Pokročilé metody řízení střídavých regulovaných pohonů s využitím umělých neuronových sítí

Abstract

The dissertation thesis focuses on sensorless control of asynchronous drives using neural networks. Sensorless control is an area that is being transitioned to due to the possible failure of sensors, in order to create a more robust and less maintenance-intensive solution. Neural networks are used to replace selected problematic control blocks in order to improve the properties of the methods used. Specifi cally, sensorless methods for estimating speed and current are discussed, which have been verified in a simulation environment and subsequently at an experimental site.

In the thesis the current state of the art in the field of sensorless methods and commonly used control methods that do not rely on speed or current sensors is discussed. Sensorless methods with a reference and adaptive model (MRAS), which are dealt with in the practical part of the thesis, are described in more detail, including their variants. The Q-MRAS, CB-MRAS and CB-MRAS methods modified by an implemented neural network, were used as speed or current estimators.

Thanks to the selected modification of the original MRAS estimators, a more reliable and error resistant solution has been created, eliminating the need for complex calculations in the form of differential equations and other mathematical functions typical for these methods. In addition accurate knowledge of the engine parameters is needed for the correctness of their calculation.

The final part of the thesis describes the process of training and implementing neural networks for individual solutions, and all results of all estimators are verified by extensive simulation tests in the MATLAB Simulink environment. The experimental curves were obtained at a measuring station built on a TMS320F28335 microcontroller from Texas Instruments and measured using a WIZnet W5300 integrated circuit designed for data collection