Skip to content

[Process] Establish a signal-processing baseline for VEST IRE onset and multi-diagnostic event composition #668

Description

@HongSik-Yun-Fusion

Summary

Establish a transparent, signal-processing-based baseline for detecting internal reconnection event (IRE) timing in VEST using routinely available diagnostics, before relying on higher-level unsupervised classification or morphology discovery.

The immediate goal is not to define one universal physical reconnection_onset. Instead, detect independently observable event markers and compose them into an IRE candidate with explicit provenance.

Primary routine signals supported by the published VEST IRE observations are:

plasma current        -> positive current spike
magnetic fluctuation  -> abrupt magnetic burst
loop voltage          -> rapid excursion / spike
H-alpha               -> emission burst
O I                   -> impurity-emission burst

Ion Doppler spectroscopy (IDS) measurements of ion heating and rotation response are scientifically important but are excluded from the production detector because they are not routinely available at sufficient temporal/spatial coverage.

Conceptually:

Ip                -> current_spike_onset
Mirnov / dBdt     -> magnetic_burst_onset
Vloop             -> loop_voltage_excursion
H-alpha           -> halpha_burst_onset
O I               -> oi_burst_onset
                         |
                         v
              multi-diagnostic coincidence
                         |
                         v
                   IRE candidate

The detector should remain interpretable enough that every IRE candidate can be traced back to the original waveform, preprocessing, threshold/persistence rule, and timing uncertainty.


Scientific basis

The primary reference is:

S. Kim, J. Y. Jang, Y. Kim, and Y. S. Hwang (2021),
Acceleration of ion rotation during internal reconnection events in the versatile experiment spherical torus (VEST),
Nuclear Fusion 61, 126011.
https://doi.org/10.1088/1741-4326/ac2b78

The paper reports the following recurring IRE signatures in the representative VEST discharges:

  • a positive plasma-current spike;
  • an abrupt magnetic fluctuation/burst;
  • a rapid loop-voltage spike/excursion;
  • bursts in H-alpha and O I emission;
  • ion heating and rotation changes observed with IDS.

The paper also shows two important constraints for detector design:

  1. A strong MHD precursor is not required. One discharge has substantial pre-IRE magnetic activity, while another remains comparatively quiet until an abrupt magnetic burst at the IRE.
  2. Loop-voltage sign is not universal. Ramp-up and ramp-down IREs show loop-voltage spikes in opposite directions, while the current spike, magnetic burst, and ion-heating response remain qualitatively similar.

Therefore, precursor activity should be optional metadata rather than a required IRE criterion, and loop voltage should be treated as a signed-independent excursion rather than a fixed-sign threshold.


1. Detect diagnostic-level events independently

Do not begin with a single monolithic detect_ire() rule.

First expose independently testable diagnostic events.

Plasma current

Candidate outputs:

current_spike_onset
current_spike_peak
current_spike_end

Use robust baseline-relative signal processing on Ip and/or dIp/dt.

The production criterion should distinguish an abrupt positive IRE-associated spike from normal current evolution, ramp-down, and isolated noise.

Reuse generic onset / sustained-excursion primitives from #409 where appropriate.

Magnetic fluctuation

Candidate outputs:

magnetic_burst_onset
magnetic_burst_peak
magnetic_burst_end

A suitable scalar may be derived from one or more validated Mirnov / magnetic-fluctuation channels, for example an RMS/envelope or derivative-excursion metric.

Do not require a long precursor mode to exist before the burst.

Optional precursor products may remain separate:

mhd_precursor_onset
mhd_precursor_amplitude
mode_frequency
mode_number

These are useful characterization quantities but not mandatory evidence for the IRE itself.

Loop voltage

Candidate output:

loop_voltage_excursion

Detect an abrupt deviation relative to the local baseline or slope.

Do not encode a fixed sign such as:

Vloop < threshold

because the 2021 VEST paper reports opposite spike directions for ramp-up and ramp-down IREs.

H-alpha and O I

Candidate outputs:

halpha_burst_onset
halpha_burst_peak

oi_burst_onset
oi_burst_peak

Treat these as optical/radiative response events, not direct definitions of reconnection onset.

Reuse signal-independent burst/onset/change-point primitives rather than introducing line-specific threshold code where possible.


2. Compose diagnostic events into an IRE candidate

After the individual events are validated, compose them using temporal coincidence.

Initial hypothesis:

primary evidence:
    current spike
    + magnetic burst

supporting evidence:
    loop-voltage excursion
    H-alpha burst
    O I burst

A first conservative candidate rule may require the current-spike and magnetic-burst events to occur within a configurable coincidence window:

[
|t_{I_p} - t_B| < \Delta t_{\rm coincidence}.
]

Supporting optical/loop-voltage events should strengthen the interpretation but should not initially be mandatory if the corresponding diagnostic is unavailable or unusable.

Do not hide missing data behind a lower threshold. Distinguish explicitly between:

no supporting event detected
supporting diagnostic unavailable
supporting diagnostic ambiguous/unusable

The coincidence tolerance must be derived from diagnostic bandwidth/timebase and corpus validation, not chosen as an undocumented constant.


3. Preserve observable timing rather than claiming exact reconnection onset

Use names tied to what is actually measured:

current_spike_onset
magnetic_burst_onset
halpha_burst_onset

Do not immediately expose:

reconnection_onset

as though it were directly measured.

