Neural network based controllers for the oil well drilling process

The oil well drilling is a complex process, requiring that the bottomhole pressure be maintained within an operational window, limited by fracture pressure and pore pressure (and/or wellbore collapse). During the drilling process, there are several phenomena that directly influence the annulus bottomhole pressure: kick, lost circulation, inefficient removal of solids, well length increase, reservoir parameters and pipe connection procedure (surge and swab effects). The main objective of this work is to develop neural network controllers able to regulate the annulus bottomhole pressure under pipe connection procedure, kick and lost circulation disturbances. For this purpose, an experimental drilling unit was built, depicting the oil well drilling process primordial characteristics. Besides, real oil well offshore data were employed for validating the neural network controllers. Neural networks were built for predicting the direct plant dynamics (inverse neural network controller) and the inverse plant dynamics (direct neural network controller). It is noteworthy to mention that direct plant dynamics neural modelling was employed as the nonlinear model of a gain scheduling control approach; in addition, the inverse plant dynamics neural modelling constituted a nonlinear controller, directly predicting the manipulated variable. Both controllers employed real-time neural network training, being, therefore, adaptive schemes for regulating the oil well drilling process, which nature is inherently transient.