Modeling naturally fractured reservoirs: from experimental rock mechanics to flow simulation

Fractures have a big impact on reservoir production but are inherently difficult to quantify. This study gives a robust and practical workflow to obtain a mechanically consistent naturally fractured reservoir model without direct sampling of the fracture network. The three tiers of the workflow are: 1) subcritical testing, 2) geomechanical modeling, and 3) flow modeling. Subcritical fracture index, a rock property, has been shown to influence fracture attributes such as length, spacing and connectivity. Subcritical tests show that the average subcritical index for sandstones in ambient air is around 62, whereas the average value for microstructurally comparable carbonates samples is 120. Thin-section analysis shows that an increase in cement content increases the subcritical index. Furthermore, sandstone samples containing more than 15% carbonate cement, sandstone samples containing more than 40% clay, and pure carbonate samples exhibit a large drop in subcritical index when the environment is changed from ambient air or oil to fresh water or brine. Geomechanical modeling has shown that the mechanical bed thickness has a large influence on fracture pattern characteristics and has the potential to overshadow fracture pattern changes due to strain level, strain anisotropy and subcritical index. Furthermore, an increase in strain anisotropy reduces the number of dominant through-going fracture sets and decreases the fracture spacing between the through-going fractures. This spacing variation not only influences the preferential drainage direction, it can also enhance the drainage efficiency, because more rock is exposed to the through-going fractures which are more likely to be intersected by a borehole. The level of detail provided by the geomechanical model greatly exceeds the level of detail normally used in reservoir simulation. Therefore, upscaling of the geomechanically generated fracture patterns is necessary for practical flow modeling. This study shows that different upscaling methods can lead to large variations in permeability prediction. A Local Grid Refinement around the well should be maintained, because it will almost always ensure accurate production prediction. This method is preferred over the dual permeability approach, which can be calibrated to match production data in some cases, but often requires using an unrealistic representation of the fracture pattern.