Engineering Mechanics

Proceedings Vol. 27/28 (2022)

ENGINEERING MECHANICS 2022

Copyright © 2022 Institute of Theoretical and Applied Mechanics of the Czech Academy of Sciences, Prague

 

Rágulík J., Sivčák M.
pages 325 - 328, full text

The paper deals with the comparison of control approaches to the height of the bellows air spring. These approaches are PI control, a deep reinforcement learning algorithm, and a combination of both control techniques, the neuro-PI controller. In the case of a neuro-PI controller, it is essentially control by a reinforcement learning algorithm, but its control neural network contains only two weights that are gained in the learning process. The advantage of the neuro-PI controller created in this way is that its functionality can be formally verified in the same way as a classic PI controller. However, such a procedure is only possible when controlling relatively simple systems, such as the described application for controlling an air spring, which is a dynamic system with one degree of freedom. All the above approaches are applied to a mathematical model of an air spring created in the Simulink environment; a deep reinforcement learning algorithm is created in MATLAB.

Ownership of copyright in original research articles remains with the Authors, and provided that, when reproducing parts of the contribution, the Authors acknowledge and/or reference the Proceedings, the Authors do not need to seek permission for re-use of their material.

All papers were reviewed by members of the scientific committee.