If later multi-diagnostic and physics analysis justifies a higher-level IRE transition time, it should be derived from the observable event set with explicit assumptions and uncertainty.


4. Relative timing as a first physics product

Once the observable events exist, provide event-to-event timing quantities such as:

[
\Delta t_{B-I}
= t_{\rm magnetic\ burst} - t_{\rm current\ spike},
]

[
\Delta t_{H-I}
= t_{\rm H\alpha\ burst} - t_{\rm current\ spike},
]

and analogous quantities for loop voltage and O I.

These should use the common temporal-relation infrastructure from #667 rather than creating IRE-specific timing utilities where possible.

Population distributions of these delays can then answer whether the routine signatures form a reproducible event sequence or only a broad coincidence class.


5. Compare against existing unsupervised IRE detection

Where labels/results from the existing VAFT-based unsupervised IRE study are available, use them as an independent comparison target rather than as ground truth.

For each method, report at least:

signal-processing only
unsupervised only
agreement

and inspect disagreements manually.

The main questions are:

  • Are most IREs recoverable from explicit current-spike + magnetic-burst logic?
  • Which weak, overlapping, or morphologically unusual events require the unsupervised approach?
  • Does unsupervised detection improve event discovery, while explicit signal processing gives more reproducible onset timing?
  • Are false positives dominated by disruptions, capacitor transients, ordinary MHD bursts, or optical spikes?

Do not automatically replace the existing unsupervised method. The purpose is to establish a physically transparent baseline and quantify where more complex methods add value.


6. Corpus and negative controls

Validation must include more than hand-picked strong IREs.

At minimum inspect:

Positive cases

  • strong IRE with quiet pre-event MHD;
  • IRE with strong precursor / tearing activity;
  • ramp-up IRE;
  • ramp-down IRE;
  • weak / incomplete IRE where available;
  • recovery after IRE;
  • disruption-associated IRE-like events where available.

Negative / confounding cases

  • ordinary current evolution without IRE;
  • smooth ramp-down;
  • isolated magnetic burst without current spike;
  • current transient without magnetic burst;
  • capacitor/power-supply transient;
  • H-alpha / O I burst unrelated to IRE;
  • failed or marginal plasma;
  • terminal disruption without a clear recoverable IRE morphology.

Power-system transients must be handled explicitly, especially because the 2021 paper notes a second OH-capacitor discharge near the current peak in the analyzed shots.


7. Architecture

Keep generic signal processing separate from IRE semantics.

Conceptually:

vaft.process.onset / generic signal primitives
    sustained excursion
    derivative transient
    burst envelope
    change point
        |
        v
IRE diagnostic event adapters
    current spike
    magnetic burst
    Vloop excursion
    H-alpha burst
    O I burst
        |
        v
IRE composition
        |
        v
#667 plasma-evolution timeline

The generic processing layer must not contain VEST/IRE-specific event names or thresholds.

The IRE-specific layer should preserve:

source diagnostic
preprocessing
criterion
parameters
native timebase
uncertainty / tolerance
quality flags
provenance

8. Relationship to existing work


Acceptance criteria

  • Define routine-diagnostic IRE event semantics from the published 2021 VEST observations without requiring IDS.
  • Implement or reuse robust current-spike onset/peak detection.
  • Implement or reuse robust magnetic-burst onset/peak detection.
  • Detect loop-voltage excursion without assuming a universal sign.
  • Detect H-alpha and O I burst timing where the signals are usable.
  • Keep a strong pre-IRE MHD precursor optional rather than required.
  • Return structured event records rather than bare timestamps.
  • Distinguish unavailable/unusable diagnostics from true absence of a supporting event.
  • Compose at least current-spike + magnetic-burst coincidence into an IRE candidate.
  • Determine the coincidence tolerance from timebase/bandwidth and corpus validation.
  • Preserve observable event names rather than claiming direct measurement of exact reconnection onset.
  • Compute relative event timing through the generic [Umbrella] Compose VAFT event/state products into a plasma-evolution timeline and physics-association framework #667 temporal-relation path where available.
  • Validate strong, weak, ramp-up, ramp-down, precursor-rich, and precursor-quiet IRE cases.
  • Validate negative/confounding cases including ordinary ramp-down and power-system transients.
  • Compare the signal-processing baseline against the existing unsupervised IRE results where available.
  • Document disagreement cases and identify where the unsupervised method provides genuine additional value.
  • Provide representative validation plots with event markers over the original signals.

Non-goals

  • IDS-based production detection;
  • defining ion heating or rotation change as mandatory IRE evidence;
  • forcing one universal physical reconnection_onset timestamp;
  • requiring a precursor tearing mode for every IRE;
  • deriving mode numbers as part of the minimum detector;
  • FAST-camera filament/phase-alignment reproduction already owned by Reproduce VEST FAST-camera fluctuation and MHD-correlated filament analysis #161;
  • equilibrium/stability interpretation of why the IRE occurred;
  • replacing the existing unsupervised IRE method before corpus-level comparison;
  • ML forecasting or causal inference.

The first target is a transparent, reproducible signal-processing baseline for observable IRE timing, with higher-level IRE interpretation and physics association layered on top.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions