ABSTRACT
Cyber-physical systems (CPS) are those types of intricate systems that are composed of physical components such as sensors, controllers, and cyber components, for instance, communication networks. Critical applications like manufacturing, transportation, and energy systems are using these systems more frequently. However, because of their complexity and dynamic nature, CPSs are vulnerable to concept drift (CD), which indicates the gradual and often unforeseen changes in their statistical features that underlie them over time. Real-time concept drift detection is required for CPS to operate safely and reliably. For the purpose of real-time concept drift detection in industrial CPS, different machine learning algorithms were developed and assessed. This chapter focuses on the machine temperature dataset from the Numenta Anomaly Benchmark, which includes time-series data from sensors tracking the temperature of an industrial machine’s interior component. Several methods for detecting concept drift have been developed using preprocessed data, including statistical techniques and machine learning algorithms. With these methods, the performance is assessed by comparing several metrics to current state-of-the-art methods. The findings show that the suggested methods have high precision and a low likelihood of false alarms for concept drift identification in real time. The results show that the proposed methodologies can identify concept drift in real time reliably and effectively with low false alarm rates.
