Artificial Intelligence Can Reduce ESP Failures

Electrical-submersible-pump (ESP) technology predominates available artificial-lift options. The risk of ESP failures can be reduced greatly with the right combination of advanced technologies, such as combining artificial intelligence with a cloud-based autonomous surveillance system.

jpt-2019-05-guestedgetty-1029508136.jpg

Electrical submersible pump (ESP) technology predominates the artificial lift options available to onshore and offshore operators for maximizing production from medium-to-deep reservoirs. Although designed, engineered, and built to withstand extreme subsurface conditions—corrosive liquids, scalding temperatures, and intense pressures—ESPs can and do fail without warning, often despite having monitoring systems in place. These failures disrupt production and operator cash flows. Ultimately, the costs of replacing an ESP and its associated production losses can be enormous.

But the risk of ESP failures can be greatly reduced with the right combination of advanced technologies, such as applying artificial intelligence (AI) combined with a secure, cloud-based Internet of Things (IoT) autonomous surveillance system. This provides operators with an early-warning system of ESP performance degradation in the form of a probabilistic, predictive maintenance model. With it, they can make better-informed decisions about the root causes of performance anomalies as well as how to mitigate, remediate, or manage them until the well’s next planned shutdown. In one pilot deployment of this application across a fleet of 30 ESPs, the failure of one was accurately predicted 12 days before it occurred.

Monitoring of ESP Fleets

Today, ESPs are deployed into reservoirs with sufficient sensors and instrumentation to enable continuous alarm-limit monitoring by technicians, who also keep a close watch on above-ground controls and motor drives. A distributed control system, often a supervisory control and data acquisition (SCADA) system, transmits an ESP’s operating data, which is recorded in a historian database. The data can then be used later for diagnostics. In practice, this approach is most often reactive, not proactive. That’s because conventional tools lack the ability to predict an ESP failure.

In contrast, consider the use of a more predictive monitoring model that can employ AI-based pattern recognition. This capability not only can identify ESP behavioral anomalies but also can provide actionable intelligence about their root causes—and, importantly, the probability of an ESP failure based on the data-fueled refinement of the AI algorithms (i.e., machine learning).

With this knowledge, an operator’s technical staff can better decide a proper course of action should they get an early warning about ESP performance issues. They can also be freed from having to constantly monitor ESP systems for alarm notifications and implement corrective actions. One benefit of this is that their time can be invested in more value-adding production activities.

As an implementation of AI, machine learning (ML) is based on the use of deep learning via neural networks. To implement it as an ESP predictive monitoring model, data scientists use ESP historical data to create a training data set. The latter then “teaches” the software-coded neural network all the highly dynamic, relational behaviors and interactions between the many variables within a process, as defined by an ESP’s various operating KPIs.

Once the ML model has learned the normal behavioral relations across all of an ESP’s varied parameters, it can start comparing massive amounts of incoming ESP operating data to its baseline data set—in near real time. Its pattern recognition algorithms can then watch for any anomalous relationships that may occur. And, by analyzing those differences, it can provide probabilities of how the system might behave in the future, similar to how meteorologists use percentage probabilities to predict the weather.

In a subject use case’s proof-of-concept and subsequent pilot implementation, ML was deployed via a highly secure, IoT-based cloud platform. The goal was to develop a probabilistic, predictive maintenance model for a large fleet of ESPs operating in an onshore oil field at various depths. The ESP pumps varied from 200–500 kW in power, driven from an above-ground variable frequency drive with a medium-voltage input drawn from a local electric utility.

The Pilot Phase

In one test during the proof-of-concept phase of the autonomous surveillance system, the ML-based model discovered significant anomalies 12 days ahead of one ESP’s failure. The predicted anomalies were carefully analyzed and verified with data from the failed ESP’s historical records.

Given the outcomes of the ML-based model’s performance in numerous tests conducted in controlled conditions, it proved the hypothesis that AI/ML technology can be applied effectively in the deployment of a practical, probabilistic predictive maintenance model for ESP fleet management.

After testing the ML-based predictive maintenance solution in the pilot deployment, the project team and the ESP fleet operator concluded that the system could detect multiple kinds of anomalies, even previously unknown ones.

Although these new, unknown types of complex ESP operational anomalies were difficult to interpret as to their root causes, they could still have led to ESP performance degradation and possible failure nonetheless, if not mitigated or remediated. The ESP fleet operator also gained fresh insights into the operation of these complex machines and has a record of those types of ESP operating behaviors—and a reason to investigate, find, and document their causes and appropriate mitigations or remediations.

In turn, the pilot project again verified the use of ML as the basis for an ESP predictive maintenance model that could provide human operators not only insights to anomalous ESP behaviors but also the probabilities of failure events. At that point, the ESP fleet operator deployed it as a supplementary expert system with no additional integration with the central SCADA system.

Ready for Deployment

In the model’s current architecture and configuration, ESP operators using it must still determine the appropriate mitigation or remediation steps to take should an operating anomaly be detected. Their actions can range from ordering an immediate shutdown to avoid a potential HSE catastrophe to taking advantage of the notification period to identify an anomaly’s root cause or causes, then carefully manage the particular ESP until the next planned production shutdown.

The autonomous surveillance system offers two advantages to ESP fleet operators. First, since the operator’s staff does not have to continually monitor the ESP fleet, they can reclaim significant amounts of time to focus on more value-adding pursuits. Second, because the autonomous surveillance system can identify and prioritize anomalies in advance and with actionable intelligence (e.g., failure probabilities), they have more time to apply mitigating action to prevent critical failures.

In many cases, anomaly notifications can occur sufficiently in advance to help ESP fleet operators avoid critical failures that could impact the operation of an entire reservoir, triggering successive control actions and alarm events in the SCADA system. By avoiding such critical failures, operators can pre-empt such situations and associated production shutdowns.

Altogether, these advantages can boost the overall operational efficiency of an entire ESP fleet, while improving ESP uptime and availability, as well as the reservoir’s ultimate profitability.


Nico Jansen van Rensburg is Siemens vice president, oil and gas upstream solutions. Previously, he was Siemens vice president, process solutions, sub-Sahara Africa. Before that he held positions of increasing responsibility with Siemens. He has 17 years in oil and gas, petrochemical, and process industries, with a specialty in process-analytical systems using artificial intelligence. He earned his BIng degree in electronic engineering from the University of Potchefstroom and a BSc in information technology from North-West University.