Development of a modified Bourgoyne and Young model for predicting drilling rate

Abstract Drilling in carbonate rock is quite challenging as the formation is permeable and porous, hence, leads to the possibility of having mud losses. Since drilling into loss formation requires controlling some of the operational drilling parameters, the rate of penetration (ROP) prediction in this area, especially in carbonate formation could be challenging. As a result, the study aims to improve the accuracy of ROP prediction in carbonate loss zones by adjusting the Bourgoyne and Young (B&Y) model, which is the most comprehensive model to date and has been used until now but is missing this feature. Modification of the drilling rate model is performed based on loss severity levels by including cutting transport ratio (RT) parameter and eliminating multicollinearity parameters. The considered elements in the modified model are loss severity level, weight-on-Bit (WOB), rotary speed (rpm), bit wear and modified hydraulics with cutting transport ratio. The North Kuwait field's drilling data was used for multiple regression analysis (MRA) with categorical coding (1,0) and generated the model. A total of 80 drilling datasets were used for model development and another 38 datasets for model verification. The modified model could predict the ROP for different loss severity levels in carbonate formation based on the results. The original model yielded an R2 value of 0.52, while the modified model yielded an R2 value of 0.81 for the first set of data, resulting in an ROP accuracy improvement of about 29%. Meanwhile, the original and modified models yielded R2 values of 0.57 and 0.88, respectively, which improved the model accuracy by 31% for the second set of data. The modified model accurately predicted the ROP with a mean absolute error (MAE) of 3.21%, 4.08% and 3.50% for seepage loss, partial loss and severe loss, respectively and outperformed all the existing four ROP models. Also, this modification reduced the number of constants used for the regression analysis. Consequently, a reduction in the number of required data points and elimination of the regression process's multicollinearity effect was attained.