Prediction of Rheological Properties of Recirculating Water-Based Drilling Mud in Geothermal Exploration using Artificial Neural Networks with Tensor Flow

Abstract

Pipe sticking in drilling operations occurs by mechanical and differential forces caused by the loss in circulation of drilling fluids or muds. Rheological properties of recirculating drilling muds (RDMs) determine the drilling hydraulics and hole cleaning effectiveness. Uncertainty in real-time data on the rheological properties of RDMs challenges estimation of potential pipe sticking problems, a problem that can be improved using machine learning methods. This study reports first-time application of a supervised artificial neural networks (ANN) algorithm with TensorFlow to estimate plastic viscosity (PV), apparent viscosity (AV), yield point (YP), flow consistency index (k), and flow behavior index (n), five rheological properties of recirculating water-based drilling mud (R-WBDM). Model input variables were density (MD), marsh funnel viscosity (MFV), and percent solids content (% S) of R-WBDM. Model performance was tested using the root mean square error (RMSE) and coefficient of determination (R2) methods. Sensitivity analysis demonstrated the strength of each predictor variable. Five optimal models showed good generalization capability in estimating the n, k, PV, AV, and YP with RMSE of 0.022, 0.270, 7.890, 8.870, and 10.149, and R2 of 0.995, 0.956, 0.756, 0.724, and 0.701 in a similar order, respectively. High sensitivity was observed in PV, AV, YP, and k models, and n-model, to changes in MFV (0.615) and % S (0.067), respectively. Results show the potential use of ANN with TensorFlow to support decision-making in geothermal drilling engineering on pipe sticking problems using easy-to-determine physical properties as predictor variables.