Whipstock casing exits are a milling operation that enable operators to sidetrack from a primary wellbore. This paper describes a prescriptive data analytics workflow that was developed and applied to optimize casing exit applications. This workflow involves three major steps:

  1. Pre-processing data to cleanse, transform, and aggregate data from past casing exit jobs

  2. Building accurate machine learning models to predict milling performance

  3. Optimizing a weighted objective function into recommend best-case operational parameters.

In collaboration with a North Sea operator, a downhole measurement-while-drilling (MWD) system was used in multiple wells during casing exit jobs to collect a rich data set of downhole measurements. Through a stringent data processing and modeling methodology, prescriptive models were developed and tested in an offset well in the North Sea. Success of the offset casing exit job resulted in greater than a 30-percent reduction in vibration, a 14-percent increase in rate-of-penetration (ROP), and a 23-percent reduction in average mill time.

Keywords:
Artificial Intelligence, machine learning, optimization problem, surface weight, whipstock length, Upstream Oil & Gas, Downhole Measurement, objective function, correlation, recommendation
Subjects:
Drillstring Design, Drilling Operations, Drilling Measurement, Data Acquisition and Automation, Information Management and Systems, Directional drilling, Measurement while drilling, Data mining, Artificial intelligence
Copyright 2020, IADC/SPE International Drilling Conference and Exhibition
You can access this article if you purchase or spend a download.