Inverse Modeling for Fluid System Characterization Through Machine Learning Algorithms

Abstract

Abstract Inverse fluid modeling facilities designing drilling, spacer, cement or fracturing fluids. This approach uses several test data available at the laboratory databases for fluid characterization. Without machine learning algorithms it is very hard for fluid engineers to rationalize a plethora of fluid system information, transform them into knowledge and make intelligent and cost-effective decisions to design a fluid system with desired properties. Therefore, trial and error method is considered to be very costly and misleading. Today, there is a need for an expert system which uses all the available fluid design data stored in a database by which we can benefit from its insights for smart fluid designs. This predictive tool suggests a composition for drilling fluids, spacer fluids, fracturing fluids and cement slurries by implementing a machine learning algorithm on imported experimental data. This study investigates the inverse modeling fluid characterization through machine learning algorithms. A laboratory dataset of measured fluid properties and their corresponding fluid system characteristics are employed as training data to the machine learning routine. GPR as a machine learning method considerably reduced the costs of testing, optimized the material use, integrated available experimental data and eliminated the user bias. This practical nonlinear regression method fosters an efficient and fast prediction analysis which do not require including complex physics of the underlying intricate chemical fluid behavior while integrating all available data from different databases.