Towards Robust and Automated Contamination Estimation During Downhole Fluid Sampling

Abstract

Accurate, robust, and real-time quantification of oil-based mud (OBM) filtrate contamination using data from downhole fluid analysis (DFA) sensors remains a challenging problem, especially under difficult sampling conditions and for advanced sampling tools with complex inflow geometries and active guarding of filtrate flows. Such tools and conditions lead to contamination cleanup behaviors that do not necessarily follow the simple power-law models that are commonly used in contamination monitoring algorithms. Moreover, the need for improved efficiency calls for more automation in sampling operations, especially in high-activity regions with many wells and sampling stations. Achieving such automation within existing contamination monitoring workflows has been proven to be challenging. In this paper, we deploy a new method for contamination monitoring, based on the inversion of DFA data using a full 3D numerical flow model of the contamination cleanup process. This flow model predicts the evolution of filtrate contamination as a function of time and pumped volume and can thus be used to forward-model the DFA sensor responses. Sensor data, such as optical density, mass density, and gas-oil ratio, are then inverted in real time to provide contamination predictions. Real-time computation is enabled through fast, high-fidelity proxy models for the cleanup process. The proxy models are trained on and validated against a library of 300,000 precomputed full-scale numerical simulations covering a wide range of hardware configurations and sampling environments. We applied the new method on 14 field sampling stations of oil sampling in OBM conditions, with four different types of sampling inlets: single probe, radial probe, dual packer, and focused radial probe. The data sets also included challenging stations with cleanup conducted over multiple 4-hour sessions with possible reinvasion of mud between sessions. We found that contamination predictions with the new method are in good agreement with results from laboratory analysis. Compared to the previous algorithms, the new contamination monitoring algorithm is applicable for all types of sampling hardware, a wider set of operating conditions, and over the full history of a sampling station (i.e., from formation fluid breakthrough to a clean sample). This latter attribute enables robust early-time predictions of the pumping time remaining to reach a clean sample. Finally, the new method is independent from manual user intervention and thus lends itself well to workflow automation.