Enhanced decision-making and intervention enabled by Artificial Intelligence Response Prioritization

Part II: Case Studies of Event-informed Responses

Avalon Surveillance AiRP (AI Response Prioritization) is a state-of-the-art real-time surveillance system for ESPs (Electrical Submersible Pumps). Previously, Part I of this series discussed how the AiRP engine proactively detects low/no flow conditions where no timely action was taken.
 
In part II, we discover how the AiRP engine identifies the type of event and quantifies its severity to facilitate appropriate actions. With six case studies, this article showcases event detection situations that help facilitate actions for event recovery, event mitigation, and operational strategy. Although not covered in this article, it is helpful to note that at a higher level, a customizable triage engine automatically ranks inter-well scores to bring attention to assets needing more immediate attention; these scores are a function of events, alarms, and other attributes – for more information on this aspect of the solution, please contact Sensia.

The AiRP is an enabler for autonomous event recovery, condition monitoring, and failure risk assessment:

• On edge devices like Sensia’s Hyperconverged Controller (HCC2), AiRP can be paired with the AiER (AI Events Recovery) application, using reinforcement learning-based AI agents to automate event recovery in real-time. More information about that concept can be found in this paper.   
• AiRP generates valuable and detailed information about a pump's operating history, such as event frequency, event severities, energy dissipation, and much more. This is a pre-requisite for the development of advanced condition monitoring and failure risk assessment solutions for ESPs.
• As an ecosystem of applications, they contribute towards Sensia’s vision for robust, autonomous, self-driving, self-optimizing, and meaningfully scalable systems.

Case studies:

This list of case studies demonstrates how manual actions were informed by AiRP events. As with Part I, it is important to re-emphasize that, in general, the engine is looking in real-time for critical rate and magnitude of changes in flow conditions and identifying potential causes – these causes (different low flow event types) reflect a range of intensity that can be observed through each event’s severity parameter (0.5 -1, where 1 is highest). For example, when the engine first raises a Pump Off event, it should be interpreted as follows: there is sufficient evidence that there is more drawdown or a restriction at the intake of the pump that could potentially escalate in the worst case to liquid levels dropping to the intake or drawing more from the formation than planned. Alternatively, the situation could recover naturally or through intervention.

Case 2.1: Insufficient Lift (IL) event resolved by increasing speed   Description: After the pump started, the motor temperature continued to rise above its expected reference value by 6 ºF at Point (1). At the same time, the intake pressure was trending upward. Finally, the motor current was 10% lower than its expected reference value. All three signatures were sufficient evidence for the engine to raise an IL event (A).   Action: Guided by AiRP, surveillance engineers started increasing the speed of the pump at Point (2) to more quickly establish the desired flow and eventually recover from the event when signals sufficiently approached their references. Event severity provides useful feedback to evaluate the impact of actions and can be observed declining as the pump was momentarily sped up. After removing the event, the pump was safely returned to the original desired/reference frequency at Point (3).

Case 2.2:  Insufficient Lift (IL) event prompting remedial actions and further adjustments to the operating point   Description:  A rapid increase in motor temperature by about 13 ºF prompted AiRP to quickly raise an IL event (A). Overheating of the motor is an indication of low-flow conditions and can damage the pump.    Action:  The surveillance team promptly increased the pump speed at Point (1) resulting in increased flow and reduction in motor temperature at Point (2) and consequently removal of the event; absent of this intervention, the situation could have easily escalated leading to a trip and shutdown. At Point (3), the pump speed was decreased to the original/reference level.    Prompted by Event (A) bringing attention to the well, the surveillance team decided on a new strategy to increase production at a lower intake pressure target. Their first attempt at increasing the pump speed was at Point (4). However, this led to a steep increase in pump discharge pressure at Point (4) above the reference signal, and hence to avoid an event, the speed was quickly reduced. A second attempt with a lower speed step happened at Point (5) but pump discharge pressure again started rising above the reference signal at Point (5). Observing a general increase in discharge pressure, the team decided to adjust (open) the surface choke at Point (6), leading to increased flow towards the new intake pressure target of about 175 PSI. The speed was adjusted to the original reference value as the choke adjustment was deemed more effective for the transition.    The reference signals were useful to help navigate the transition to this new operating condition without triggering events. Even if events were raised, an attentive team could react to resolve them quickly.

Case 2.3:  Insufficient Lift (IL) after startup requiring ramping up in speed   Description:  The pump was started at a lower speed at Point (1) compared to the previous reference level, and instead of continuing to decline, intake pressure began rising at Point (1). Consequently, the AiRP engine raised an IL event (A), indicating a lack of lift compared to the last normal operating condition before the pump was shut down.   Action:  In response to the event, the surveillance team could initiate a ramp-up to target speed at Point (2), which promptly resolved IL event (A). During the ramp-up, there was a brief Dead Head DH-2 (borderline 0.5 severity) event (B), which was due to a slug causing discharge, intake, and tubing head pressures to rise momentarily. Operations made further adjustments to the surface choke as pressure started building up and more flow was established, avoiding further events. Overall, the learnings from this case are used to improve future startups by starting at a higher initial frequency.

