Expert system for early sign stuck pipe detection: Feature engineering and fuzzy logic approach

Stuck pipe represents one of the main anomalies of oil well drilling operations, responsible for economic, operational, and safety issues. However, it is possible to infer early signs of stuck pipe from the process variables, which improves the performance of the operation teams and avoids its occurrence. In this scenario, machine learning techniques have been used to prevent stuck pipe. However, such techniques strictly depend on the characteristics of the available data set, which, for the problem at hand, are typically unbalanced. On the other hand, expert systems are an alternative to machine learning algorithms because they capture the tacit knowledge of operators. This work presents a system for preventing stuck pipe based on fuzzy logic applied to special features derived from the process variables. In addition, a deep analysis of the behavior of the process variables in real time is presented. The features of the process variables allow generalization, identifying early signs of stuck pipe in different wells. The application of the proposed early sign fuzzy detection system to fourteen cases of stuck pipe in four different offshore oil wells on the Brazilian coast demonstrated its effectiveness in predicting stuck pipe's occurrence. The performance of the proposed system was measured using precision/recall score and the advance time for the stuck pipe to occur, showing that more than 92% of the stuck cases could have been avoided. Finally, a comparison with four well-known classifiers highlighted the best overall performance of the proposed approach.

Automated summary

This paper presents a fuzzy logic-based system for preventing stuck pipe, which can identify early signs of stuck pipe and avoid its occurrence. By analyzing special features and applying machine learning techniques, the system shows high effectiveness in predicting stuck pipe and can avoid more than 92% of stuck cases.