Wireline formation testers (WFT) play a key role in reservoir characterization and are the first tool to make dynamic contact with reservoirs and collect reservoir fluid samples. However, there is an industry challenge to collect clean samples due to mud filtrate contamination. Because of that, the time to sample could last from hours to over half a day, which can scale up for long reservoirs with multiple sampling depths. Therefore, this work aims to reduce the CO2 footprint and risks related to sampling jobs by optimizing the sampling time.
The workflow developed uses petrophysics information, usually acquired in previous logging runs, to apply a machine learning proxy that generates a smart guide with optimal sampling flags. This is then used to define in the acquisition program the best depths for sampling based on the depth of the drilling fluid invasion and the optimal WFT technology for maximum efficiency. The workflow integrating the petrophysical data, the machine learning proxy, and operations planning are used to reduce the cleanup time to collect a low-contamination reservoir fluid sample, therefore reducing the CO2e emissions for achieving the same quality of reservoir fluid sample.
The study has been conducted in pre-salt wells, including two of the largest reservoir producers in Brazil; for this, we integrated a quick petrophysical interpretation with the machine learning proxy to create a smart guide to help the decision process of sampling depth. The machine learning proxy has been trained on 350 thousand reservoir simulations, reflecting a remarkably high accuracy. With petrophysics and the predicted depth of invasion, we use numerical reservoir simulation to estimate and match the actual time to sample and compare it to nearby depths with similar reservoir quality but a shallower invasion. The results show that for achieving the same quality of the fluid sample, when using this smart guide in the operations planning, there is a potential to reduce up to half of the sampling time per well, therefore reducing the same amount of the CO2 footprint from these activities related to Scopes 1 and 3 from Greenhouse Gas Protocol, and the risks related to tool imprisonment inherited from this type of job.
The study presented here, enabled by a machine learning proxy, provides a method to improve sampling operation efficiency and contributes to achieving safer, economically improved, and lower CO2 footprint operations. When reducing the cleanup time, the proposed methodology addresses one of the main challenges in decision-making when designing an acquisition program due to the long-lasting sampling jobs.