Case 2.4:  Multiple Insufficient Lift (IL) and Pump Off (PO) events after startup caused by initial speed feedback control   Description:  During restart, the pump was under intake pressure feedback control mode between a frequency range of 54 and 62 Hz. At Point (1), there were rapid increases in discharge pressure (>200 psi above its reference value) and tubing head pressure, prompting AiRP to raise IL event (A). Throughout the highlighted region, the pump went through a series of IL and PO events (B, C, and D) primarily triggered by 20 ºF swings in motor temperature and 20% fluctuations in motor current. Intake and discharge pressures also had noticeable fluctuations.    Action:  Based on these events, the surveillance team realized the pump was under unnecessary thermal and mechanical stress. Therefore, the feedback control mode was disabled and once the target frequency was fixed at Point (2), pump operations soon stabilized with signals approaching their expected reference values, and no further events were raised by the engine.

Case 2.5:  Multiple Insufficient Lift (IL) events caused by initial feedback control setpoints for pump speed   Description:  The pump was originally operated at a constant speed of 53 Hz. Operations were stable, and AiRP did not detect any issues. In the highlighted region, the pump was put on feedback control mode based on a new (lower) intake pressure setpoint. This was done to increase production close to the limit. However, the trade-off led to unstable operations where AiRP detected multiple IL events. These events are representative of thermal and mechanical stress that could lead to cumulative damage and impact run life.   Action/Impact:  This case study demonstrates how AiRP can serve as feedback for the impact of various operating or control strategies. Capturing these events automatically will be important for further analytics to enable more informed trade-offs between production vs run life.

Case 2.6:  Appropriate feedback control strategy helps prevent Gas Interference (GI) events   Description:  The pump was originally operated in a feedback control mode. As shown in the left half of the figure, pump operation remained relatively stable, with only a few GI events detected by the AiRP engine. In the highlighted region, the pump was switched to operate at a constant speed. This led to unstable operations, as indicated by significant fluctuations in all signals and corresponding GI events. Once the pump was put back on the feedback control loop at the right side of the figure, the frequency of events was significantly reduced.   Action/Impact:  This case study again demonstrates how AiRP can serve as feedback for the impact of various operating or control strategies. Capturing these events automatically will be important for further analytics, to enable more informed trade-offs between production vs run life.

We hope you found Part II helpful in understanding how AiRP can support your decision-making capabilities. For more information on our AiRP products, click here. And remember, this is the second of a three-part series. Look out for our next edition in the coming weeks.
 
Learn more about our authors:
 
Harshkumar Patel
As an R&D Data Scientist at Sensia, Harshkumar’s interest is in developing artificial intelligence and physics-based digital and automation solutions. He has over 10 years of experience conducting R&D across diverse domains within the oil & gas, geothermal, and offshore industries. Harsh (as he is known) holds a PhD in Petroleum Engineering from the University of Oklahoma, USA. He has published over 30 research papers and has received several scholarly awards, including the 'US Einstein Visa' for extraordinary research impact and international recognition in his field.
 
Jonathan Chong
Advanced Technology R&D Manager at Sensia, Jonathan is an expert in scalable operationalization of artificial intelligence and physical modeling techniques in digital and automation systems. He achieves increasingly autonomous operations that are robust and able to adapt to complex, evolving environments. Jonathan holds a PhD from the National University of Singapore (NUS) in the Innovation in Manufacturing Systems and Technology program under the Singapore-Massachusetts Institute of Technology (MIT) Alliance. He has 18 years of combined experience with SLB and Sensia in various roles within technology organizations, from conducting R&D and engineering to managing large technology programs.
 
Tiago Meira de Borba
With over 15 years of industry expertise, Tiago started a dynamic career primarily with SLB, navigating diverse oil environments ranging from offshore and deep-water subsea to onshore and unconventional settings. By 2023, he started a new role as Global ESP Domain Champion at Sensia. Here, Tiago is responsible for overseeing and optimizing artificial lift systems. He possesses in-depth knowledge of technologies like ESP (Electric Submersible Pump), Rod Pumps, Gas Lift, and other artificial lift methods. His primary role involves designing, managing, and supporting digital products and addressing technical challenges to achieve production needs.

Interested in reading more in the series? Click here to read:

Article 1 - How Artificial Intelligence Response Prioritization facilitates action-focused surveillance at scale

Article 3 - Managing data quality with Artificial Intelligence Response Prioritization