How starryroses/turbofan-rul-predictor works
This repository is an interactive turbofan predictive-maintenance dashboard. A user selects an engine from the FD001 test data or enters sensor readings manually; the app runs the readings through a stacked RUL predictor, then presents health, anomaly, uncertainty, explanation, and maintenance information. The README also documents training and validation methods, but the sampled source is the dashboard rather than the training implementation; those methods are therefore represented only as documented model provenance.
Main entry points: streamlit_app.py Inference pipeline · xgb_model.pkl XGBoost · scaler.pkl Feature scaling · rf_model.pkl Random Forest
Components 21
Dashboard interaction
- Streamlit dashboard
src/streamlit_app.py - Engine selection
src/streamlit_app.py - Manual sensor input
src/streamlit_app.py - Health report
src/streamlit_app.py
RUL inference
- Inference pipeline
src/streamlit_app.py - Feature scaling
src/models/scaler.pkl - Random Forest
src/models/rf_model.pkl - XGBoost
src/models/xgb_model.pkl - LightGBM
src/models/lgbm_model.pkl - Ridge meta-model
src/models/meta_model.pkl - XGBoost calibration model
src/models/xgb_calib_model.pkl
Prediction insights
- Anomaly detection
src/models/iso_model.pkl - Conformal intervals
src/streamlit_app.py - SHAP explanations
src/streamlit_app.py - Maintenance guidance
src/streamlit_app.py
Data and model assets
- Feature metadata
src/info/feature_cols.json - Sensor selection
src/info/sensors_to_use.json - FD001 test data
src/data/test_FD001_processed.csv - True RUL values
src/data/test_FD001_true_rul.csv - Calibration residuals
src/models/calibration_residuals.npy
Other components
- Dashboard user
Connections 22
- Dashboard user → Streamlit dashboard: uses
README.md - Streamlit dashboard → Engine selection: offers mode
src/streamlit_app.py - Streamlit dashboard → Manual sensor input: offers mode
src/streamlit_app.py - Engine selection → FD001 test data: loads
src/streamlit_app.py - Engine selection → True RUL values: loads
src/streamlit_app.py - Engine selection → Feature metadata: loads
src/streamlit_app.py - Engine selection → Sensor selection: loads
src/streamlit_app.py - Engine selection → Inference pipeline: submits readings
src/streamlit_app.py - Manual sensor input → Inference pipeline: submits readings
src/streamlit_app.py - Inference pipeline → Feature scaling: scales features
src/streamlit_app.py - Inference pipeline → Random Forest: predicts with
src/streamlit_app.py - Inference pipeline → XGBoost: predicts with
src/streamlit_app.py - Inference pipeline → LightGBM: predicts with
src/streamlit_app.py - Inference pipeline → Ridge meta-model: combines predictions
src/streamlit_app.py - Inference pipeline → Anomaly detection: scores anomaly
src/streamlit_app.py - Inference pipeline → Health report: returns results
src/streamlit_app.py - Health report → Conformal intervals: computes interval
src/streamlit_app.py - Health report → Calibration residuals: uses calibration
src/streamlit_app.py - Health report → SHAP explanations: explains prediction
src/streamlit_app.py - SHAP explanations → XGBoost: explains with
src/streamlit_app.py - Health report → Maintenance guidance: shows guidance
README.md - Inference pipeline → XGBoost calibration model: loads model
src/streamlit_app.py