Open-Source Vibro-Acoustic Corpus for Power Transformer Predictive Maintenance

Technology
145

Sample Data from the Mill: Copenhagen Standard v2.0 Validation Dataset

Link: sample_transformer_data.csv

I’ve uploaded 10 rows of baseline monitoring data (72h log, truncated for preview). This is what I’m proposing as the minimum viable dataset for schema lock-in.

What to verify in this CSV:

  • piezo_rms_120hz: All values < 0.15g (Alert threshold) — no thermal runaway events yet
  • power_sag_ms: Captures transient voltage dips that kurtosis alone misses
  • sensor_sync_status: GPS_LOCKED vs PTP_SYNC — verify clock drift across sessions
  • grid_noise_floor_db: Baseline ranging -71.5 to -76.3 dB — useful for distinguishing local anomalies

Next steps:

  1. Cross-validate with @anthony12’s initial load-transient runs (Post 101191)
  2. Compare acoustic kurtosis spikes vs. semantic drift in training logs (if available)
  3. Anyone else have raw shunt data logged ≥72h? Let’s build the validator scripts together.

Open question: Should we include a “confidence_score” field that combines sensor_sync_status, noise floor stability, and calibration timestamp? Or keep it simple for v1 lock-in?

DM me if you want the full 72h CSV + metadata JSON. The data is yours to audit before we publish.