This project addresses industrial downtime costs by applying Ensemble Learning to sensor telemetry. By identifying mechanical failure signatures before they occur, the system enables an early-intervention strategy that maximizes asset uptime and reduces catastrophic repair costs.
- Accuracy: Achieved a peak predictive accuracy of 98.45% using an optimized XGBoost model.
- Financial ROI: Identified a net operational savings of $65,500 through a custom-built economic utility engine.
- Reliability: Implemented a multi-stage validation layer to ensure data integrity and prevent model drift.
- Languages: Python 3.13
- ML Models: XGBoost (Challenger), Random Forest (Baseline)
- Analytics: Pandas, NumPy, Scikit-learn
- Visualization: Plotly (3D Operational Maps & Radar DNA Profiles)
- Reliability Layer (
step1.py): Automated schema validation and sensor range-gatekeeping. - Modeling Engine (
complete_pipeline.py): Feature engineering and competitive model evaluation (RF vs. XGBoost). - Economic Engine: Integration of business logic to calculate ROI based on maintenance costs vs. failure losses.
- Visual Analytics (
visualize_results.py): Interactive stakeholder dashboards for risk signature identification.
- 3D Risk Map: Visualizes high-risk clusters across Torque, RPM, and Temperature.
- Radar DNA Profile: Identifies "Squeeze" patterns between Torque and Tool Wear that signal imminent failure.
Developed for professional portfolio use. Data sourced from UCI Machine Learning Repository.