Controle inteligente de sistemas subatuados com aplicações em problemas de mecânica do contato

Abstract

In the present work, a nonlinear control strategy based on sliding mode method and artificial neural networks is presented. This approach can be applied to stabilize nonlinear under actuated systems subjected to torsional vibrations. In order to evaluate the proposed approach, the control law was implemented in a drill string. Drill strings are addressed due to their nonlinear and underactuated dynamics. The drill strings are discretized in n parts. The control output is given at the top drive whereas the other string parts are not directly actuated. Torsional vibrations employed in this work are stick-slip, which is the more critic way of torsional vibrations. Stick-slip is a nonlinearity usually caused by friction between the drill string and rock formations and can occurs at any point along its length. Sliding mode control is a very robust technique even with parametric uncertainties and external perturbations. There is no need of previous knowledge of uncertainties and nonlinearities. The discontinuous function adopted at the control law can lead to the undesired chattering effect. The discontinuous function can be replaced by a smooth function to avoid chattering, but this also implies on decrement of the system performance. In order to attenuate this effect, a compensation is added to the control law. Two different artificial neural network architectures are investigated: multilayer perceptrons and radial basis functions. The sliding surfaces are defined as a linear combination of tracking errors. The neural network is employed to mitigate the generated limit cycle. Numerical results are presented in order to demonstrate the performance of the control system.