Authors:
Davi Neves
1
and
Ricardo Augusto Rabelo Ricardo Augusto Rabelo
2
Affiliations:
1
Department of Production Engineering, University Federal of Ouro Preto, Ouro Preto, Brazil
;
2
Department of Computing Science, University Federal of Ouro Preto, Ouro Preto, Brazil
Keyword(s):
Dynamic Systems, Koopman Operator, Reinforcement Learning, Neural Networks, Topological Measures.
Abstract:
Manufacturing processes are generally modeled through dynamic systems, whose solutions establish a tool for control theory, essential in the elaboration of industrial automation, a pillar of the fourth revolution. Understanding and mastering these technological procedures correspond to the ability to determine and analyze the solutions of a system of differential equations, in order to deploy smart devices in a production line, such as the robotic arm, because this trajectories can be always associated with the running of any equipment. Currently there are many formulated methods to determine (or forecast) these curves, through numerical or stochastic tools, the focus in this work are those capable of reconstructing a state space, such as the Koopman’s operator, convolutional neural network and reinforcement learning technique. Therefore, based on the solutions provided by these methods, a benchmark will assembled to compare them, using topological measures such as Shannon entropy, L
yapunov exponent and Hurst coefficient, thus defining the effectiveness of each one.
(More)