Experimental and Simulation Investigation on Ball-Sealer Transport and Diversion Performance Aided by Machine Learning Method

Abstract

Summary Ball-sealer diversion has been proven to be an effective and economical way to increase fractures and fracturing volume in multistage hydraulic fracturing and matrix acidizing treatments. However, designing and implementing a successful ball-sealer diversion treatment is still challenging. Typically, operators rely on empirical data to determine diversion parameters and need an understanding of accurate ball transport and diversion behaviors. A model for optimizing operating parameters, including fluid and ball properties, and predicting the diversion performance of ball sealers before treatment is needed for designing the fracturing process. In this work, we systematically investigated ball-sealer diversion using experimental and numerical methods. The resolved model of computational fluid dynamics (CFD) and discrete element method (DEM) is first developed to simulate the transport of a large ball in a horizontal wellbore with side holes. The experimental results validated the numerical model. The effects of the ball position in the pipe, flow ratio of the hole to pipe, injection flow rate, and ball density on the diversion performance were studied under field parameters. The results show that the ball sealer easily misses the heel-side perforation due to the inertial effect and travels to the toe side due to the large inertia and turbulent flow. The ball position and flow rate ratio are crucial for the diversion performance. There is a threshold value of the ball position under the specific condition, and the ball successfully turns to the perforation only when the threshold distance is met. A ball sealer closer to the perforation will have a larger probability of blocking the hole than the ball at the other side of the wellbore. The larger the flow rate ratio, the more the drag force on the ball, and the ball can successfully divert to the perforation despite the ball being far from the hole. The injection flow rate and ball density negatively correlate with the diversion performance due to the large inertia and gravity. The best classification result with the F1 score of 87.0% in the prediction set was achieved using the random forest (RF) algorithm. It provides new insight into developing ball sealers and adjusting fracturing parameters based on machine learning (ML) methods.