Edge-Compute AI Surveillance for ESP Fault Detection
Real-time playback of Isolation Forest scoring on ESPset vibration records. Paper: SPE-235017-MS.
Results: AUC-ROC 0.9585 and F1 0.921 on the ESPset real offshore vibration dataset.
ESP Edge Monitor
Simulated playback · ESP #4 · real ESPset feature records
Local Wi-Fi · connectedT+00:00
Speed
sample 1 / 150ESPset record id #2
▤ ANOMALY RISK SCORE
NORMAL
0.000/ threshold 1.15
◉ LIVE SPECTRAL ENVELOPE
Normal
reconstructed from peak1x, peak2x & decay features (a, b) — not raw signal
▲ AI EVENT LOG
No anomalies flagged yet — press Play
SENSOR FEATURE READOUT
Ground truth (ESPset expert label): Normal
Feature values and expert labels are the real published ESPset records (ESP #4). Anomaly scoring here is a simplified mean-|z-score| approximation for demonstration —
the deployed system uses a trained Isolation Forest (AUC-ROC 0.9585 on held-out data). Timeline is simulated: original acquisition did not record continuous timestamps between samples.