Implementation of adaptive systems and machine learning models in industrial automation without plc overload

Abstract

In the context of accelerating digitalization and the introduction of Industry 4.0 concepts, adaptive systems and machine learning (ML) models play a key role in improving the efficiency and reliability of industrial processes. However, the integration of ML into programmable logic controllers (PLCs) faces a number of limitations, including limited computing power and the need for real-time. This article is devoted to the analysis of approaches to the implementation of adaptive systems in industrial automation that can be operated on the fly without significantly increasing the load on the PLC. Distributed learning architectures are considered, which include data processing on peripheral devices (edge computing) and the use of cloud technologies for analyzing and training complex models. Approaches to reducing computational load are discussed, including the use of lightweight algorithms such as linear regression or decision trees, and the use of specialized libraries, such as TensorFlow Lite.