Electrical submersible pumps or ESPs are one of the most important types of artificial lift methods used globally – more than 30% of oil is produced using ESP wells. Considering the importance of ESPs in the industry and the significant savings to operational expenditure from improved performance monitoring and run life maximization there has always been interest in improving surveillance, condition monitoring and failure prediction. Due to the growing number of wells and business need for one production / well surveillance engineer to monitor performance of hundreds of wells, it is important to streamline the process and provide effective tools for efficient analysis.
Knowing well ahead of time when an asset is expected to fail leading to unplanned downtime and assets failing prematurely is extremely important and valuable. Some studies indicate that on average predictive maintenance increases productivity by 25%, reduces breakdowns by 70% and lowers maintenance costs by 25%. The study referenced in the whitepaper [1] provides a summary of condition-based maintenance and predictive maintenance. The goal of predictive maintenance is to run assets as long as possible without interruptions / unplanned downtime and maintain them before an unexpected failure occurs.
This paper uses a combination of analytical methods for ESP Surveillance, detect ESP fault conditions and ultimately improve ESP run life and reliability. The work done in this paper is categorized into Predictive Analytics using Machine Learning (ML) analytics and fault tree analysis techniques. The ML analytics workflow within the application is followed to load, process, and develop a training dataset for the in-built Machine Learning model. Once the trained model has been developed it is evaluated and tested before reviewing the results. The in-built Machine Learning model is called "Moving Mean Principal Component Analysis" and as indicated by the name is a variation of the well-known Principal Component Analysis (PCA) method that has been applied in previous work on ESP pumps and other industrial applications. The fault tree analysis technique applies rule-based logic for detecting ESP conditions e.g. Deadheading. The model processes real-time data and generates events / alerts based on the rule logic configured.
