This study presents an AI-driven approach to stage categorization for hydraulic
fracturing operations, addressing deviations from pumping designs that result in
increased costs, reduced efficiency, and operational unpredictability. The dataset
includes 1,595 stages, of which 970 exhibited at least one issue such as mid-stage
shutdowns, screen outs, and rate reductions. The objective is to improve
operational decision-making by automating stage categorization, distinguishing
between surface and subsurface issues, and lay the foundation for future
predictive machine learning models to anticipate trouble stages.
The stage categorization model evolved through eight iterations to refine detection
of operational issues and differentiate between surface- and subsurface-driven
problems. Version 1 implemented basic pressure slope algorithms, while Version 2
improved accuracy with steady-rate pressure checks. Version 3 expanded
diagnostic capabilities by incorporating proppant and chemical concentrations.
Version 4 adjusted the search for specific activities like pad for ball-seat stages,
flush, and pressure test. Version 5 introduced dynamic thresholds to account for
friction effects from varying casing sizes and stage measured depths. Version 6
optimized the overall model, removing factors that reduced Precision, resulting in
significantly improved performance. Version 7 improved Clean Sweep remediation
detection using stage activities, such as flushes, to determine the most accurate
period to scan. Screen outs were updated based on a sensitivity analysis around
the required pressure per minute reading. Finally, in Version 8 logic was added to
connect subsequent subsurface-related issues to one that occurred earlier in the
stage’s progression.
The finalized stage categorization model analyzes time-series data, including
pressure, rate, proppant and chemical concentration, and wellbore design
parameters, capturing key variables affecting treatment behavior. It achieved 98%
accuracy and 88% precision in identifying issues like mid-stage shutdowns and

screen outs. By distinguishing surface from subsurface problems, operators can
optimize resource allocation and treatment designs. The study also revealed
significant correlations between operational challenges and geological variability,
highlighting the importance of integrating rock and operational data. These
insights facilitate improved treatment consistency, reduced costs, and enhanced
decision-making.
Bio:
Jessica Iriarte is the General Manager of Completions at Corva. Jessica is a data
science and energy leader, has held various leadership positions in oil and gas,
including international experience in data, research, and operations. Jessica is an
inventor, a distinguished lecturer, and has 17 publications with SPE, JPT, and URTeC.
Jessica holds a Bachelor of Science degree in Petroleum Engineering from
Universidad del Zulia and a Master of Science degree in Petroleum Engineering from
Colorado School of Mines.