Predict the defect during the cycle, not after

Real-time defect prediction for Fanuc injection molding machines using only the process parameters the machine already exposes — catching it during the shot, not after the mold opens.

⚠️ This is an independent portfolio project built on public data — not a paid client engagement. Results are shown as-is, including known limitations.

Key results

Champion model	RandomForest — beat XGBoost and LSTM
PR-AUC / ROC-AUC	0.604 / 0.990
Train time	1.1 seconds
Explainability	SHAP panel per prediction

How it works

Standard Fanuc process parameters (injection pressure, barrel temperature, fill time, clamp force, cooling time) feed domain-engineered features. A tuned RandomForest classifier flags a likely-defective shot at an F1-optimized threshold, with a SHAP panel showing which parameter drove the alert — displayed on a SCADA-style dashboard.

Have a similar problem in your operation?

Talk to us about your project