# AIRISE Hack the Chatter: Anomalous Time series — Data

来源：https://www.kaggle.com/competitions/airise-image-regression-challenge-challenge/data

采集时间：2026-10-19T16:40:18Z

## Acquisition context

Contents 
This dataset contains labelled data from vibration measurements during the milling process, 
in the context of the MCCi experiment that was funded under the AIRISE project. The data 
labelling concerns the existence of chatter in the machining process.  
 
Equipment 
Data was collected using a Micromega-IAC-CM-U-03 triaxial MEMS accelerometer, interfaced via a Labjack T8 data acquisition system. All measurements were sampled at a consistent rate of 10 kHz. Vibration signals were recorded along the X and Y axes of the machine.
Two types of milling machines were used:
A robotic machining cell based on a Yaskawa GP225 industrial robot.
A Haas VF-2SS vertical machining center.
In both setups, the accelerometer was mounted on the spindle head.

## Acquisition Procedure:

All experiments involved side milling operations. In some cases, ramp coupons—where the depth of cut gradually increases—were used to capture the onset of chatter more accurately and improve labeling precision.
Chatter labeling was based on:
Visual inspection of the machined surface.
Auditory cues from the machining process.
To aid understanding, the following excerpt from Schmitz and Smith, Machining Dynamics (2nd ed., p. 243) illustrates the role of sound in chatter detection:
“Stable cuts sound different from those that chatter. Stable cuts generate sound at the tooth passing frequency, runout frequency, and their harmonics. These cuts sound 'clean' and produce 'pure' tones. Unstable cuts, however, emit sound at the chatter frequency, which is generally not a harmonic of the tooth passing or runout frequencies. As a result, they sound 'harsh' or 'raspy'—a mix of non-harmonically related frequencies. Machinists can often distinguish between good and bad cuts by sound alone.”

## Materials and Parameters

Yaskawa Robot: Machined 7075-T6 aluminum and 316L stainless steel (produced via Direct Energy Deposition).
Haas VF-2SS: Machined low-alloy steel coupons.
All operations used climb milling without coolant. Process parameters varied across experiments and are reflected in the filenames.
Each experiment corresponds to a single machining pass:
If the process remained consistent, one file was generated.
If the process transitioned (e.g., from stable to chatter), two files were created with identical parameters but different labels.

## Visualisations

Example of a stable sample:

![](assets/af41f580915d930ec28ecb36.png)

Example of a chatter sample:

![](assets/021845af2a7dc4e797c9f38e.png)

## Files

*   **train.csv** - the training set
*   **test.csv** - the test set
*   **Documentation_short** - a brief description of the acquisition protocol

## Columns

*   `id` - id of the sample point
*   `label` - either 'chatter' or 'stable', only available in train.csv 
*   `axix`- either x or y. Two axes ard provided for each sample point
*   `rpm`- spindle rotations per minute
*   `adoc_start` - angle depth of cut (start of cut)
*   `doc_end` - angle depth of cut (end of cut
*   `rdoc` - rotational depth of cut
*   `fpt` - feed per tooth (mm)
*   `0.XXXX` time since start of sampling, in seconds