Controle inteligente de oscilações stick-slip em colunas de perfuração de poços de petróleo utilizando aprendizagem de máquina

Abstract

Stick-slip oscillations represent one of the main causes of problems and the drop in performance of an oil well drilling process, in addition to limiting the life and productivity of the drill bit. Although major improvements have been made to overcome this dysfunction, the design of controllers for this type of system is very challenging, mainly due to the large amount of uncertainties involved. In this work, a biologically inspired framework of an intelligent controller is proposed to control torsional vibrations of drillstrings due to the stick-slip phenomenon. The structure of the intelligent controller is based on the combination of a nonlinear control technique with computational intelligence, which allows the system to make reasonable predictions about the dynamic behavior of the plant, adapt itself to the changes that happen during its operation, learn by interacting with the environment and be robust to both structured and unstructured uncertainties. These characteristics, which are associated with the most fundamental attributes of biological intelligence, are taken into account in the controller design, which is based on the sliding mode control technique together with an adaptive neural network. A reinforcement learning algorithm, based on the Upper Confidence Bound (UCB), is used to be part of the neural network training, which is completed by online updating its weight vector by minimizing a composite error signal. The boundedness and convergence properties of all closed-loop signals are proved using a Lyapunov-like stability analysis. Finally, numerical simulations are presented to validate the efficacy of the proposed approach in controlling the speed of drillstrings and consequently in attenuating the torsional vibrations induced by the stick-slip phenomenon.