Industrial AI & edge computing
ESP Edge Monitor
Vibration fault detection for electrical submersible pumps.

The project
Overview.
Edge-computing research and an interactive dashboard for detecting faults in electrical submersible pumps using vibration data.
The research connects an Isolation Forest model, a Wi-Fi-enabled microcontroller and a local operator interface. The linked browser demo uses published vibration records in a simulated timeline.
01 / The problem
Monitoring pumps without a cloud connection
Electrical submersible pumps are critical to oil production, but continuous condition monitoring can be difficult to justify on smaller or remote fields. The research investigates an edge-computing approach that can work with vibration measurements and a local interface.
The proposed system pairs a low-cost Wi-Fi-enabled microcontroller with a locally hosted progressive web app, so operators can inspect condition information without depending on an internet connection.
02 / The model
Learning normal operation from vibration data
An Isolation Forest was trained on normal-condition records and used seven frequency-domain vibration features to identify anomalies. Validation used ESPset, a real offshore dataset containing 6,032 records from 11 pumps and five operating conditions.
The paper reports an AUC-ROC of 0.9585 and an F1-score of 0.9129 on 2,192 held-out records. It estimates hardware costs below US$50 per monitoring node.
03 / The interface
A field dashboard and a public demonstration
The field concept connects equipment data, edge processing and a local operator dashboard. The public browser demonstration makes the sensor features, risk score and event log explorable without connecting to industrial hardware.
The demonstration uses published feature records, a simplified scoring approximation and a simulated timeline. It is not a live feed from a pump and does not run the paper’s trained model. The demo displays an F1-score of 0.921; the paper reports 0.9129.
The interface
Desktop and mobile.


Result
The published study reports 0.9585 AUC-ROC and 0.9129 F1 on held-out offshore ESP data.
Have a project in mind?