Tryolabs logoTryolabs logo compact
  • Services
  • Our work
  • Company
    • About us
    • Team
    • Careers
  • Blog
  • Hire us
  • Services
  • Our work
  • About us
  • Team
  • Careers
  • Blog
  • Contact
Halliburton logo

Predictive maintenance
for oil well pumps

Halliburton is one of the world's largest providers in the energy industry. With around 60,000 employees in more than 80 countries, it helps its customers maximize value throughout the lifecycle of the reservoir. We partnered with them to implement a monitoring system for their oil pumps scattered across the United States.

Case hero imageCase hero image
Image description


An in-house solution for data collection was necessary, as well as a system able to automatically analyze machinery data and determine the condition of their many thousands of wells at the same time.

Until then, the process was time consuming and the ability to detect faults with accuracy and at scale was very limited. The company was relying on a third-party service to detect anomalies in the performance of oil well pumps, and potential failures prediction. They were gathering very little information from the wells: the data consisted of just 8 data points and was taken every 15 minutes. This didn't enable the use of machine learning.

Data was collected from oil wells pumps and then handed to engineering operators at Halliburton. They analyzed the data manually, focusing on any deviations from the standard performance measures.
Read more
Image description


The ultimate goal of this project was to provide a system that helps technicians to improve maintenance by monitoring the wells close to real-time and see how they are performing. To do this, we developed a solution that detected potentially hazardous scenarios for an oil well and warned the technicians in the control center to act upon these. This approach minimizes the cost of unscheduled maintenance and maximizes the component's lifespan.

Our team of engineers implemented a custom system with the ability to collect data from sensors of oil well pumps remotely. This not only allowed Halliburton to capture more metrics than with the previous third-party solution, but also to customize and structure this information to be used by machine learning models.

The implemented system is able to detect failures that the human eye cannot see given the nuances in deviations from the standard values. The performance results of the machinery, as well as warning signs, are displayed in a user-friendly dashboard for Petroleum engineers. In the case of warning signs, the operators get an automatic notification, which enables them to take real-time actions based on actual data.
Read more

About Halliburton

Headquarters: Houston, TX & Dubai

Founded: 1919

Market Cap: 18.78B (Dec 2019)

Image description

2000+ wells

monitored in real time


false positive alerts

Data domain

Total data control for the client

Absence of required data for a machine learning solution.

By building a custom data collection system, Tryolabs set the foundation for predictive maintenance of machinery, which was leveraged in a second stage of the project.

Image description


We implemented a solution using IoT devices attached to the already existing equipment, which allowed us to collect, analyze and later visualize data.

Stage 1

Data collection

We developed a scalable system for autonomously collecting data from oil well pumps, where Internet connectivity is often very limited. An IoT device was installed to collect this data.
The solution includes two control systems: one is placed in the oil field to gather data and the other is carried by a technician working in the field. After an evaluation phase, we chose to implement ARM devices, given their suitability for the extreme climatic conditions that are present in the desert or arctic. The two systems, when in proximity to each other, connect and initiate a protocol to transfer the data gathered. Modbus is used to collect measurements from pumps and MQTT for transmitting them between the control systems. The data is then uploaded to a SCADA system where it is stored and analyzed.
Technical details

Stage 2

Feature extraction

Using a combination of time series analysis to detect possible events, and a classifier that defines their severity, we built a rules engine system to automatically detect events.
From those events, it generates different types of alerts related to the health of the wells. The engine is able to ditch the old threshold method that caused several hundreds of false-positive per day. The devices in the fields would upload 5 minutes snapshot of the well sensors and if the algorithms detected a scenario that could potentially lead to a shutdown raised an alert for the technicians to check.
Technical details

Stage 3

Dashboard for metrics & feedback loop

Lastly, we built a dashboard that displays current metrics for the different oil well pumps and the generated alerts.
Additionally, it allows petroleum engineers to use their expert knowledge to tag healthy/unhealthy time ranges, along with other possible predefined points of interest. The labels provided by the operators are leveraged to implement an automatic feedback loop, which adjusts the classifier and thus improves the system accuracy over time.
Technical details

Get in touch

Tell us a bit about your project, we'd love to see how we can help.
You can send us an email or just fill out the form below.

Email icon

Email us!

Paper sheet icon

Contact us!


Our work


  • Phone: (+598) 2716 8997
  • US: 44 Montgomery St,
    San Francisco, CA
  • Luxembourg: 9 rue du Laboratoire
  • Uruguay: Rambla Gandhi 655/701,
© 2021. All rights reserved.