A New Method to Determine Friction Factor of Cuttings Slip Velocity Calculation in Vertical Wells Using Neural Networks

摘要

The slip velocity of cuttings is a calculation of the settlement velocity of a particle in a stagnant fluid. In dynamics, the slip velocity is defined as the difference between the solid and the fluid velocities. If the cuttings velocities are low, the cuttings may not be transported efficiently to the surface, causing a variety of inefficiency issues. For example, they may accumulate at the bottom of the wellbore and get regrinded by the bit. Determination of the correct value of this parameter in vertical wells has been the concern of many investigations in the oil and gas industry. API RP 13D (2010) standard uses Walker and Mayes' study (1975) to calculate the slip velocity for vertical wells but as discussed below, this model does not show accurate results compared to other models. Moore's model (1974) provides better slip velocity calculation compared to Walker and Mayes' model. These methods are based on providing three sets of equations for laminar, transition, and turbulent to correlate Reynolds number to friction factor for an average value of sphericity for the drilling cuttings. This paper describes a new method to determine the cuttings slip velocity in vertical wells using neural networks. Artificial neural networks have been successfully implemented in many disciplines in the oil and gas industry. The interconnected network between the neurons provides a nonlinear approximation of the data to solve complex models. The "Reynolds number vs. friction factor" data at different particles' sphericity graphs has been used to determine the friction factor and then apply it to the Stokes slip velocity equation. Moore's model does not consider the sphericity of the particles because it uses a generic value for this figure, however, the new technique covers a wider range of the sphericity of particles from 0.125 to 1.0. In addition, this model gives higher accuracy compared to the afformentioned models.