In recent decades, with the development of the Industry 4.0 concept, Internet of Things (IoT), Machine Learning (ML) and Artificial Intelligence (AI) technologies, there has been an active adoption of these technologies in various industries. One such area is smart manufacturing, where prediction and troubleshooting play an important role. Similarly, the use of neural networks with reinforcement learning is a powerful tool for solving prediction problems. Neural networks can identify complex patterns in production data, and reinforcement learning allows for real-time data analysis, improving the interaction of systems based on experience. This study examines the GEMMA (Guide d’Etude des Modes de Marche et d’Arret) model for process control, which is divided into zones where the procedures for normal operation of installations, diagnostics and equipment failures, and procedures for restoring the system after an accident are implemented. The equipment diagnostics zone is considered, data processing is carried out using neural networks with reinforcement learning. The article presents the application of Proximal Policy Optimization (PRO) and Deep Network (DDN) algorithms. The networks are configured to analyze industrial data received from the Modicon M340 series controller (Schneider Electric). The simulation results and experiments were conducted at the Industrial Automation Lab of Kazakhstan-British Technical University JSC.