Experimental and 3D Numerical Investigation on Proppant Distribution in a Perforation Cluster Involving the Artificial Neural Network Prediction

Summary Uniform proppant distribution in a cluster and a stage with multiple clusters is a primary objective to optimize fracturing parameters and improve the production from each cluster. Because fracturing slurry is typically pumped at high pressure and rate in fields, it is a big challenge to study proppant transport behavior and distribution characteristics through laboratory experiments. There is still a lack of an effective model to quantitatively evaluate proppant distribution based on an actual wellbore configuration. The objective is to propose a novel method to accurately evaluate the distribution uniformity and quickly optimize fracturing parameters based on field conditions. This paper conducts particle transport experiments in a horizontal pipe with six holes at the helical distribution. A 3D numerical model coupling of the computational fluid dynamics (CFD) and discrete element method (DEM) is used to study proppant distribution. Proppant distribution is quantitatively evaluated by the proppant transport efficiency (E) and normalized standard deviation (NSD). The effects of 10 parameters are investigated. An artificial neural network (ANN) model is developed to predict proppant distribution in a cluster. The results identify that proppant distribution among perforations is generally toe-biased in a horizontal wellbore due to a high pumping rate. Proppants with large inertia easily miss the heel-side holes and are suspended to the toe side. The complex vorticity flow carries them to the toe-side perforation regardless of hole orientation. Fluid distribution can significantly change proppant distribution regardless of fluid velocity. The heel-biased fluid distribution leads to the same bias of proppant, and the downward perforations receive more proppants. Proppant transport reaches equilibrium quickly, and the distribution is hard to change unless the injection condition varies. It is a good choice to increase fluid viscosity, add perforation sealers, and inject small mesh proppant, especially for the low density. The ANN model trained by extensive experimental and numerical samples can accurately evaluate proppant distribution uniformity. The study provides an efficient way to optimize injection parameter design and achieve real-time optimization coupled with the fiber-optic downhole diagnostic. It can be a crucial part of artificial intelligence hydraulic fracturing.