Prediction of Cutting Concentration in Horizontal and Deviated Wells Using Support Vector Machine

Abstract

Abstract Hole cleaning, or drilled-cutting transportation, is one of the main concern in the petroleum industry. This is due to the high impact of improper downhole cleaning during drilling operations. To illustrate, many drilling problems can happen because of inefficient cleaning of the wellbore. These problems may include premature bit wear, slow drilling rate (i.e. low ROP) and, in most severe cases, a stuck pipe which in some cases can lead to complete loss of the well. Moreover, measuring the cleaning efficiency using field or experimental measurements is highly costly and time-consuming which makes it a very complicated problem. Therefore, a lot of studies have been conducted to understand cutting transport efficiency in drilling operations. However, most of these studies are experimental and try to come up with the best measures including experimental models or empirical correlations. In this study, artificial intelligence techniques were used to estimate the cutting concentration in the wellbore. The purpose of this study is to use Support Vector Machine (SVM) technique to indirectly measure the hole cleaning efficiency by predicting the cutting concentration in the wellbore from other operational parameters. Based on 116 experimental data points collected from the literature, the cutting concentration in the borehole was predicted from the properties of the mud itself such as the mud rheological properties (e.g. yield point and plastic viscosity) and mud density (mud weight) and other operational parameters during drilling including the drill pipe rotary speed (RPM), pipe eccentricity (i.e. the axial location of the drill pipe), hole inclination angle, the rate of penetration (ROP), flow rate (GPM), temperature and annular size. The results obtained from SVM show the ability of this method to accurately predict the cutting concentration in the wellbore with an average absolute error (AAE) of less than 5% and a correlation coefficient (R) of more than 0.9. For this specific group of data, comparing the results obtained from this technique with a correlation presented in the literature shows that the SVM method provides a better prediction of cutting concentration and higher accuracy than that in the literature. Finally, the method developed in this study to predict the cutting concentration is based on continuously measured parameters during any drilling operation. Therefore, integration of the developed model into the drilling system will allow for real-time prediction of the concentration of the cuttings (i.e. the amount of cuttings present in the wellbore) and, hence, the cleaning efficiency during the drilling operation.