The paper presents some important conclusions obtained by analysis of data from many hydraulic fracturing treatments monitored over the past five years. These conclusions involve some major changes from conventional concepts about hydraulic fracturing, many of which have been based on inadequate models of the process and a commensurate lack of adequate data or motivation to check those models. Recommendations are. therefore, also made about simple, low-cost procedures for adequate data collection, such as the use of flow-rate changes and/or multiple injection/shut-in cycles for stringent model evaluation. Our conclusions from such careful analysis include: shorter, wider fractures; relative insensitivity of fracture width to frac-fluid rheology; dangerously fast convection vs. settlement of proppant in imperfectly-contained fractures; and a potential role of proppant in imperfectly-contained fractures; and a potential role of natural fractures in explaining many phenomena formerly regarded as evidence for long contained fractures. Although serendipity may sometimes compensate for poor job design, we recommend that many existing approaches to fracture design and execution be re-considered and that more credible efforts at treatment optimization be achieved by more careful (on-site) analysis of properly monitored and more flexible field execution schedules.
The theory and practice of hydraulic fracturing abound with concepts, analysis techniques and procedures for creating "optimum fracture geometries" in representative reservoirs.
These ideas have evolved somewhat over the past twenty years, as may be seen, for instance, by comparing Ref. 1 and Refs. 2, 3: the most obvious, but still hard-won, recognition has been that fractures are typically not well-contained in the pay-zones of the target formations. However, many other misconceptions and inadequate models or procedures still pervade the industry — despite large amounts of data which, directly or indirectly, refute those approaches. This situation is still possible because adequate data-sets are not often collected and are rarely analyzed properly, if at all; and unsatisfactory explanations are often found when anomalies are noted (e.g. Ref. 4). Indeed, clearly questionable models are still purveyed (e.g. Refs. 2,3) as adequate representations of the physical process: in particular, they do not often come close to matching the actual observed pressures, even in well-defined job situations. Results vary from vague (e.g. Ref. 5) to demonstrably erroneous recommendations for optimum job design. One of the authors (Cleary) has been working on this problem for over twelve years, first conducting laboratory (Ref. problem for over twelve years, first conducting laboratory (Ref. 6) and computer (Ref. 7) simulations, then applying the resulting practical computer-based models (Refs. 8,9) to field applications, practical computer-based models (Refs. 8,9) to field applications, using actual data and on-site analysis (e.g. Ref. 10). Some of our original concepts (e.g. the role of rheology) have changed quite dramatically (e.g. from those presented in Ref. 11), but others (such as the roles of stress, modulus and rock nonlinearity) have been borne out by the data. More importantly, although we can still represent conventional models as special cases of our formulation, the models that we employ for practical analysis have changed a lot from those in Ref. 12, despite the widespread adoption of those early models by industry. These changes have been forced, most of all, by the careful analysis of field data-sets, made possible by a large-scale research effort undertaken by the Gas Research Institute (GRI), especially over the past five years.