Neural-net Modeling of Shale-swelling Behavior in Aqueous Drilling Fluids

Abstract

Abstract Managing shale swelling is critical when drilling with water-based muds (WBMs). Excessive swelling can lead to shale sloughing, borehole collapse, stuck pipe, and shale disintegration. The fine solids content can increase and cause difficulties controlling rheological properties. Linear swell meter (LSM) testing is a well-known laboratory procedure used to characterize shale swelling in a WBMs. A mathematical modeling tool known as the artificial neural network (ANN) was used to model shale swelling in a WBM. The ANN model establishes complex relationships between a set of inputs and an output based on computational modeling. For ANN modeling of the shale-swelling, the shale mineralogy and fluid composition constituted input parameters, while the output was represented by an experimental characteristic swelling parameter derived from the "% swelling vs. time" data from the LSM test for the respective shale-fluid combinations. Experimental data for building the ANN model was obtained by performing about 250 standard LSM tests on different shales with varying mineralogy and WBMs with varying salt concentrations. This shale swelling ANN model provided excellent correlation with R2 > 0.9. The ANN model was then successfully validated for an independent set of shale-fluid conditions. Using the shale swelling ANN model, shale-swelling in WBM was predicted for a given shale mineralogy and fluid composition. This reduced the number of trials necessary to determine an optimized WBM formulation. Mud engineers can use this model in real-time as the shale chemistry varies with the depth of the formation drilled. The model provides a helpful measure of fluid performance to determine the optimization necessary to obtain the desired shale behavior. Using this method can help minimize drilling risks and costs associated with unpredictable shales.