Session: 01-03 Advanced Analytics for Condition Monitoring and Asset Reliability

Paper Number: 174431

174431 - Streaming Analytics for the Digital Oilfield: Incremental Learning for Predictive Maintenance 

The digital age is now.  Deep insights reside within data that inform on challenges when integrated with theory.  Within the domain of predictive maintenance this means the goal is to help determine a more optimal time to perform maintenance. Many of us must pay to get our car inspected every year, even though there’s nothing wrong with it. While it’s good to perform these checks, it would be better if we could bring our car to the shop just before something bad happens. We save money, and our car doesn’t break down on the road. In other words, we fix it before it breaks.

While we can’t know the future for sure, just like the weather, we can predict it based on scientific, data-driven evidence.  MathWorks enables the digital oil field with an end-to-end ecosystem to determine a more optimal time to perform maintenance.  Today, we discuss a solution for both anomaly detection and remaining useful life estimation as applied to rotorary motor equipment.  Specific focus will be on: (1) managing data drift on oilfield equipment for more reliable operational insights, (2) applying streaming data to optimize predictive maintenance and minimize downtime, and (3) deploying solutions to both cloud and edge devices for seamless integration and performance.  

This enables more cost-effective maintenance compared to regularly scheduled check-ups, since check-ups mean downtime for the machine, and downtime for production.  Furthermore, time spent watching the line for anomalous behavior is reduced.  The blueprint of our solution addressing both anomaly detection and remaining useful life estimation is framed around: leveraging incremental learning to keep anomaly detection accurate in the presence of data drift, the stream processing system to enable data scientists to focus on algorithm development and data analysis, as well as integrating the analytics with open tool chains to easily transition from prototype-to-production.  

Presenting Author: Ayon Dey MathWorks

Presenting Author Biography: Dr. Ayon Kumar Dey | Senior Applications Engineer, Energy Resources, MathWorks

Ayon is a Senior Application Engineer at MathWorks based in Texas. Ayon’s experience spans over 25 years working in multiple Geosciences roles for oil, gas, and chemical companies based in the USA, Canada, South Africa, Saudi Arabia, and the UAE, as well as for research institutions in Canada and The Netherlands. Ayon holds a PhD degree in Applied Physics from Delft University of Technology (Netherlands), an MSc degree in Geology and Geophysics from University of Calgary (Canada), and a BSc in Applied Mathematics from Memorial University of Newfoundland (Canada).