Neural Networks in Control Systems for Multi-Drive Facilities of Metallurgical Enterprises

Abstract

Metallurgical plants use mechanisms with multiple motors, and their control systems must account for the specifics of the process and the impact of external and internal disturbances. Standard controllers don't always achieve the desired results, especially when systems contain various uncertainties, such as signal, parametric, and structural ones. Therefore, it has been proposed to integrate neural network-based control systems to account for and minimize nonlinearities and disturbances present in the systems. The study developed an algorithm for creating a nonlinear control system that takes into account processes occurring in the mechanical part of the object, as well as random disturbances. An algorithm for creating neural controllers for multi-drive systems was developed. Compared to a standard neural controller, its use allowed for a reduction in the number of neurons in the inner layer and the number of internal layers, reducing the training set by approximately 28 percent. The proposed universal structure of neural controllers and observers enables the creation of stable control systems for multi-drive mechanisms, while maintaining the operational capability of the devices and ensuring high-quality process control even when unmeasured parameters or distorted signals are involved in their formation.