A TESTBED FOR EVALUATION OF MACHINE LEARNING TECHNIQUES IN ANOMALY DETECTION IN INDUSTRIAL CYBER PHYSICAL SYSTEM

Abstract

The increasing digitalization of industrial environments, driven 
by Industry 4.0, has fostered the integration of Cyber-Physical 
Systems (CPS), which combine Information Technology (IT) 
and Operational Technology (OT) to enhance efficiency, 
automation, and real-time decision-making. However, this 
convergence has also increased the exposure of Industrial 
Control Systems (ICS) to cyber threats. Thus, it is necessary to 
develop tools and frameworks that contributes to detect 
anomalies and cyber-attacks. In this context, the present work 
proposes a testbed to evaluate Artificial Intelligence 
techniques for automatic detection of anomalies in industrial 
processes. This testbed enables the simulation of both normal 
and anomalous conditions following the GEMMA methodology. 
Furthermore, a use case is presented in which a supervised 
machine learning approach is applied —specifically, the K
Nearest Neighbors (KNN), Random Forest (RF), and Support 
Vector Machine (SVM) algorithms—to data generated in the 
Testbed for Anomaly Detection and Protection of Industrial 
Cyber-Physical Systems (DAyPSCI). The results obtained 
demonstrate the feasibility of the proposed approach to 
classify parts based on their operational behavior, achieving 
high accuracy in identifying normal parts and actuator failures, 
and acceptable performance in detecting sensor failures. 
These findings validate the testbed as a realistic and effective 
environment for the development and evaluation of intelligent 
anomaly detection systems in industrial CPS. Moreover, the 
testbed provides a solid foundation for future research in 
predictive maintenance, process optimization, and industrial 
cybersecurity.