Applying Big Data Analytics to Detect, Diagnose, and Prevent Impending Failures in Electric Submersible Pumps

Abstract

Abstract More than 90% of producing oil wells require some form of artificial lift for pumping production fluids to the surface (Bates, Cosad et al. 2004). The electrical submersible pump (ESP) is widely used and is currently the fastest growing form of artificial-lift pumping technology. About 15 to 20 percent of almost one million wells worldwide produce oil with the help of ESPs (Breit and Ferrier 2008). ESPs are usually considered more efficient and reliable among all oilfield lift systems and enable recovery of hydrocarbon fluids from greater depths at higher temperatures while handling a range of viscosities, gas-liquid ratios, and solids production. Over the years, the most common concern of operators using ESPs in their assets has been high workover costs and inadequately low system run life (Vandevier 2010). It is often observed that ESP performance declines gradually and reaches the point of service interruption due to a number of factors such as high gas volumes, high temperature, and corrosive environments. Should an ESP fail, the financial impact would be substantial, in terms of both lost production and replacement or intervention costs. Given the high cost of an ESP failure, operators are increasingly investing in real-time surveillance sytems to monitor ESP performance using downhole measurements and raise alarms in case of abnormal events such as trippings or failures. However, such systems are reactive in nature, i.e. action is taken after an event occurs. Consequently, there is a need and an opportunity to utilize the large amount of data being collected in real time from ESP operations, and to create solutions that advance from a reactive to a more proactive approach that would ideally detect issues well in advance, diagnose causes, and suggest corrective action. This paper offers such an approach. Specifically, a data-driven analytical solution is proposed to proactively monitor and transform vital statistics related to ESP performance into actionable information using multivariate statistical techniques for dimensionality reduction and pattern recognition. This real-time framework combines engineering principles with mathematical models to detect impending problems long before they occur, diagnose potential causes, and prescribe preventive action. This can safeguard ESP operations and increase uptime, extend ESP life expectancy, reduce intervention costs, and optimize production.