Abstract
Introducing rotary steerable systems (RSS) to the drilling industry has extended the directional drilling envelope to new horizons. Its role in better hole cleaning, faster rate of penetration (ROP), and less stuck pipe incidents is unquestionable. However, it comes with more economical challenges in terms of operating rates and Lost-in-Hole (LIH) charges. Many factors control the operator decision to run RSS rather than positive displacement mud motors (PDMs) — which have been the industry standards for decades — and vice versa. The intent of this paper is to introduce an advisor system based on machine learning that makes the selection process easier and more straightforward.
The system predicts total section time (including casing running) and cost for both RSS and PDM, the user may use either section time or cost criterion to assess which technology is preferred based on drilling campaign needs. The system input features include offset data of interval measured length, formations encountered, ROP, dogleg severities, operating cost of both technologies, rig operating cost, and many other variables. Input data to the system are divided into training and testing data for more reliable modeling. Many algorithms are tested to avoid over-fitting or under-fitting of training data.
The past performance of both RSS and PDM technologies in different formation plays a major role in determining the result. The result comes in both visual and tabulated forms to give the drilling engineer both quick impression and detailed insights about the selection process. As the result depends on formation-specific performance, it would change between wells depending on the formations thickness at these wells. This means that the result is not fixed per field and may vary considerably based on small changes of input.
The use of this advisor system should help in speeding up operator companies’ decisions and improving drilling performance and cost-effectiveness. This occurs through automating the whole process and making best use of offset drilling data thought implementation of machine learning algorithms.
