Future Industrial Production Work Designusing Clustering Approach: Transitioning to Cyber Physical Systems

The proliferation of Internet of Things (IoT) devices and infrastructure has prompted the development of several IoT data analytics methods for detecting hostile cyberattacks and keeping IoT infrastructure safe. The intrinsic instability of IoT data, however, may cause concept drift concerns in IoT data analytics, which in turn can cause model deterioration and the inability to detect assaults. This is due to the fact that static models used in conventional data analytics cannot accommodate for shifts in the distribution of data. In a smart factory, industrial equipment is outfitted with thousands of interconnected devices or sensors that collect and transmit massive amounts of data about the machines' states to a cyber physical system in the cloud. The system then use a variety of CBM approaches to forecast when machines will begin to behave oddly, enabling timely and cost-effective preventive maintenance or component replacement. Unbalanced data (where data containing faults is a tiny proportion of total data) and concept drifts (where the pattern of the distribution of fault patterns may alter over time) are two problems that afflict CBM. A high-performance answer to these issues may be found in ensemble learning, which incorporates the variety of several classifiers. In reality, most businesses likely lack the resources to set up a stable architecture to support actual time online classification algorithms, but they may already be using commercially available offline classifiers. In this study, we provide an IoT data stream analytics framework based on Performer Weighted Probability Averaging Ensembles (PWPAEs) for drift-adaptive IoT anomaly identification. The efficacy of our proposed PWPAE approach is shown experimentally on two publicly available datasets, where it is compared to state-of-the-art methodologies.