Data-driven stuck pipe prediction and remedies

Abstract

Stuck pipe incidents are considered a very common challenge in the drilling phase, which can result in increasing non-productive time. Common recommended practices are used to prevent or reduce the severity of these incidents. The ability to predict these incidents based on some measured parameters has been applied in the industry by using different non-physical techniques such as Artificial Neural Networks. In this work, recursive partition analysis was used to develop classification trees. The data was collected from 385 wells drilled in Southern Iraq in different fields. A total of 1015 data points were collected and divided into three data sets: training, validation, and testing. The main objective of this work is to develop a model that consists of easily adoptable logical conditions that predict stuck pipe events and suggest an appropriate remedy to free the stuck pipe. The developed method was able to predict stuck pipe events with an accuracy of 90% using simple and limited input parameters. For the stuck pipe remedy model, the accuracy of the prediction for freeing the stuck pipe reached 84%. The proposed models for stuck pipe events and remedy predictions provide logical criteria based on simple quantities that can be easily applied in the field.