Industrial AI System
An end-to-end industrial AI system that predicts machine failure from real-time synthetic telemetry.
The Digital Factory
To train and validate an industrial predictive maintenance system without access to proprietary factory data, this project begins by creating a synthetic industrial digital twin.
The simulator generates a factory containing thousands of machines, producing real-time sensor telemetry (temperature, vibration, pressure, load) that can be consumed by the AI inference engine at a 2Hz broadcast rate.
System Architecture
Machine Learning Pipeline
The project uses a highly calibrated XGBoost ensemble (100 trees), processed through `StandardScaler` and `OneHotEncoder`, and finalized with Isotonic Calibration. The classification boundary is mathematically locked at `0.31`.
Explainable AI (XAI)
The system doesn't just predict failure. It explains why.
SHAP FEATURE CONTRIBUTIONS
Macro-level analysis of feature contributions across the entire 100-tree ensemble.