A surface-based solution for ESP monitoring

Motor current signature analysis and Machine Learning algorithms find “fingerprints of failure”

Electrical submersible pumps (ESP) play a crucial role in oil and gas operations. When a reservoir has insufficient energy to produce oil at economic rates an artificial lift method is required to increase fluid flow. Electric submersible pumping is one of the most versatile and adaptable options for moderate to high fluid volumes.

Unfortunately, the conditions encountered in some wells can be very hostile and this often has an adverse effect on both a pump’s reliability and any monitoring sensors associated with it. ESP failures may be caused by the presence of solids (fine rock particles) from the reservoir, abrupt changes in well conditions, the presence of free gas in the pump, corrosion or elevated operating temperatures.

When an ESP fails it can have a catastrophic effect on operations, incurring high costs associated with lost production and repair or replacement. It is critical, therefore, to mitigate these risks. A new surface-based monitoring method uses advanced algorithms to analyse current and voltage data and provide early diagnosis of ESP problems.

Monitoring and maintenance strategies

Most of the maintenance regimes currently used by the oil and gas industry depend on time-based strategies or traditional monitoring methods for condition-based maintenance.

Time-based maintenance introduces additional costs (unnecessary shutdowns when checks are made too early) and accepts unplanned downtime as a feature (component failure when checks are made too late).

In contrast, the aim of condition-based maintenance is to perform repair work before failure occurs – typically when a drop in ESP performance is recorded – but not during normal operation. Condition-based maintenance (CBM) is the optimal maintenance strategy for ESPs because it mandates that maintenance is performed before breakdowns occur or when performance decreases, but not before. CBM requires accurate, reliable and cost-effective condition monitoring tools. This is where traditional tools fall short: Because they have to be installed on the motor and pump, installation of sensor-hardware is expensive as it requires sensors to be installed on the motor and pump itself, including wiring up the system with power and data-cables. What’s more, because these sensors and cabling are subjected to hash conditions increasing the risks of failure.

In traditional condition-based maintenance systems vibration or temperature sensors are installed subsurface, on or near the pump. These measure parameters such as motor temperature, pump discharge temperature, pump intake pressure, pump discharge pressure and motor vibration. However, installing sensors on a submerged pump that operates in harsh conditions below the surface of the earth is often challenging and expensive. It can be difficult to ensure the physical integrity of the components such as the sensors and the cables that transmit data from the ESP to a surface station.

Furthermore, time-based and traditional condition-based maintenance both fail to leverage the latest technologies in sensing systems and artificial intelligence. This places the burden of interpretation on technical staff who will doubtless have many other demands on their time.

Fortunately, there is an alternative. Online condition monitoring tools based on a combination of motor current signature analysis (MCSA) and machine-learning algorithms offer an efficient and cost-effective solution that addresses the unique challenges facing oil and gas operators.

Motor current signature analysis

The MCSA concept originated in the early 1970s when it was proposed as a tool for monitoring motors in hazardous areas or harsh environments within nuclear power plants. It is a condition monitoring technique that can diagnose problems in induction motors by analyzing current and voltage data [Ref 1]. MCSA sensors are installed inside the motor control cabinet (MCC) and data is collected online without interrupting production and with motors running under load at normal operating conditions.

Sensors installed inside a Motor Control Cabinet

For engineers the recognition of motor current fault signatures would require a considerable degree of expertise and experience, but the modern MCSA tools take care of that. The new system delivers an automated interpretation using powerful artificial intelligence algorithms that detect and diagnose imminent failure in AC induction motors and pumps.

A surface solution for downhole systems

In contrast to traditional ESP condition-monitoring tools, which place sensors downhole, all of the hardware for an MCSA system is installed on surface in the motor control cabinet (MCC). This is a much less demanding environment for precision instruments.

MCSA sensors can be installed in less than an hour per motor and will collect data under all operating conditions. This ensures the continuous streams of high quality data that automated condition monitoring tools rely on. In contrast, replacing downhole sensors that have failed may be technically impossible or prohibitively expensive (Table 1).

