Vibration dataset: end-mill wear monitoring on a desktop CNC at 64 kHz

Vibration recordings from three uncoated tungsten carbide end mills (3.175 mm diameter, two flute) machining pockets in 6061 aluminium on a Lunyee 3018 desktop CNC router. Two single-axis ADXL1002 MEMS accelerometers were sampled at 64 kHz per channel via Analog Devices CN0532 and CN0540 boards and Cora Z7 FPGAs, giving usable bandwidth well above the sub-2 kHz ceiling typical of low-cost tool condition monitoring. Each tool was run until the programmed 18 mm depth was reached or the collet approached the pocket wall, giving 3.6 to 4.2 hours of cutting per tool. The deposit contains the raw vibration records (14 500, 12 500 and 11 000 acquired blocks of 8192 samples per accelerometer), extracted feature tables (70 features per block), the MATLAB analysis code, and microscope images of the cutting edges before and after each test. These data support a study of whether wear-related spectral information is concentrated in the 4–8 kHz band, above what low-cost acquisition chains can resolve, and whether a health index fitted on one tool transfers to unseen tools. See the included README for acquisition parameters, feature definitions, pipeline configuration and known issues.