SEPTEMBER 20, 2021 Author: Abhishek Bihani Category: Thought Leadership
Like any manufacturer, upstream oil and gas producers continually seek the next great efficiency improvement. They understand that even fractional gains in core equipment efficiency can yield substantial benefits to production, revenue, and equipment life. For oil producers, sucker rod pumps (SRPs), also known as beam pumps, are under constant scrutiny for this reason.
SRP systems are the most widely used method of artificial lift in onshore oil production. Designed for use on wells that cannot produce the well fluid on their own, which is by far the majority, the systems consist of many surface and subsurface components that must be periodically maintained to optimize operational efficiency.
Sub-optimal production in a single well can cost a producer thousands of dollars per day in lost revenue. Additionally, damaged rod pumping systems require extremely expensive repair processes called workovers, making predictive maintenance essential. A new approach was needed to accelerate efficiency improvements.
Tignis developed a machine learning (ML) model using PAICe Builder to demonstrate how easy it is to apply advanced analytics to automatically detect sudden efficiency losses and emerging equipment issues. The SRP analytics built into Tignis’ PAICe Builder monitor the rod pump cycle, identify anomalies, trigger alerts, and quantify the production losses and impact on revenue on a pump-by-pump basis. The potential impact on the bottom line is considerable.
Performance and diagnostic challenges
Any number of pump or rod issues can lead to failure or a drop in SRP process efficiency, such as pump valve leaks, bent rods, wear, corrosion, insufficient liquid supply, fluid buildup, gas interference, and sand production. These failure modes can present themselves in different ways for each pump or failure, making it difficult to write legacy condition-based monitoring alerts.
A dynamometer is one commonly used device to monitor SRP operation. These devices plot the SRP’s rod load versus position through every cycle, which can be compared to an ideal dynamograph to monitor if a pump is behaving normally. Although this method is considered to be very adept at catching SRP issues, the solution can be difficult to implement and maintain. Each SRP’s ideal dynamograph must be tuned individually. Additionally, dynamographs must be re-tuned as the reservoir and well properties, such as crude viscosity or fluid ratios, change over time, resulting in significant maintenance costs.
One proposed methodology of monitoring SRPs is utilizing pure physics-based models of the pump and well to produce the ideal process state and comparing the ideal to real-time sensor data from the well. However, in practice, this solution is not practical. Physics-based models of sucker rod pumps are computationally taxing and cannot be run in real time on currently available hardware. Calculations of a single cycle of the pump using readily available physics-based models can take more than a minute to compute, meaning that the calculation cannot keep up with the incoming real-time data.
Real-time analytics alternative
Tignis enables an automated approach to SRP monitoring by providing a tool that allows process/operations engineers to quickly build real-time ML analytics on physical systems and bring data-driven decision support to challenges across the operation. To provide an example, a new rod pump analytics model was built into PAICe Builder, powered by our proprietary Digital Twin Query Language (DTQL), which helps automate detection and notification of SRP inefficiencies by converting the SRP surface load and position into time-series data and running ML-based predictive models against it in real time.
A key advantage of this method is that ideal operation of the SRP is based on the pump’s historic behavior, meaning the analytic does not require manual setup and tuning for each pump it is applied to. This analysis can be broadly applied to many SRP manufacturers and applications. In addition, the model continues to tune itself over time using new incoming data. This method allows the quick and efficient detection of changes in well efficiency or operation.
Another advantage of using PAICe Builder for SRP analytics is that the embedded machine learning drastically increases calculation times of ideal variables. When trained on historic data, machine learning can predict ideal process states, similar to a physics model, but much more efficiently. This enables monitoring to happen in real time as data is being produced.
Not only does PAICe Builder uncover efficiency anomalies and immediately alert the operations crew, but it also allows engineers to translate the efficiency loss to the loss in barrels per day (bbl/day) produced, and then translates that oil loss to the loss in potential revenue per day. All this supporting data is included with the alert to help prioritize actions on SRP abnormalities.
Quantifying substantial business value
The SRP analytics model targets conditions that cause insufficient pump rates, such as friction between the pump and other subsurface components, equipment wear, and corrosion. The figure above illustrates how dynamograph data reflects the abrupt change in position and load behavior when a pump’s rate, or strokes per minute (SPM), drops after six minutes of normal operation. Using a simple rule set created in the PAICe Builder app, deviations like this are handled in three steps:
1. Available sensor data, in this case the annular fluid height in the well, flow rate out of the well, and surface pump position, is monitored to detect in real time when a significant change in surface load behavior occurs and its relationship to fluid production behavior, at which point an alert is issued with this information and the following supplemental data points.
2. The analytics determine the ideal flow rate (846 bbl/day) had the anomaly not occurred, as compared to the actual flow rate (506 bbl/day) as a result of the anomaly, which translates to a production loss of 340 barrels of oil per day.
3. The resultant daily revenue loss is computed by multiplying the production loss amount by oil price data that is either manually input or streamed into PAICe Builder. In this case, at $60 a barrel, the theoretical 40% loss in pump efficiency equates to $20,400 of lost revenue per day.
Armed with this knowledge, corrective actions can be prioritized based on their value to the operation. Simple connectivity to data visualization tools such as OSIsoft’s PI Vision allows operators to rank order the rod pumps by those causing the most immediate or substantial revenue loss, allowing them to concentrate their efforts to maximize production and profitability.
From this example, it is easy to see how early detection of efficiency anomalies can save thousands of dollars per day. Additionally, the SRP analytics reveal issues that don’t immediately impact oil production but can shorten pump life or cause expensive repairs, allowing predictive maintenance to occur before failure. One such example is a pump-off condition where the pump fills with insufficient fluid during upstroke. This condition can lead to fluid pound, causing accelerated stress and fatigue of subsurface SRP equipment, ultimately leading to a premature need to perform an expensive workover of the pump and other subsurface equipment.
The simplicity of the solution is also evident in how the analytics are applicable to a broad range of rod pumps. The custom ML model can be deployed to any rod pump that has sensor data, regardless of its manufacturer, rating, or geographic location.
The SRP model is just one example of how, using PAICe Builder, customers can quickly build a wide range of analytics for themselves to address whatever process challenges they are facing. ML analytics use cases like this one make it clear why artificial intelligence (AI) is the future of process control