Manufacturing Intelligence Series | Module 2 of 6

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.

Scrap rate from 4.5% to 3.4% – on one lever

Every avoided scrap part saves material AND the rework that would otherwise follow.

Batch data
Process parameters
Gradient boosting
Risk score
Intervention before scrap

02 The model – Risk score per batch and parameter set

Classification across all process and material features known at production time

▸ Output
ROC-AUC Ausschuss-Klassifikation: 0.912
Top-Treiber: ['material_charge_B', 'werkzeug_temp', 'vorschub', 'werkzeug_standzeit_h']
Kritische Kombination -> Ausschussrate: 14.8%
Scrap rate by parameter combination – top drivers
↳ The strength lies in the interactions

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

€419,000
Savings / year
4.5% → 3.4%
Scrap rate
−25%
Less rework
Material value of scrap – before vs. with model
Model calculation · How the figure is built – derived transparently

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.

ItemValue
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.

↳ From finding to rule

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

① Link inspection data

We connect process parameters from the MES with the inspection results from quality assurance – the data already exists.

② Model per product family

For the highest-scrap product family a pilot model is built. You see the drivers validated on real batches.

③ Live warning

Risk score before production start: "This batch with these parameters has elevated scrap risk – recommendation: adjust parameter X."