An End-to-End Solution for In-situ Stress Estimation Using Downhole Drilling Dynamics Data

Knowledge of the in-situ state of stress is essential to ensure safe and cost-effective subsurface operations for carbon storage, geothermal and oil and gas applications. In carbon storage operations, reliable characterization of the in-situ stress state in the complex storage reservoirs and the cap rock formations is critical for safe storage of CO2 and minimizing potential environmental hazards related to fluid leakage and induced seismicity. Also, the integrity of the borehole is directly dependent on the stresses profile along the well trajectory and appropriate design of the drilling and stimulation operations. In-situ stresses are notoriously difficult to determine. Several methodologies have been used to date to estimate in-situ stresses. Seismic-based methods offer the advantage of covering a larger volume of subsurface formations, but do not provide sufficient vertical resolution for the required subsurface characterization, especially for carbon storage purposes. Log-based methods that are commonly used for stress estimation rely on the use of costly well logs such as image, density, porosity, sonic etc., together with oversimplified models or correlations to estimate stresses. Another drawback is the typical absence of the required logs (e.g., dipole sonic) and the lack of them outside the pay zones. Logs are also rarely available in unconventional horizontal wells, where the lateral changes in the state of stress is crucial to optimize the stimulation design. These limitations and shortcomings identify an essential requirement for new methodologies and sources of data for stress evaluations, which provide full coverage of the well length at a higher resolution. This project investigates the use of downhole drilling dynamic data together with advanced signal processing, data analytics and machine learning techniques to calculate subsurface in-situ stresses along vertical, deviated, or horizontal wells in real-time. The promising results of the Phase I study successfully proved the concept behind the proposed technology. We found the frequency band of the signals carry information about the stress field and used it to generate reasonably accurate 1-D profiles of the principal in-situ stresses. Building on our findings in Phase I, our main objectives for the Phase II are: (i) validating the findings over a wider range of data acquired in different rock types, bit types and well trajectories, (ii) exploring alternative signal processing approaches, (iii) extending the model to estimate rock mechanical properties as well, (iv) upgrading the machine learning regressor model to deep neural network, and (v) developing and commercializing a software platform hosting this technology.