THULab/nasa_milling

NASA Milling Wear Traces (TsFile)

This dataset is an Apache TsFile conversion of jonasmaltebecker/nasa_milling. It contains run-level metadata and six aligned sensor traces from the NASA milling tool-wear experiment. The underlying NASA source is linked from the dataset card at data.nasa.gov.

Modalities: Time-series.

Overview

• Source revision: 386c47697daa8ea1609db0f6372b5c9ddb5bd5f7
• Source file: data.parquet
• Source rows/runs: 167
• Converted observations: 1,509,360 rows
• Converted files: 2 TsFile shards
• Array lengths: 166 runs contain 9,000 samples and one run contains 15,360
• Sampling interval: 4 ms (250 Hz), as configured from the NASA milling data
• Split: train

The source Parquet has one row per run. Each run stores six same-length arrays: smcAC, smcDC, vib_table, vib_spindle, AE_table, and AE_spindle.

TsFile schema

The converted table is nasa_milling; all runs are represented as devices by the pair (case, run).

File	Rows	Size (bytes)
nasa_milling_1.tsfile	1,048,576	12,766,020
nasa_milling_2.tsfile	460,784	5,753,363

Column	Role	Type	Meaning
Time	TIME	INT64 (ms)	Sample offset 0, 4, 8, ... within each run
case	TAG	STRING	Source case identifier
run	TAG	STRING	Source run identifier
vb	FIELD	DOUBLE	Source VB tool-wear metadata; nulls are preserved
experiment_time	FIELD	INT64	Source time run-level duration (not a sample timestamp)
doc	FIELD	DOUBLE	Source DOC depth-of-cut metadata
feed	FIELD	DOUBLE	Source feed metadata
material	FIELD	INT64	Source material code
smcac, smcdc	FIELD	DOUBLE	Source smcAC and smcDC motor current traces
vib_table, vib_spindle	FIELD	DOUBLE	Table and spindle vibration traces
ae_table, ae_spindle	FIELD	DOUBLE	Source AE_table and AE_spindle acoustic-emission traces

The six sensor arrays are expanded at their common length, and each scalar sample repeats the run metadata. case and run are string TAGs even though the source stores them as integers.

Conversion notes

• The source arrays are flattened to scalar fields; no sensor samples are dropped or interpolated.
• Time is synthesized from the sample index using the 4 ms interval and restarts at zero for every (case, run) device.
• Source time describes run duration, so it is retained as experiment_time rather than incorrectly using it as the sample clock.
• The source VB column has 21 missing run values. Those nulls are repeated for the corresponding run and remain TsFile nulls.
• No source metadata columns are silently discarded. Array structure is the only representation change; the importer normalizes field identifiers to lowercase as shown in the schema table.

Read example

from tsfile import TsFileReader

path = "nasa_milling_1.tsfile"
with TsFileReader(path) as reader:
    print(reader.get_all_table_schemas().keys())
    with reader.query_table(
        "nasa_milling",
        ["smcac", "vib_spindle", "vb"],
        batch_size=4096,
    ) as result:
        batch = result.read_arrow_batch()
        if batch is not None:
            print(batch.to_pandas().head())

Source & license

• Original dataset: https://huggingface.co/datasets/jonasmaltebecker/nasa_milling
• Original NASA data: https://data.nasa.gov/Raw-Data/Milling-Wear/vjv9-9f3x/data
• Author / publisher: jonasmaltebecker; original data from NASA
• License: not declared by the original dataset; please defer to the original

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("nasa_milling_1.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())