ABSTRACT
The research presents machine learning for securing industrial IoT systems. It describes the implementation of an accompanying architecture and demonstrates its effectiveness for cyber-attack detection and motor status monitoring. The inclusion of Local Outlier Factor and Actor-Critic Methods support an average cyber-attack detection rate of 95%, enhancing its application for early defense against new threats. Gradient Boosting Machines, in turn, provides a predictive accuracy of 92% effectiveness for forecasting motor failures to schedule maintenance and optimize equipment for reduced downtime. In addition, the architecture uses Transformer models and Seasonal Decomposition of Time Series to interpret the temporal structures and contextual information. As a result, the developed solution's accuracy to detect irregular equipment activities is 90%, enabling intervention and preventive measures before any threat to industrial applications. The architecture is adaptable to multiple standard industrial environments and is capable of scale as it can be integrated into the existing IIoT infrastructure. Finally, the results show how the proposed architecture has a high potential of transforming industrial cybersecurity and asset management. The incorporation of machine learning and real-time analytics has made it possible to detect even those threats and equipment failures that have not been yet identified, thereby making industrial IoT systems reliable and resilient to modern challenges.
