Rafail Manafov

Lost circulation remains a major challenge during well construction, often leading to non-productive time, increased material consumption, and additional treatment costs. In field practice, the selection of lost circulation materials (LCMs) is still largely based on empirical rules or laboratory testing; however, these approaches are not always suitable for rapid decision-making under variable downhole conditions. This study presents a physics-guided surrogate modeling framework for predicting fracture sealing performance and supporting injection strategy selection. The approach combines laboratory observations with coupled computational fluid dynamics and discrete element method (CFD-DEM) simulations to represent both measured behavior and a broader range of mechanically consistent sealing scenarios. The final dataset included 300 cases, comprising 45 physical experiments and 255 CFD-DEM-generated synthetic cases. A hybrid machine learning architecture based on Temporal Convolutional Networks and Artificial Neural Networks was developed to predict sealing pressure under different material and fluid conditions. The model achieved an R² of 0.89 and a mean absolute percentage error of 6.4%, while showing 94% agreement with laboratory-based recommendations for injection strategy. The proposed framework can therefore serve as a rapid engineering support tool for preliminary formulation screening and a more computationally efficient digital workflow for fracture sealing design in drilling operations.