Date of Award

5-2026

Degree Name

Doctor of Philosophy

Department

Industrial and Entrepreneurial Engineering and Engineering Management

First Advisor

Lee Wells, Ph.D.

Second Advisor

Ilgin Acar, Ph.D.

Third Advisor

Jim Burns, Ph.D.

Fourth Advisor

Shameek Bhattacharjee, Ph.D.

Keywords

Control charts, cyber security, manufacturing, process control, quality control, statistics

Abstract

The increasing integration of digital technologies and industrial control systems in modern manufacturing has introduced new cybersecurity vulnerabilities within cyber–physical production environments. Malicious actors can exploit these vulnerabilities to induce subtle process deviations that degrade product quality while remaining undetected by conventional statistical monitoring tools. Such attacks can be deliberately engineered to manipulate process behavior through transient shifts that vary in magnitude, duration, and frequency. Despite extensive research on transient shifts caused by assignable causes in Statistical Process Control (SPC), limited attention has been given to intelligently designed cyber–physical attacks that exploit the structural characteristics and limitations of control charting schemes. This research evaluates the ability of commonly used SPC monitoring schemes to detect malicious process manipulations and proposes enhancements to improve their resilience against adversarial behavior. Several attacker-driven transient attack patterns are investigated, including alternating, constant, and trending mean shifts. The performance of multiple monitoring schemes, including the Shewhart 𝑋chart, πΈπ‘Šπ‘€π΄x, πΈπ‘Šπ‘€π΄x2, and their combinations, is evaluated through extensive simulation experiments. Performance is assessed using both traditional SPC measures and newly introduced evaluation metrics to better capture the operational and quality impacts of malicious attacks. Results indicate that combined monitoring schemes provide improved detection performance across multiple attack scenarios by leveraging the complementary strengths of different charting statistics. To further reduce attacker advantage, a randomized monitoring approach is introduced that limits an adversary’s ability to tailor effective attack strategies by reducing predictability in the monitoring structure. Additionally, a comparative study between the πΈπ‘Šπ‘€π΄x, πΆπ‘ˆπ‘†π‘ˆπ‘€, and the Shewhart 𝑋chart is conducted using new assessment metrics, including Directional Average Run Length and the percentage of simulations with false ignored signals (%SFS), providing deeper insight into detection behavior under adversarial conditions. Using these metrics, the results show that πΆπ‘ˆπ‘†π‘ˆπ‘€ achieves superior peak detection efficiency when optimally tuned, whereas πΈπ‘Šπ‘€π΄x demonstrates greater robustness to parameter misspecification. The Shewhart 𝑋chart shows inferior performance relative to πΆπ‘ˆπ‘†π‘ˆπ‘€ and πΈπ‘Šπ‘€π΄x for small to moderate shifts; however, its performance improves substantially for large process shifts, where its simplicity allows for faster detection.

Access Setting

Dissertation-Open Access

Download

Share

COinS