Predicting filter cake thickness in drilling fluids using machine learning techniques

Predicting filter cake thickness in drilling fluids is critical for improving drilling progressions and minimizing operational subjects such as pipe sticking and reduced permeability. This study investigates the performance of several machine learning models, including Decision Tree, Random Forest, AdaBoost, MLP-ANN, and Ensemble Learning, for accurately modeling filter cake thickness. A dataset of 354 experimental samples, derived from peer-reviewed studies, was employed to assess the relationships between input parameters such as nanoparticle type, nanoparticle concentration, salinity, temperature, and polymer characteristics. Model evaluation was performed using metrics such as Mean Squared Error (MSE), Coefficient of Determination (R²), and Average Absolute Relative Error Percentage (AARE%). Results indicate that the MLP-ANN model outperformed other algorithms, achieving an R² of 0.9269 and an MSE of 0.0741 during testing. Cross-validation was implemented to ensure robust model training and evaluation, reducing overfitting observed in models like Decision Tree and AdaBoost. Additionally, SHAP investigation recognized nanoparticle concentration and type as the most influential factors impacting filter cake thickness, revealing their negative correlation with the target variable. These discoveries highlight the potential of advanced machine learning procedures to enhance drilling fluid design by identifying key parameters and optimizing formulations to reduce filter cake thickness.