Using Artificial Intelligence to Predict Contamination During Formation Fluid Sampling

Formation fluid samples acquired using advanced wireline or LWD tools are the most representative because the reservoir fluid is maintained in single phase all throughout the acquisition. The foremost issue with these samples is the mixing of formation fluid with the mud filtrate used while drilling. While immiscible fluids can easily be differentiated and tracked, miscible fluids make it difficult to quantify the fraction of mud filtrate in the mixture. Until recently, the cleanup trend in miscible fluids was thought to be along an exponential curve. Fitting an exponential curve to the fluid properties showing the cleanup trend would provide the current contamination level and could then be used for predicting the time and volume needed to achieve a desired level of cleanliness. Recent studies, as well as simulation studies, though, have debunked the idea that a single exponential fit can explain the cleanup trend completely. The early and late time data fit appears to have different constants and exponents. Thus, explaining the cleanup process for all reservoir and wellbore conditions with a single exponential parameter is subjective, highly user-dependent, and nonrepeatable. To overcome these challenges, an artificial intelligence (AI)-based method has been developed using more than 15,000 fluid sampling simulation models. These simulation sets represent the sample cleanup process in varying reservoir, wellbore, and drilling fluid invasion conditions. The trends observed in actual measured data are then compared to the trends from the simulated data to determine a statistically significant number of best simulation model sets, which are then used to determine the expected contamination level and the associated uncertainty. The cleanup process of miscible fluids may vary significantly due to the potentially wide range of rock and fluid properties and mudcake efficiency. The AI technique presented in this study is based on the fluid flow dynamic and accounts for the most critical variations in reservoir conditions. As a result, it has been successfully applied to estimate the contamination level in real time during the sampling process, and the determined contamination levels are within the range of those from PVT lab analysis. This new method, using an advanced AI-powered algorithm, offers a more robust, reliable, and repeatable analysis of contamination levels over previous methods.