Data-Driven Machine Performance Assessment by Adjusting for Product Mix Effects on Scrap: A Case Study in the Wire and Cable Industry
Data-Driven Machine Performance Assessment by Adjusting for Product Mix Effects on Scrap: A Case Study in the Wire and Cable Industry
Parvaneh Ahmadi1 Payman derakhshan barjouei2
1) M.Sc. in Industrial Engineering (Systems Optimization), Quality Control Specialist, Shirkooh Wire and Cable Company, Yazd, Iran
2) Ph.D. Student in Industrial Management – Operations Research, Department of Management, Islamic Azad University, Yazd Branch, Yazd, Iran; Planning and Quality Manager, Shirkooh Wire and Cable Company, Yazd, Iran
IOI Code
:
ACEF-ACAZKG
Publication :
6th. International Congress On Engineering, Technology and Applied Sciences(etas2027.com)
Abstract :
Reducing production scrap and accurately assessing machine performance are important challenges in multi-product manufacturing systems. However, directly comparing machines based on raw scrap values may provide a misleading picture of actual equipment performance due to differences in production volume and product mix. This study proposes a data-driven approach for adjusted machine performance assessment in a wire and cable manufacturing plant. Real production and scrap data collected from 2024 to 2026 were cleaned, integrated, and analyzed at the machine-shift level, resulting in a final dataset of 1,240 observations. Variables including machine identity, production length and weight, product mix, number of production runs, operator information, and temporal features were used to estimate expected scrap. The performances of Random Forest, CatBoost, and XGBoost models were compared. XGBoost achieved the best predictive performance, with a Mean Absolute Error (MAE) of 5.64 kg, a Root Mean Squared Error (RMSE) of 12.10 kg, and a coefficient of determination (R²) of 0.488. The difference between actual and expected scrap was then calculated as an adjusted performance indicator. The results showed that machines A and B generally performed better than expected, whereas machines C and D exhibited scrap levels above the expected values. SHAP analysis further indicated that temporal features, production length, machine identity, and the number of production runs were among the most influential variables in scrap estimation. The proposed framework can support fairer machine comparisons, identification of high-risk production conditions, and data-driven process improvement decisions in the wire and cable industry.
Keywords : production scrap performance assessment product mix machine learning wire and cable industry