Original scientific paper

https://doi.org/10.17559/TV-20251213003201

Enhanced Classification of Shewhart Control Chart Patterns Using Hybrid Features and Adaptive Weighted Ensemble Voting

Waseem Alwan Zaboon ; Mechanical Engineering Department, College of Engineering, Wasit University, Wasit, 52001, Iraq
Yousif Raad Muhsen ; 1)College of Computer Science and Information Technology, Wasit University, Wasit, 52001, Iraq 2)Technical Engineering College, Al-Ayen University, Thi-Qar, 64001, Iraq
Adnan Hassan ; Industrial Engineering Department, Faculty of Engineering, Islamic University of Madinah, Madinah 42351, Saudi Arabia
Dragan Marinković ; 1)TU Berlin, Department of Structural Analysis, Strasse des 17. Juni 135, 10 623 Berlin, Germany 2)University College, Korea University, 145 Anam-ro, Seongbuk-gu, Seoul 02841, Republic of Korea 3)Institute of Mechanical Science, Vilnius Gediminas Technical University, 10105 Vilnius, Lithuania *
Tahsien Al-Quraishi ; School of IT, Victorian Institute of Technology, Australia
Nor Hasrul Akhmal Ngadiman ; Faculty of Mechanical Engineering, Universiti Teknologi Malaysia, UTM Skudai, Johor Bahru 81310, Malaysia
Szabolcs Fischer ; Department of Transport Infrastructure and Water Resources Engineering, Széchenyi István University, 9026 Győr, Egyetem Square 1, Hungary *
Dragan Pamučar ; 1)Transport and Logistics Competence Centre, Vilnius GediminasTechnical University, Vilnius, Lithuania 2)School of Engineering and Technology, Sunway University,Selangor, Malaysia
Darko Božanić ; Military Academy, University of Defence in Belgrade, Veljka Lukića Kurjaka 33, 11000 Belgrade, Serbia

* Corresponding author.

Abstract

Control Chart Pattern Recognition (CCPR) is essential for effective monitoring and fault detection in industrial processes. However, traditional manual interpretation methods face challenges such as vulnerability to noise and difficulty in capturing subtle variations in control chart patterns, limiting their reliability. This study aims to develop a robust automated CCPR system that enhances classification accuracy and reliability through a novel hybrid feature extraction approach combined with an adaptive weighted ensemble voting mechanism. The proposed approach comprises five main phases: Generate synthetic data with varying noise, hybrid feature extraction, feature selection, classifier model with adaptive weighted ensemble Voting - introduction of a dynamic weighting scheme that assigns confidence-based weights to each base classifier's prediction, enabling improved robustness and accuracy, especially under noisy conditions, and accuracy evaluation, the output of each phase is input for next phase. Experimental evaluation on 1,200 synthetic Shewhart chart samples covering six pattern types demonstrated that the proposed weighted ensemble classifier consistently outperformed individual models, achieving classification accuracies of up to 99.1% under noise-free conditions and maintaining high accuracy (98.3%) at realistic 10% noise levels. The ensemble also showed superior inference times and robustness, confirmed by strong confusion matrix diagonal dominance and low misclassification rates. This study presents a highly effective CCPR framework that combines rich hybrid features with an adaptive ensemble mechanism, significantly enhancing accuracy, interpretability, and suitability for real-time deployment. This work presents an adjective approach to developing industrial process monitoring systems that contribute to the early detection and resolution of faults, addressing the shortcomings of previous methods.

Keywords

adaptive weight; control chart pattern recognition; ensemble classifiers, hybrid feature extraction; statistical process control

Hrčak ID:

348715

URI

https://hrcak.srce.hr/348715

Publication date:

30.6.2026.

Visits: 0 *