Real-Time Detection of Stuck Pipe Utilizing Hybrid AI-Physical Prediction Models

Abstract

Abstract Effective stuck pipe prediction becomes more challenging and requires real-time advanced analysis of all available drilling data. This paper presents an innovative model to predict stuck pipe incidents. A machine-learning model based on intensive feature-engineering integrated with physical models has been developed. It automates real-time drilling data collection, analysis, and detects the patterns for the most dominating drilling parameters values to achieve the success criteria of early warning signs of stuck pipe incidents. It has been applied on two equal sets of wells either stuck or non-stuck incidents. The model triggers alarms reliably and early before the stuck pipe incidents happen and therefore corrective actions could be taken properly in advance.