Computational intelligence models for particle direction and velocity estimation in multi-phase flows

While in the early days, studies on cuttings transport are made for steep wells and single phase fluids; nowadays they are made for angled drills and multi-phase flows. In the multi-phase flows, motion direction and velocity of the particle are two important aspects for estimating the behaviour of the drill particle. However, in the literature there is still a lack of information on the estimation of direction and velocity of the particle in multi-phase flows. Main aim of this study is to try to predict the the direction and velocity of cuttings in eccentric tubes. For this purpose, we studied on the data which is obtained from the experimental setup in Middle East Technical University, Petroleum and Natural Gas Engineering Department, Cutting Transport and Multi-Phase Flow Laboratory. Direction and velocity characteristics of cutting transport is tried to be predicted for tubes with no angle (horizontal), 15 and 30 degrees of angles. In order to create a prediction model there are several methods are used from conventional methods such as linear and nonlinear regression to computational intelligence methods such as neural networks, support vector machines and support vector regression. Results obtained from these methods are compared to each other by using some performance metrics. It is found that for both direction prediction and velocity prediction, support vector machines and regression are the most successful methods among others.