Data-Driven Experimental Design of Rheological Clay–Polymer Composites

Natural clays are used as thixotropic agents and viscosity controllers. When dispersed in water, laminar clays create a fluid suspension which can be further thickened by incorporating acrylic and cellulosic polymers. Typically, the identification of formulations exhibiting the optimal characteristics for specific rheological applications requires experimental testing of a large number of prototype mixtures, for which yield points and apparent and plastic viscosities are obtained from the measured shear-stress curves. Herein, we report an alternative approach involving high-throughput virtual experiments aimed at discovering high-performing clay–polymer rheological agent formulations at a fraction of the time and cost of experimental campaigns. Specifically, we developed customized feature vectors to represent the rheological formulations and combined them with Random Forest models trained on experimental data points. The latter were identified by latin hypercube sampling with a multidimensional uniformity designer algorithm to generate homogeneous information and amplify the initial historical data set. The final Random Forest models (R² > 0.91) of rheological targets were used to perform multiobjective optimizations of the formulations based on the Pareto front algorithm. The three highest-performing formulations identified by the screening process were prepared, and their water dispersions were characterized in terms of their rheological behavior. The prototypes create viscoplastic fluids with a yield point and apparent and plastic viscosity ranges of 54.5–57 Pa (26–27 lb/100 ft²), 28.5–30 mPa s, and 15–18 mPa s, respectively, reflecting a suitable gel strength to be employed in engineering operations such as horizontal directional drilling.