Basic principles

ESPs are powered by AC induction motors. Electrical power moves a rotor inside a stator to turn electrical energy into mechanical energy. There is an air gap between these components and disturbances of this gap show up as tiny ripples on the sine-wave of the AC power consumption.

Mechanical problems in the motor or the pump cause additional vibration relative to equipment that is operating normally. This vibration causes ripples on the current sine-wave that are measurably different from those recorded when the ESP is operating in a healthy state. Current sensors measuring at high frequencies and high resolutions, can pick up this analogue signal and turn it into a digital data stream for further analysis.

Algorithms convert the data into a pattern of behaviour that defines the range of sine wave shapes and ripples that occur under normal, healthy circumstances. Anomaly detection algorithms can then track changes over time and identify “unhealthy” deviations that cannot be explained by operational factors such as changes in load and power.

Anomaly detection

When anomaly detection algorithms flag behaviour outside the normal “healthy” pattern for the ESP it indicates behaviour that warrants inspection of the motor and pump.

The commercial value of the MCSA approach becomes apparent when considering a site where the engineering team has to monitor dozens or even hundreds of pumps. In a traditional scenario, this would require visits to each pump to conduct manual inspections. This is a labour-intensive process that makes inefficient use of scarce technical staff resources and may carry additional safety risks. The MCSA-based system enables operators to avoid these issues.

Anomaly detection algorithms can indicate which pumps are behaving normally and which have shown the highest deviations from the expected patterns. This helps the maintenance team prioritise the assets that are most likely to require urgent attention. But that’s not all; the MCSA-based system can also indicate the likely cause of a problem.

Example patterns of a Pump in normal operating conditions vs operating with a damaged shaft

Fault classification

MCSA classification algorithms recognize patterns associated with specific failure mechanisms. For example, pump cavitation shows a distinctively different pattern from bearings damage, or stator failures of motors for that matter. The different failure mechanisms leave distinct marks on the AC sine and, as such, are like “fingerprints of failure”. And just as in the courthouse, a partial match can be enough to identify the culprit and connect the pattern to an electrical or mechanical root-cause.

The major faults of electrical machines that can be identified by MCSA include:

  • Air-gap eccentricity: a non-uniform air gap between the rotor and the stator
  • Broken rotor bars that can cause sparking and overheating
  • Bearings damage
  • Cavitation
  • Impeller damage
  • Shorted turns in stator windings
  • Load effects
  • Equipment wear effects

The latest developments in artificial intelligence can leverage what is called ‘transfer learning’ to abstract the meaningful parts of a failure fingerprint that hold up regardless of asset brand, type or even class. This means that the fingerprint of stator failure in a large motor can be used to detect the same issue in a smaller system. In some cases it is even possible to use a failure mechanism pattern from one equipment type to identify the same issue in a different equipment category.

Conclusion

An MCSA-based condition monitoring tool for ESPs can reduce unplanned downtime, lower maintenance costs and help operators minimize safety and environmental risks. The latest generation of MCSA-based condition monitoring tools offer unique monitoring capabilities for ESPs because the sensor system is installed inside the MCC, rather than on the asset.

The emergence of powerful artificial intelligence algorithms, improved data transfer protocols and cheap, powerful computing have made MCSA-based systems an attractive option. MCSA-based systems have the potential to greatly improve data accuracy and reliability whilst lowering the costs of monitoring at scale. The elimination of unplanned downtime is now within reach.

Author:

Simon Jagers

Founder Semiotic Labs

simon@semioticlabs.com

www.semioticlabs.com

About Semiotic Labs

Semiotic Labs offers SAM4, a Smart Condition Monitoring tool for AC induction motors and rotating equipment – including pumps- that detects upcoming mechanical and electrical failures at an early stage. SAM4 is a plug & play tool that includes hardware, analytics and an online dashboard that provides actionable information about the health, performance and energy consumption of assets.

Semiotic Labs is based in Leiden, The Netherlands.

References

1 Brief review of motor current signature analysis, D. Miljković. (2015) https://www.researchgate.net/publication 304094187_Brief_Review_of_Motor_Current_Signature_Analysis

Accessed 26.07.2018

Like what you read? Give Semiotic Labs a round of applause.

From a quick cheer to a standing ovation, clap to show how much you enjoyed this story.