HFTC is next week, and it looks like a great lineup. This week, I skimmed through about 40 of the papers that will be at the conference, and read some of them closely. Here are some papers that I found particularly interesting: “Monitoring the Pulse of a Well through Sealed Wellbore Pressure Monitoring…” by Haustveit, Elliott, et al. (Devon), “Proppant Flowback: Can we Mitigate the Risk?” by Chuprakov, Balyokova, et al. (SLB), “Developing Upscaling Approach for Swarming Hydraulic Fractures…” by Fu, Morris, et al. (LLNL), “Acoustic Imaging of Perforation Erosion…” by Robinson, Littleford, et al. (DarkVision, Encana, and Petronas), “Evaluating Limited Entry Perforating & Diverter Completion Techniques…” by Murphree, Kintzing, et al. (SM Energy, DarkVision, and Halliburton), “Investigating Near-Wellbore Diversion Methods for Refracturing Horizontal Wells” by Zhang, White, et al. (ConocoPhillips), and “A Cost-Effective Evaluation of Pods Diversion…” by Ugueto, Huckabee, et al. (Shell).
But – this is the ResFrac blog, so I mainly want to talk about the two papers where we are coauthors. Naturally, they’re two of my favorites. I mean, I can’t help it – everybody thinks their baby is the cutest!
Our two paper are: “Nuances and Frequently Asked Questions in Field-Scale Hydraulic Fracture Modeling” by McClure, Picone, et al. (ResFrac Corporation) and “Case History of Drainage Mapping and Effective Fracture Length in the Bakken” by Cipolla, Litvak, et al. (Hess Corporation). Discussed below.
In our conversations with operators and colleagues, we’ve noticed that there are certain ‘hot button’ topics that come up over and over. The “Nuances and FAQ” paper lays out our perspective on these topics.
First, we discuss the interrelationship between data-driven and physics-based models. We believe that both are important, and they are complementary. Data-driven approaches recognize patterns and trends. But they cannot predict ‘out of sample’ behavior. Operators may have pumped hundreds of thousands of frac stages, but this data isn’t as ‘big’ as it seems. Most of these stages used similar designs, with gradual changes over time. As a result, only a tiny fraction of the full multidimensional design space has been explored. There are challenges like correlated inputs, confounding factors, and geologic variability. Physics-based models can predict out of sample behavior and thus guide companies to find new designs that are different from what was used previously. They can also help make sense out of new diagnostic tools. But they are less efficient at integrating large volumes of data. Also, there is not universal agreement about the physics of hydraulic fracturing, and so physical models must be used with ‘critical thinking’, and not as a black box. Both statistical and physics-based techniques are important for operators, and they are complimentary. We cite a wonderful paper by Starfield and Cundall, “Towards a Methodology for Rock Mechanics Modeling” (1988). My favorite line from the paper: “a model is an aid to thought, rather than a substitute for thinking.”
Second, we talk through our overall workflow. Our workflow is standard, and widely used. But workflow is important, and so it is worth repeating. (1) Set up a geologic model, wellbore geometries, and boundary conditions (injection schedule and production schedule). (2) Run an initial model and compare with data. (3) Modify model inputs to achieve consistency with the data. (4) Predict response to new designs. (5) Look back and subsequently compare predictions with results, and iterate.
Third, we discuss this hot topic of ‘planar fracture’ modeling versus ‘complex fracture network’ (CFN) modeling. Planar fracture models are the classical approach for modeling hydraulic fracturing. CFN models became popular with the rise of shale, because of the idea that we need to create a lot of fracture surface area to maximize production (I even wrote my PhD thesis on the topic). However, recent studies with direct, in-situ observations have led to a rethinking of the conventional wisdom.
