СИНТЕЗ ДВОМАСОВОЇ ЕЛЕКТРОМЕХАНІЧНОЇ СИСТЕМИ З НЕЙРОРЕГУЛЯТОРОМ СПРОЩЕНОЇ СТРУКТУРИ ТА НЕЛІНІЙНИМ ФРИКЦІЙНИМ НАВАНТАЖЕННЯМ

• Дата: 2025
• Автори: Клепіков, В.Б., Бєляєв, О.С., Обруч, І.В.
• Формат: Стаття
• Мова: Ukrainian
• Опубліковано: Інститут електродинаміки НАН України, Київ 2025
• Теми: двомасова електромеханічна система; електропривод; системи управління; передатна функція; характеристичний поліном; метод кінцевих різниць; характеристика тертя; фрикційні автоколивання; нейронна мережа; квазінейрорегулятор
• Онлайн доступ: https://techned.org.ua/index.php/techned/article/view/1735
• Назва журналу: Technical Electrodynamics

This paper proposes a synthesis methodology for an electromechanical system controller that ensures the required control performance. When linearizing individual sections of a nonlinear load, the structure of the controller resembles a neural network, leading to its designation as a quasi-neurocontroller (QNC). Unlike a PERCEPTRON-type neural network, the proposed controller does not include hidden layers, and its weight coefficients, instead of being determined through multiple iterative calculations (105 – 106), are derived analytically. These relationships are universal and valid for any linearized section. The use of generalized dimensionless parameters makes these relationships applicable to a wide range of electric drives used in machines and mechanisms. A modified set of dimensionless generalized parameters has been introduced, allowing for a simple transition to expressions for systems with absolutely rigid couplings. One of the key advantages of the proposed controller is that it eliminates the need to measure hard-to-obtain system coordinates, such as elastic torque and the need of state observers. Instead, it operates using only a single output variable. The essence of quasi-neurocontrol consists in the construction of the feedback transfer function in the form of a polynomial complement, which ensures that the characteristic polynomial of the closed-loop system matches a polynomial with the desired root locations. To compute the required derivatives, the finite difference method is proposed for use within the quasi-neurocontroller. The methodology of quasi-neurocontrol is illustrated by the example of eliminating friction-induced self-oscillations in a two-mass electromechanical system with a quasi-neurocontroller and a nonlinear frictional load. References 23, figures 11.