Geopark Colombia Predicts ESP Failures with 92% Precision using AI

Artificial lifts are a vital part of many wells around the world. Using artificial lifts can prolong the life of wells and increase oil recovery significantly [1]. Electric submersible pumps (ESP) are the most commonly applied artificial lift systems for pumping production fluids to the surface. ESP systems are especially beneficial in wells that have a low bottomhole pressure, low bubble point, high water cut, low gas/oil ratio, or low API gravity fluids.

Over the last decade, the technology for ESP systems has gained reputation as a cost-efficient and low-maintenance alternative to vertical turbines, positive displacement pumps, and split cases in several fluid movement surface applications. However, despite their extensive use, ESP systems can be prone to mechanical, electrical, and operational failures.

When ESP systems fail, unscheduled downtime increases which directly affects the total productive hours and revenue. Interestingly, 30 percent of pump failures are avoidable (Figure 1) – 40 percent of failures come from the pump and 22 percent occur due to motor failures. More than half of the failures can be reduced with a combination of automation and early detection [2]. The challenge is to shift from a reactive operational model to a proactive model to reduce the turnaround time required for identifying, fixing, or replacing failed pumps with minimum unscheduled downtime.

Figure 1: 30 percent of ESP pump failures are avoidable

To predict electrical failures in ESP systems, Halliburton has developed an intelligent alert model using artificial intelligence and machine learning. The proof of concept was developed for GeoPark Colombia to predict and bring down their ESP system failure rates.

GeoPark is a leading independent oil and gas company in Latin America with assets in Ecuador, Colombia, Peru, Chile, Argentina, and Brazil. It is presently the first private oil and gas producer in Chile and the third-largest oil operator in Colombia.

How Halliburton’s ESP Failure Prediction Solution Helped GeoPark Colombia

We developed the ESP failure prediction methodology for an asset by processing well and pump data and configuring the system to generate two alerts viz. the Early Alert and the Imminent Failure Alert using AI – specifically a consequence of neural networks plus non-supervised algorithms. The solution will be expanded from the current Proof of Concept to supervise more Geopark Colombia wells. The deployment will use Digital Field Solver®, a DecisionSpace® 365 cloud application.

During the Proof of Concept, the Early Alert model provided warnings with an average forecast window of about 3 months while the Imminent Failure Alert provided warning of imminent electrical failure of ESP systems with an average forecast window of about 25 days. Together, these models can help bring down reduction losses by up to 10 percent. Similar outcomes are expected from the deployed solution where more wells and data will be included.

The parameters for training the models were as follows:

• All wells belonged to the same field
• All reservoirs were under similar conditions, production regime, and had similar completion cycles

The Workflow

Figure 2: Workflow for predicting electrical failures in ESP units using AI

The workflow (Figure 2) comprised of the following stages:

1. Collection of data from GeoPark Colombia’s repository.
2. Preprocessing of data, dropping the missing and outlier values.
3. The data discovery process to help identify normal and abnormal variations within the data.
4. Training and predicting early alerts using the neural network by detecting abnormal behavior.
5. Training and predicting imminent failure using non-supervised algorithm by detecting anomalies that reflected an imminent failure.

A salient feature of this approach is that the methodology to train and predict the model was applied to every pump and it was independent of variables for each pump such as reservoir conditions, type of well, pump model, well production, and so on.

Outcome

Figure 3: Results of our SmartDigital® co-innovation service to predict electrical failure in ESP systems

With this data, Landmark developed and trained prediction models during the Proof of Concept that could:

1. Forecast early warnings of pump failures up to three months in advance.
2. Generate warnings for critical cases with an imminent failure up to as early as 25 days.

The model was able to create these alerts (Figure 3) with nearly 80 percent accuracy and 92 percent precision within the scope of the Proof of Concept. With such high fidelity, well site engineers can plan the logistics well in advance for repairs or replacements of ESP systems. Similar outcomes are expected from the solution when more wells and data will be included.

This proactive operational model eliminates the need to act at the point of failure and ensures there are preventive measures taken well in advance. This can bring down unscheduled downtime and deferred production losses by up to 10 percent.

Mr. Carlos Escobar, ALS Manager, GeoPark Columbia recalls his experience of collaborating with us, “Predicting ESP failure is really challenging. But right planning, collaboration, knowledge transfer, and excellent teamwork between Halliburton and Geopark Colombia helped make the journey to developing the solution easy and trustworthy. We will continue pushing this project in the right direction to reduce deferred production and rig intervention time.”

Summary

Our study with GeoPark Colombia suggests that ESP systems failure is predictable by shifting from a reactive to a proactive model. As part of our SmartDigital® co-innovation service, we can collaborate alongside you to show how AI and ML can help enhance your wellsite operations.

References:

1. Rigorous review of electrical submersible pump failure mechanisms and their mitigation measures: Sherif Fakher, Abdelaziz Khlaifat, M. Enamul Hossain, & Hashim Nameer
2. Operating Electric Submersible Pumps: How to achieve long lifetime in challenging reservoirs: B. Viguerie (TEP Qatar); E. Toguem, P. Lemetayer, P. Perusat, (Total S.A)