Real-Time Drillstring Vibration Characterization Using Machine Learning

Abstract

Abstract Downhole shocks and vibrations have been identified by many operators as one of the biggest causes of Non-Productive Time, the most significant factors limiting rate of penetration (ROP) and the leading cause of premature failure of downhole tools. Today most of the existing methods for detection and characterization of downhole dynamics rely on costly downhole sensors integrated in bottom hole assembly (BHA). This paper presents a new technique for detecting and characterizing drillstring shock and vibrations in real-time using solely surface measurements and a machine learning method. Using historical offset well data and simulated well data, this new technique provides a method to build a classifying model that can be used during drilling operations to characterize real-time drilling data. Validation of the new technique on recorded data demonstrates the method’s capability to detect and characterize downhole dynamics such as stick-slip and lateral shocks from surface measurements.