Machine Learning Aids Early Detection of Stuck Pipe in Extended-Reach Wells

Abstract

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 206516, “Stuck-Pipe Early Detection in Extended-Reach Wells Using Ensemble Method of Machine Learning,” by Rushad Ravilievich Rakhimov, SPE, Schlumberger, and Oleg Valerievich Zhdaneev, SPE, and Konstantin Nikolaevich Frolov, SPE, Ministry of Energy of the Russian Federation, et al. The paper has not been peer reviewed. The objective of this paper is to describe the experience of using a machine-learning model prepared by the ensemble method to prevent stuck-pipe events during construction of extended-reach wells. The tasks performed include collecting, analyzing, and cleaning historical data; selecting and preparing a machine-learning model; and testing it on real-time data by means of a desktop application. The idea is to display the solution at the rig floor, allowing the driller to take actions quickly for prevention of stuck-pipe events. Problem Analysis Packoff is the blockage of the annular space with drilled cuttings or parts of rocks fallen from wellbore walls. In a packoff situation, circulation is severely limited or impossible. With an inadequate response to packoff, fracturing of rocks below the bottomhole assembly (BHA) can occur, which leads to losses of drilling fluid and possible kick. In addition to time spent and technological risks, high expense is possible. In a worst-case scenario, the BHA can be left in hole and the bore must be redrilled, leading to a multimillion-dollar financial loss caused by nonproductive time. Packoff events fall mainly into two categories: - Caused by a cuttings bed that occurs when drilled cuttings are caught up while backreaming out of hole too promptly - Caused by wellbore collapse because of insufficient mud properties During the past 10 years, 28 wells have been drilled in the field of interest. Two hundred and twenty-seven incidents associated with stuck-pipe events have been registered. The light cases required an additional rereaming of the problematic interval, with a loss of time of up to 20 minutes. The most severe cases resulted in multimillion-dollar financial losses because of the BHA being lost in hole and the time spent on redrilling the wellbore. Classification of the incidents showed that 24 cases (11%) belong to the differential type of sticking, while 17 cases (8%) belong to the wellbore-geometry type and 186 cases (81%) to packoff events. The data collected at the field show that packoffs were encountered only during backreaming operations. Packoffs evolve rapidly. Thus, in most cases on the studied field, the time interval between trouble-free operation and sticking events is between 20 and 90 seconds (Fig. 1). Unlike physical modeling methods, machine-learning methods serve as a base for development of tools that work solely from historical data, allowing a more-precise simulation of real parameter behavior. Another advantage of machine-learning methods is the ability to select only the best historical cases for use. However, the use of historical data for prediction of downhole accidents in the well-construction process is not a novel approach. It has been used in the industry for more than 30 years. The complete paper includes an extensive discussion of examples in the literature.