Loading...
SVM-based control chart for Quality 4.0
Abdelwahid, Issra M.
Abdelwahid, Issra M.
Files
35.232-2025.65a Issra M. Abdelwahid.pdf
Adobe PDF, 1.79 MB
- Embargoed until 2027-03-04
Description
A Master of Science thesis in Engineering Systems Management by Issra M. Abdelwahid entitled, “SVM-based control chart for Quality 4.0”, submitted in November 2025. Thesis advisor is Dr. Mahmoud Awad. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).
Abstract
The emergence of Quality 4.0, driven by Industry 4.0 technologies, has introduced complex, high-dimensional, unstructured, and non-linear data that challenge the effectiveness of traditional Statistical Process Control (SPC) methods. Hence, this thesis proposes a novel machine learning-based control chart, utilizing Support Vector Machine (SVM) to address these limitations and enhance quality monitoring systems. In contrast to conventional SPC methods, which struggle with multivariate data and lack adaptability, the proposed model integrates advanced machine learning techniques to create a flexible and robust framework suitable for modern industrial environments and smart manufacturing processes. Furthermore, the proposed methodology follows an eight-step pipeline: Data collection, Data preprocessing, Develop control chart using one-class methods, understanding out of control process shifts, generation of out of control points either independently or dependently via kernel density estimation (KDE) sampling of joint tails, development of a control chart using a two-class method with fine-tuning of cost and weight parameters either manually or ROC-guided.
Three case studies demonstrate the approach: a Chemical process, Tobacco, and Penicillin fermentation process were used to validate the proposed method. Results show no single chart dominates across all settings. For low-dimensional, small, sustained shifts, MEWMA outperformed the other methods with 100% true positive and negative rate detection. For the Tobacco case, MEWMA and one-class SVM provided a 100% true positive rate and 85% true negative rate. Finally, for Penicillin, where dimension and correlation are stronger, two-class SVM with KDE achieved the best balance with 100% true positive rate and 98% true negative one. Overall, the proposed method can be used to improve out-of-control events detection in complex and multivariate systems.