Casing Failure Using Machine Learning Algorithms: Five Case Studies

2018 Society of Petroleum Engineers. All rights reserved. Recent casing failures in the Granite Wash play in the western Anadarko Basin have sparked deep concerns to operators in North Texas and Oklahoma. Hydrostatic tests made in the field show that present API standards do not assure adequate joint and bursting strength to meet deep-well requirements. Past and present literature has been infested with numerous casing failures incidents. Despite the extensive documentation and recommendations, a mounting trend of failure is still on the rise. In an attempt to find possible solutions for these failures, this study is a continuation of an on-going effort to minimize the likelihood of failure using Data Mining and Machine Learning algorithms. The study applied both descriptive visual representations such as Mosaic and Box Plots and predictive algorithms including Artificial Neural Networks (ANN) and Boosted Ensemble trees on eighty land-based wells, of which 20 possessed casing and tubing failures. The study used a predictive analytics software and python coding to evaluate twenty-six different features compiled from drilling, fracturing, and geologic data. This work attempts to shed light on operational problems and implement a data analytic approach to find out the possible factors contributing to casing failures using both descriptive and supervised ML algorithms.

Casing Failure Using Machine Learning Algorithms: Five Case Studies Conference Paper