| Resumo: | Nowadays, we have been witnessing an abrupt growth and development of the industry, reflected in the high level of complexity and intelligence that the current production systems present, in which the logistics systems stand out. This incessant search for innovation and continuous improvement are very common today, reproducing into constant changes in the product quality concept. In this sense, the need to optimize the factory layouts emerges, leading to an increase in flexibility because of their dynamic behaviours. In this segment, there is an essential need to improve the behaviour of the associated autonomous vehicle, to reach common objectives such as increasing the productivity and minimizing costs and lead times. In this context, this dissertation, beyond the implementation of the simulation model of the logistics system, develops, in an initial phase, elementary behaviours to be applied to the vehicle, implemented in the simulation environment itself. Subsequently, given that the Machine Learning area has been so successful in other technological areas, the challenge of introducing the concept of the neural network appears, through the creation of a new entity called Agent and characterized by the Reinforcement Learning technique. Finally, in this dissertation, in addition to concluding that the Reinforcement Learning-based approach provided the best productivity results, conclusions were also drawn regarding the robustness of these models, in order to assess their flexibility when subject to different contexts, simulating a real environment.
|
|---|