Computer-based data processing approaches to production scrap management

The aim of this study was to design and empirically validate a predictive approach for classifying scrap types and forecasting scrap volume in brake calliper remanufacturing, oriented toward operational decision-making. We addressed three research questions: (1) which “red-flag” defects most strongly increase the risk of high scrap (“Large”); (2) whether scrap category is driven by the accumulation of multiple defects or by single catastrophic defects; and (3) whether logistic regression provides sufficient performance and interpretability for industrial use, or whether more complex machine-learning models are justified. The analysis used real industrial data (435 cases, 34 defect-related features with discrete values: 0/1 and, in some cases, counts). The baseline model (logistic regression) achieved 79% accuracy and an AUC of 0.773, confirming its practical usefulness in a production environment. Coefficient analysis indicated, among others, that the defect “power-supply leakage” is a strong predictor of the “Norm” class (OR ≈ 4.64 for “Norm” vs. “Large”), while defect-profile analysis revealed an accumulation trajectory (mean 7.3 defects for “Large” vs. 3.5 for “Norm”), distinct from a catastrophic trajectory (immediate disqualification). The study shows that interpretable models can support decision processes in remanufacturing, reducing scrap and supporting circular-economy and Industry 4.0 goals.