Research on the wellbore cleaning mechanism and prediction of cleaning ability of well-flushing fluid based on experiment-molecular dynamics simulation-machine learning

Abstract

In order to reveal the cleaning mechanisms of ultra-deep well drilling fluids and predict the efficacy of flushing fluids in complex conditions, this study combined laboratory experiments, molecular dynamics simulations, and machine learning to investigate the micro-scale cleaning phenomena. A 1 % SEO (temperature-resistant surfactant) solution was used as the well-flushing fluid and demonstrated superior cleaning ability on various drilling fluids. Experimental results revealed that the contact angle of the flushing fluid on metal surfaces ranged from 7° to 16°, notably lower than that of the contaminants, indicating enhanced wettability. Post-cleaning experiment, SEO molecules occupied adsorption sites on the metal, effectively blocking contaminant re-adsorption. Molecular dynamics simulations further demonstrated that the adsorption energy of SEO molecules (–290 kcal/mol to –337 kcal/mol) was substantially higher than that of contaminant molecules (–60 kcal/mol to –300 kcal/mol), promoting a “Stripping-Dissolution” process. Diffusion coefficients for contaminant clusters in the SEO solution were recorded at 1.995 × 10⁻⁶ and 4.723 × 10⁻⁶, highlighting effective dispersal within the flushing fluid. Based on simulated and experimental data, a machine learning-based predictive model for flushing efficacy was developed, achieving an accuracy of over 85 % with the K-Nearest Neighbors (KNN) algorithm. This study offers theoretical guidance and technical support for designing and optimizing intelligent well-washing strategies in oil field operations.