Supported Data

Exact supported layouts and limitations.

Vnuli 0.1 supports documented engineering time-series layouts. A supported file extension does not mean every file with that extension is supported.

At a glance

CSV

Supported CSV files have a header row, a first column named exactly time, finite strictly increasing time values in seconds, and numeric channel columns.

time,pressure,temperature,vibration
0.000,10.2,75.1,0.03
0.001,10.4,75.2,0.04

• One file is one dataset, one run run_001, and one timebase.
• CSV timestamps are preserved as explicit irregular time. Vnuli does not infer a regular sample rate from evenly spaced rows.
• Units are always unknown for this layout.
• Missing time, blank channel values, duplicate headers, non-finite time, non-monotonic time, pivoted data, and non-numeric channels are not silently accepted.
• Bounded requests scan rows incrementally and collect only requested channels and rows in the requested time window.

HDF5

Supported HDF5 files use the documented Vnuli hierarchy and declare vnuli_layout="timeseries.v1" at the root.

/
  runs/
    <run_id>/
      timebases/
        <timebase_id>/
          time
          channels/
            <channel_id>

• Each /runs/<run_id> is one run.
• Each timebase group is one canonical timebase.
• Regular timebases use start_seconds and sample_interval_seconds.
• Irregular timebases use an explicit numeric strictly increasing time dataset.
• Numeric one-dimensional channel datasets are supported; channel lengths must match their timebase.
• Channel unit attributes are preserved when present; otherwise units are unknown.
• Scalar representable attributes are preserved as metadata.
• Bounded channel values are read using HDF5 dataset slicing where practical.

MATLAB MAT

Product MAT support is limited to SciPy-readable MATLAB v5 MAT files with a top-level vnuli_timeseries structure.

• Runs contain id and timebases.
• Timebases support regular timing with start_seconds and sample_interval_seconds, or irregular timing with a time vector.
• Channels contain id, numeric values, and optional name, quantity, unit, and scalar metadata.
• Units are preserved when present; otherwise units are unknown.
• Valid MAT workspaces without vnuli_timeseries are unsupported.
• MATLAB v7.3 is unsupported/deferred in the product MAT adapter.
• SciPy MATLAB v5 loading materializes arrays when the file is read. Vnuli documents this limitation and does not claim HDF5-style slicing for MAT v5.

NASA Milling mill.mat

NASA Milling remains a separate special/reference MAT path. A NASA mill.mat file registers through the NASA adapter, not the product MAT adapter.

• Dataset name is NASA Milling Dataset.
• Time is generated zero-based regular time.
• Sample rate is 250.0 Hz.
• Source-native units, run IDs, channel IDs, metadata, evidence, and ground truth are preserved.

TDMS

Supported TDMS files use native TDMS file/group/channel structure with numeric channels and sufficient waveform timing properties.

• TDMS file properties become dataset metadata where scalar and representable.
• Each TDMS group is one run.
• Channels are partitioned into timebases by equivalent timing signature.
• Regular waveform timing is supported when channel properties provide finite start or relative start plus sample interval or sample rate.
• Unit-related channel properties are preserved when present; otherwise units are unknown.
• Numeric channels without timing metadata need clarification rather than guessing.
• Bounded values are read using native nptdms channel slicing where practical.

What Vnuli reports

Vnuli classifies sources as supported, clarification-needed, unsupported layout, or invalid/unreadable. Missing optional units do not automatically make a file unsupported, but missing timing or channel structure is not guessed.

Next steps