, squeeze_me=True)
omega, b = float(m["omega"]), float(m["b"])   # rpm, m (omega is stored as uint16)
t, x, Fx = m["time"], m["xtpos"], m["Forcex"]

Sampling is about 12.4–12.5 kHz and depends slightly on spindle speed, so take the time step from time.