Back to projects
PredictMOP, predictive maintenance

5
models benchmarked
ISO
vibration severity
| Problem | Unplanned stops were caught late: vibration data and thermal images were reviewed by hand, without a common severity criterion. |
|---|---|
| Solution | Machine Learning models (Random Forest, Gradient Boosting, XGBoost, Ridge, ElasticNet) with Isolation Forest anomaly detection, Kalman filtering and IQR cleaning validated with k-fold; computer vision (ORB, homography, RANSAC) reading temperature from thermal images; ISO vibration severity and an asset-health board on FastAPI and PostgreSQL. |
| Results | Early warnings on equipment condition and one view of severity and criticality to decide where to act first. |
| Technologies | Python, scikit-learn, XGBoost, OpenCV, FastAPI, PostgreSQL |
