Fuzzy logic control with reinforcement learning on slow time-varying electro-mechanical systems
Roel Pantonial, Sandra Mae W. Famador · 2014
This paper demonstrates the design and implementation of a Fuzzy Logic Controller (FLC) with Reinforcement Learning (RL) on Electro-mechanical system in the industrial and commercial setting. FLC, which is a nonlinear controller, is based on the linguistic description of the system and not its mathematical model. RL on the other hand is a class of learning task wherein an agent maximizes a scalar evaluation by environment interaction. Combining these two schemes, an online adaptation is achieved on FLC without the need of a training set. FLC design is presented first in which expert knowledge are incorporated at the start of a system's life to reduce heavy learning phase. Afterwards, RL is added to fine tune the conclusion part of the FLC. Then, a C-based Simulation is demonstrated on a position controlled time-varying electro-mechanical system. The experimental results show the plausibility and applicability of such design in the industry.