Failures of electric submersible pumps (ESPs) are common occurrences in the oil industry, resulting in production disruption that amounts to thousands and in some cases millions of barrels of lost or deferred oil production annually. Timely detection of ESP developing problems will enhance the capability to timely and proactively attend and fix operational ESP issues that may cause ESP shutdowns. Early warning automated surveillance systems will significantly minimize production losses associated with ESP downtime.
Traditional ESP surveillance systems provide alarms when ESP performance signals drift out of a preset range (i.e., low pressure, high current, high temperature); however, finding the root cause of the problem requires time, engineering expertise and sometimes the use of specialized tools to properly diagnose the underlying abnormal condition. After years of experience, highly trained ESP surveillance engineers are able to diagnose ESP problems by cross-referencing performance signals and finding patterns like slope increase, signal stability, unexpected step changes, etc. Pattern recognition is one of the most common applications of artificial intelligence (AI) and machine learning algorithms, which can be naturally trained to automatically detect patterns and irregularities in ESP data.
This paper presents an account of the development and testing of an ESP surveillance system based on AI with the ability to automatically detect and correctly diagnose ESP problems in the early stages of development.