Multivariate Statistical Analysis of Stuck Drillpipe Situations

Abstract

Summary. Stuck drillpipe accounts for losses of millions of dollars each year in the petroleum industry. Major efforts have been made to determine the causes of stuck pipe and to outline operating practices that will minimize the occurrence of the problem. No previously published efforts, however, have attempted to predict the chances of the pipe sticking on the basis of existing conditions. An analytic technique was therefore developed that evaluates 20 common drilling parameters to determine the probability of the pipe sticking in a particular well. probability of the pipe sticking in a particular well. Introduction Historically, stuck drillpipe has been the most costly drilling problem operators face in the Gulf of Mexico. In a 3-year period problem operators face in the Gulf of Mexico. In a 3-year period between 1981 and 1984, we experienced 131 instances of stuck pipe in this area. To reduce the frequency of stuck pipe occurrences and the associated costs, a study was initiated to determine how and why drillpipe becomes stuck and to recommend changes in operating practices to minimize stuck pipe incidences. practices to minimize stuck pipe incidences. The approach to the investigation of the problem began with a review of published literature. This review did not reveal any analytic techniques designed to prevent the sticking of drillpipe. Most papers presented general guidelines for operating practices to avoid stuck pipe and included general statistics of some key drilling parameters. We noted that none of the statistical analyses combined more than two variables into an analytic technique that could be used to predict the likelihood of the pipe becoming stuck. In most cases, histograms and statistical pipe becoming stuck. In most cases, histograms and statistical counts were prepared for a particular variable only for stuck wells. These statistical counts are of little use in preventing stuck pipe unless a comparative set of statistical counts are prepared for wells that never experienced stuck pipe. prepared for wells that never experienced stuck pipe. Data Collection To study our Gulf of Mexico stuck pipe problems, we collected all information from stuck pipe occurrences that might be pertinent to the phenomenon. Because of the anticipated amount of data that would be required to analyze the 131 stuck pipe occurrences, we determined that a large-scale computer analysis would be required. Data were collected with personal computers, and available database software was used to maintain data files. This use of personal computers allowed local control of the data and database structures and thus facilitated easy manipulation and analysis of very large data sets. Collection of the data by the engineers personally involved in the study and use of personal computers to gather, to manipulate, and to present data is believed to have contributed greatly to the success of this study. Data collected for the 131 cases of sticking included 96 drilling parameters recorded on a daily basis for at least 3 days before and including the day of sticking and 77 items recorded once for each case. The parameters collected included mud properties, depths, hole-geometry information, costs, hydraulics, bit data, bottomhole assembly (BHA) data, operational remarks, etc. Most data came from daily reports and consisted of normal drilling parameters collected and measured in accordance with API parameters collected and measured in accordance with API recommended practices. While data on stuck pipe occurrences were being collected, we were classifying the cases as either differentially stuck or mechanically stuck on the basis of the observed behavior of the well and operations at the time of sticking. In addition, a third set of data was gathered from wells that had not become stuck for use as a control in the analysis. The nonstuck data were collected from randomly selected wells in the same general areas of operation as the stuck wells. These three general groups of data-mechanically stuck, differentially stuck, and nonstuck-form the basis of the analysis presented here. Although they were not the main thrust of this analysis, both histograms and crossplots of the raw data were produced and studied. The histograms and crossplots revealed relationships similar to those previously published in the literature. These techniques did not provide any new information or shed any light on differences between the three groups, but they did serve to reinforce confidence in the data. P. 237