Data Driven Approaches to Enhance Choke Valve Reliability

Abstract A choke valve is a crucial component in oil and gas production, serving as the primary equipment responsible for handling substantial pressure reduction from the wellbore to surface facilities. This research evaluates the reliability of choke valves through statistical analysis, machine learning techniques, and case studies from the Gulf of Thailand. Numerous failure records have been examined, focusing on operational limitations and recommended design enhancements from manufacturers. Numerous approaches for monitoring choke valves have been comprehensively analyzed using statistical and machine learning methodologies. The methodologies include monitoring valve coefficients, comparing velocity limits with hypothetical tests, performing causality analysis to determine significant variables from both surface and subsurface data, and detecting anomalies to provide warnings for the maintenance team to analyze conditions. The objective of the project is to incorporate data insights into operational and maintenance workflows with the goal of improving the reliability of valve functions. Consultation with manufacturers revealed the absence of a definitive industry standard for choke valve velocity restrictions, in contrast to flashing services in control valve application as stated in the International Society of Automation (ISA) guideline where a suggested throttling velocity is 23 m/s. Although a cage velocity limit of 45 m/s was proposed for multiphase flow by the manufacturer, validation results with field data indicated that there was no significant difference in failure rates for valves operating at speeds below or above 25 m/s. By varying the velocity thresholds, it was found that a limit of 15 m/s produced the most statistically significant outcomes using the student t-test, which in turn served as a warning signal for future operations. Nevertheless, the results of causality tests revealed that subsurface data, including true vertical depth, porosity, and other subsurface data had a greater predictive power for valve damage compared to velocity limitations alone. To enhance the monitoring system and provide maintenance teams with actionable information, a machine learning pipeline was created to generate anomaly warnings based on these factors. This study introduces new data-driven approaches for analyzing choke valve performance, challenges existing industry standards, and offers practical solutions for improving reliability. Lessons learned from the Gulf of Thailand cases provide valuable insights for industry professionals and valve manufacturers.