Obtaining representative clean fluid samples in the least amount of rig time is the primary objective for open-hole sampling. The estimation of contamination levels within the pumpout fluid in real time poses major challenges to accomplish this operation, particularly during the early stages of field exploration and appraisal. The objective of this paper is to introduce new real-time monitoring and control algorithms to improve sampling quality and improve estimates of key formation properties such as pore pressure, mobility and relative oil/water saturation effects using machine learning based on an extensive database created from a parametric simulation study. A database can be created from field data but is typically too sparse to be used to create an analysis method, as many of the parameters are unknown.
After detailed analysis of formation pumpout cleaning behavior and oil-well sampling, a parametric study was designed and used to conduct an extensive matrix of simulations to generate the comprehensive database. The study was required to determine the sensitivities to various parameters related to sampling and contamination. The results from this database were used to create a new tool for real-time monitoring and control. The database contained the simulated temporal cleaning trends of pumpouts over a variety of reservoir and operating conditions. Parameters affecting the cleaning behavior during the formation testing pumpout were evaluated and incorporated into a three-dimensional (3-D), compositional simulation model. Reasonable variations of the parameters were selected based on a comprehensive study and data from available literature. The parameters included reservoir rock and fluid types, reservoir pressure and temperature, different oil-based muds (OBM), active mud-filtrate invasion, and operating conditions. Nearly one hundred thousand scenarios were evaluated based on full factorial and a one-factor-at-a-time (OFAT) experimental design and were subsequently used to study the effects of each parameter and its variations.
Statistical tests, such as mutual information and analysis of variance (ANOVA), were used to determine the significance of different parameters on the simulation results. These simulations reveal a more complex pumpout behavior than has previously been published. This paper discusses the effects of different parameters on the cleaning process and uncertainty analysis for various scenarios. Trends of density and contamination during pumpout are evaluated, and new guidelines and equations are provided for trend-fit cleanup prediction. Additionally, the effect of active mud-filtrate invasion is considered and its effect on endpoint contamination is described (i.e., lowest contamination possible). The workflow and data can be used for pre-job planning as well as during real-time operations with various wireline-formation-tester (WFT) and logging-while-drilling (LWD) tools to optimize cleanup and sampling of formation fluids. Simulations of different realizations of reservoir properties, drilling mud invasion profiles, and cleanup operations also helped develop a useful and diverse cleaning behavior database for data-driven modeling for a variety of reservoir and operating conditions.
