Scrap forecast:
Know the defect before the batch runs
Which combination of material, machine parameters and conditions leads to scrap? Gradient boosting learns it from your inspection and process data – and warns before expensive material is processed.
01 The problem – Scrap is detected too late today
Final inspection tells you something went wrong – not why
Scrap costs twice: the lost material and the rework. Today it usually surfaces only at final inspection – once material, energy and machine time are already invested.
Yet the causes sit in the data: batch properties, temperature, pressure, tool wear, feed rate. A model detects the critical combination a human cannot see in the parameter space.
Every avoided scrap part saves material AND the rework that would otherwise follow.
02 The model – Risk score per batch and parameter set
Classification across all process and material features known at production time
ROC-AUC Ausschuss-Klassifikation: 0.912 Top-Treiber: ['material_charge_B', 'werkzeug_temp', 'vorschub', 'werkzeug_standzeit_h'] Kritische Kombination -> Ausschussrate: 14.8%
Individual parameters in the green zone does not mean "all good": only the combination of material batch B, elevated tool temperature and high feed rate tips quality over. Gradient boosting (XGBoost) models exactly these interactions – and names the concrete driver via SHAP values.
03 Business impact – Lower material and rework cost
The lever: reduced scrap rate × material input, plus reduced rework
No round number: every assumption comes from the sample plant and is stored centrally. With your real figures only the input changes, not the method.
| Item | Value |
|---|---|
| Material input / year | €24,000,000 |
| Material savings (Scrap rate: 4.5% → 3.4%) | €264,000 |
| Rework cost / year | €620,000 |
| Rework reduction (25%) | €155,000 |
| Result: savings / year | €419,000 |
Assumptions of a sample plant – in a real project your data replaces these values.
The model does not stop at the warning: the drivers become process windows that feed back into the controls. "Batch B + hot + fast" becomes a concrete parameter limit – permanently.
04 Next steps
From risk score to closed-loop control
We connect process parameters from the MES with the inspection results from quality assurance – the data already exists.
For the highest-scrap product family a pilot model is built. You see the drivers validated on real batches.
Risk score before production start: "This batch with these parameters has elevated scrap risk – recommendation: adjust parameter X."