Core-across studies have shown that hydraulic fractures are very densely spaced, but are largely linear and consistently oriented in the direction of SHmax. The paper “Analysis of a Drained Rock Volume: An Eagle Ford Example” by Raterman, Liu, and Warren (ConocoPhillips) at URTeC-2019 was a watershed. They cored an offset well through an SRV. They encountered a very large number of hydraulic fractures, far more than the number of perf clusters. However, RA tracer indicated that only a small percentage of them contained proppant. The fractures were consistently oriented in the direction of SHmax – they were not in zig-zagging flow pathways, as hypothesized by the ‘complex fracture network’ modeling approach. They placed pressure gauges outside the (cased and cemented) offset well and produced. Over time, they observed that the amount of pressure depletion at the pressure gauges was closely predicted by the distance from a propped fracture – depletion was uncorrelated with any other variable. These observations, in my view, strongly support the planar fracture modeling approach. There is only a small number of major, mostly linear, proppant-filled fractures. Their spacing is irregular, but there are roughly as many major fractures as perf clusters. The secondary, unpropped fractures contribute to leakoff during fracturing, but not production (except perhaps for very near the well). Also supporting this view, recent papers like “Can you Feel the Strain…” by Ugueto, Huckabee, et al. (Shell) at the 2019 ATCE demonstrate direct fiber optic observations at offset wells showing that fractures appear to be propagating in a narrow corridor, linearly in a consistent direction, and not branching out into a wide region of fracturing. Without question, not every formation is the same. But observations appear to suggest that in the great majority of major shale plays, fractures are not branching into broad, zigzagging flow pathways. This explains why operators have moved away from openhole completions and long stages, and towards tight cluster spacing with limited-entry.
With planar fracture modeling, we do not claim that this is a literal representation of the exact fracture geometry. Rather, the argument is that if you zoom out to reservoir scale, the fracture geometry looks like a planar fracture. We use constitutive equations to capture field-scale consequences of these sub-model-scale processes. For example, the Fu et al. paper at HFTC this year discusses specifically how we might modify flow equations if we consider that ‘one’ planar fracture in a model actually represents a population of several strands.
As an example of ‘constitutive relations to capture small-scale processes,’ our HFTC paper discusses how we model a ‘proppant trapping’ process, where we hypothesize that proppant gets hung up on small-scale irregularities and trapped as it flows through the fracture. We discuss direct and indirect evidence for this mechanism. At this year’s HFTC, there’s a new paper (that I just read yesterday) that provides support for the ‘proppant trapping’ mechanism – “A Data Analytics Framework for Cored Fracture Imaging…” by Maity and Ciezobka (GT). They discuss results from the HFTS core-through study and how ‘local screenout’ is observed in the fractures, correlated with fracture roughness. This process in included in ResFrac modeling, and we consider it to be an important part of achieving part of achieving realistic results. We came up with this approach back in 2018, when ResFrac Corp was first getting started, based on conversations with engineers in QEP.
Our paper goes through several other topics, which I won’t recap here. These topics are: fracture length and toughness, asymmetry, integration with RTA, impact of net pressure, ISIP trends along the wellbore, and the effect of low stress zones.
We are also coauthors on the paper “Case History of Drainage Mapping and Effective Fracture Length in the Bakken” by Cipolla, Litvak, et al. (Hess Corporation). Most of the work was done by the other authors. We contributed on the DFIT interpretation and by providing a supplementary modeling study using ResFrac, which was performed after most of this work had already been done. We will probably write a follow-up paper giving more detail on the ResFrac work.
It is a really cool, rather unusual dataset. They have a Bakken well that has been producing for ten years. As part of a field pilot, they drilled two vertical wells nearby, performed a series of DFITs, and hydraulically fractured one of them (with proppant and HVFR). The fracture treatment at one of the vertical wells hit the original horizontal. Simultaneously, they reinjected into the horizontal well. This scenario is well-suited for ResFrac because it involves reinjection/reopening of fractures at the original lateral, frac hits, poroelastic stress changes, and DFITs. Because ResFrac is a genuinely integrated fracture and reservoir simulator, it can model the entire history of all the wells, over many decades in a single, full-physics simulation. We watch how pressure and stress change during production, calibrate inputs such as Biot coefficient, describe the multiphase flow of the reinjection into the horizontal and the frac hit, and model how these changes impact the DFITs and the fracture treatments. Pretty cool stuff! The dataset provides a great testing ground and calibration for the ResFrac model.