Artificial Intelligent for Real-Time Prediction of Rheological Drilling Mud Properties

Abstract

Rheological characterisitcs are very important for drilling fluid functions since it is directly influenced the cutting transportability, hole cleaning ability, lubricity, and filtration loss behaviour while drilling. The Plastic Viscosity (PV), Yield Point (YP), Apparent Viscosity (µa), and gel strength are mud's rheological attributes that can contribute to achieve the drilling fluid functions, but they are only measured once or twice a day, leading to insufficient performance of the drilling fluid functions. In contrast, mud density, marsh funnel, and solid percent are frequently and continuously measurable. There are few attempts to link between the most frequent mud measurements to the less frequent mud measurements which also have a high error rate in their predictions. The Artificial Neural Networks (ANN) and multiple regression analysis (MRA) were conducted using real field measurements to develop new models to determine the rheological mud properties. The ANN was optimized using training and validation datasets of 70% and 30%, respectively. The coefficient of determination (R2) and the root mean square error (RMSE) are found for each model. Upon creating a plot of the predicted PV versus the actual PV, the obtained R2 and RMSE of the ANN model are 0.955 and 0.694 for training datasets, and 0.954 and 0.634 for validation datasets, respectively. In contrast, the results of the MRA are R2 of 0.91 and RMSE of 0.914. The results of other models for determining YP, µa, and gel strength are also successfully checked. These results showed that ANN is able to predict the rheological mud properties with high accuracy, which outperformed the obtained results of MRA. The presented models will help to track in real-time the rheological mud properties that allows better control for the drilling operation problems.