Modeling Lost-Circulation in Fractured Media Using Physics-Based Machine Learning

Abstract

This work provides a novel machine learning approach to model lost-circulation in a naturally fractured formation. As input, the modeling tool requires the observed mud rate, mud physical properties, pressure flowing bottom-hole condition, if available. The deep neural network tool is trained using a physics-based model based on full-physics Cauchy momentum equation for non-Newtonian fluid, which can serve an accurate and quick estimate of the effective hydraulic aperture of natural fracture, and predictions for cumulative mud loss volume, and final stopping time leakage behavior. Such information can help take the preventive/corrective decision, such as the optimum drilling additive design for the lost circulation material. To our best knowledge, the proposed machine learning workflow is applied for the first time for modeling lost circulation events in fractured formations.