When it comes to applying machine learning to improve operations, Teck is reaping big benefits thanks in part to partnerships with accelerators like Google and Pythian.
It all began in spring 2017, when the Technology and Innovation and Teck Digital Systems (TDS) groups began looking at ways to leverage the power of machine learning. Very early on, the team realized that in order to build in-house expertise, including learning how to spot and evaluate opportunities, they would need a partner with extensive machine learning experience.
And what better partner than industry giant and machine learning pioneer Google.
“Once we started working with Google, things began to move really quickly,” says Peter Cunningham, Director, Digital Operations, TDS. “Their tools are powerful and leading-edge, and their people are authoring a lot of machine learning applications.”
And while Google provided Teck with the machine learning framework on their Cloud platform—which includes a range of tools and services that provide the infrastructure to power machine learning—it became quickly apparent that the Teck team would need more training if they were to ‘dabble’ in Google’s space and fully realize the power of the Google Cloud platform.
Enter Pythian, an Ottawa-based IT consulting and managed services company. Pythian was engaged to mentor Teck’s data analytics team to help advance their expertise in this area, improve their understanding of the processes for building machine learning applications, and provide an opportunity to see first-hand how the applications work with end-users.
To bring a real-world application to the training, Pythian and TDS worked with Teck’s Reliability Engineering group to look at how machine learning could be used to identify and predict potential mechanical and system issues in haul trucks at our steelmaking coal operations using the millions of data points generated by the mobile fleet.
This project involved two key phases. The first was uploading the data—faults that had already happened—to Google Cloud. During the second phase, that data was taken by Pythian and used to create custom algorithms that can predict outcomes, e.g., when a failure of an electrical system would occur.
“Progress was slow at first,” says Peter. “The reliability of predicting failures started out at less than 50%. But over a period of three months, during which time we learned more about the data and were able to optimize parameters and include critical inputs into the algorithms, the reliability increased to about 85%.”
Based on the success of that project, Peter says they’re now working on using data from the haul trucks to understand and optimize the relationship between vehicles, haul roads and drivers. Using data that tracks information from multiple sensors, things such as elevation changes, curves on a truck route and the truck speed, the hypothesis is that safety and productivity can be improved by matching the right truck to the right haul route and to the experience of the driver.
This time around, however, instead of Pythian writing the algorithms, they’re being developed by a team of five TDS analysts who attended a four-week course on machine learning in March and April at Google’s head office in Mountain View, California.
“Our partnerships with Google and Pythian have had great results,” says Kalev Ruberg, Vice President, TDS and Chief Information Officer. “In just a year of working together, we’ve built significant internal machine learning capacity that will open up a lot of opportunities across Teck.”
The Big Benefits of Machine Learning
Machine learning algorithms for detecting electrical failures are deployed across Teck’s fleet of 930E trucks at our steelmaking coal operations. In just four months, these new algorithms have prevented more than $550,000 in maintenance costs and are expected to save more than $1.5 million each year.
The Google Machine Learning Intensive
Carlos Viejo, Advanced Mining Analytics Specialist with Teck Digital Systems (TDS), who is originally from Chile but now calls the Elk Valley of British Columbia home, was one of five TDS analysts selected to travel to Google’s headquarters in California in March and April to learn about machine learning.
“It was an excellent experience,” says Carlos. “The first part of the course involved integrating the students into Google’s culture and understanding how they operate. The second was an overview of machine learning, from basic principles to more complex applications. It also provided an opportunity to conceptualize ideas.”
(L – R): Chris Ison, Supervisor, Operational Analytics, Corey Carlson, Supervisor, Asset Health Systems, Abra Gurnett, Senior Analyst, Technical, Carlos Viejo, Specialist, Advanced Mining Analytics and Ben Danic, Senior Analyst, Mining Systems, all with the Teck Digital Systems group, at the Google campus in California.
“We had access to Google’s experts at their campus and got to experience their unique culture and how they work. We discussed potential ideas, showed them data, and received feedback. It was an amazing learning opportunity.”