Machine Learning‐Based Prediction of Board Warping in High‐Speed Corrugated Board Manufacturing: An Industrial Case Study towards IIoT‐Enabled Predictive Quality Control
Quality and Reliability Engineering International · 22.11.2026
DOI: 10.1002/qre.70493 ↗Kurz gefasst
Fachartikel in der Zeitschrift Quality and Reliability Engineering International, online erschienen am 22.11.2026. Verfasst von Zahran Abd Elnaby und 3 weiteren Autor:innen. Die Arbeit nennt 23 Literaturangaben.
Abstract (Originalsprache)
This study examines board warping as the dominant quality defect in a high‐speed corrugated board plant and as a candidate for predictive, sensor‐driven control. A seven‐month production audit (January–July 2026, 9,175,600 sheets) showed an overall defect rate of 9.38%, with warping alone accounting for 440,000 rejected sheets (5.0% of total output and 51.1% of all recorded defects), ahead of delamination, board brittleness, and excessive moisture. We trained and compared seven machine learning (ML) classifiers (logistic regression, k‐nearest neighbours, decision tree, support vector machine, Random Forest, gradient boosting, and a small multilayer perceptron) on 100 real production records covering seven process‐related predictors. Neural network and logistic regression models achieved the highest raw test‐set scores (accuracy 0.96 and 0.92; AUC 0.976 and 0.968, respectively), while Random Forest, retained as the primary interpretable model, reached 84.0% test accuracy, an AUC of 0.921, and a cross‐validated accuracy of 90.0% (± 5.5%). Contrary to the moisture‐centred explanation of warping common in the mechanics literature, machine speed and temperature instability emerged as the strongest predictors in the real dataset ( n = 100). To test these empirical findings against potential small‐sample artifacts, we ran a physics‐based numerical simulation of the Warping Index ( n = 2000). The simulation robustly validated the observed pattern: a Random Forest regressor trained on 2000 simulated observations reached R 2 = 0.943 (MAE = 0.409 mm), with the two liner‐temperature terms jointly accounting for 76% of feature importance, confirming that thermal boundaries dominate defect formation under practical operating spans. A conceptual closed‐loop Industrial Internet of Things (IIoT) architecture is proposed, not yet field‐deployed, as an extension
Zitieren
APA
Elnaby, Z. A., Zaher, A., Abdelhameed, A. W., & Elkhouly, H. I. (2026). Machine Learning‐Based Prediction of Board Warping in High‐Speed Corrugated Board Manufacturing: An Industrial Case Study towards IIoT‐Enabled Predictive Quality Control. Quality and Reliability Engineering International. https://doi.org/10.1002/qre.70493
Vancouver
Elnaby ZA, Zaher A, Abdelhameed AW, Elkhouly HI. Machine Learning‐Based Prediction of Board Warping in High‐Speed Corrugated Board Manufacturing: An Industrial Case Study towards IIoT‐Enabled Predictive Quality Control. Quality and Reliability Engineering International. 2026. doi:10.1002/qre.70493
BibTeX
@article{Elnaby2026,
author = {Elnaby, Zahran Abd and Zaher, Amal and Abdelhameed, Abdelhameed Wazeer and Elkhouly, Heba I.},
title = {Machine Learning‐Based Prediction of Board Warping in High‐Speed Corrugated Board Manufacturing: An Industrial Case Study towards IIoT‐Enabled Predictive Quality Control},
journal = {Quality and Reliability Engineering International},
year = {2026},
doi = {10.1002/qre.70493}
}
EUVIMED