Teck

The Solution: Improving Asset Availability using Predictive Analytics

ClearObject delivered 10 machine learning models in three months to help Teck Resources predict haul truck engine failures before they occur, ranging from specific engine component failure modes to general anomaly detection and engine performance monitoring.

Speed to value was vitally important for Teck, so ClearObject used a phased approach to model development. ClearObject also continues to deliver model improvements and ongoing model sustainment services. This has allowed the models to use in-service feedback from technicians to tune the model and improve performance by reducing false notifications and delivering more accurate prediction notifications.

“Teck has years of experience capturing near real-time analytics from assets in operation. ClearObject brought their machine learning capability, a product-centric approach to digital solution development, and subject matter expertise in rare event anomaly detection to allow us to better understand asset health. ClearObject is a partner concerned with the long-term success of our digital transformation initiatives.”–Alex Creagh, Manager, Maintenance – Business Improvement

The Customer

Teck Resources Limited is committed to responsible mining and mineral development around the world with business units focused on coal, copper, zinc and energy as well as steelmaking. Teck is headquartered in Vancouver, British Columbia, Canada, and currently owns or has interests in 11 operating mines, an industry-leading metallurgical complex, and development projects throughout the Americas. Teck has a well-define digital transformation initiative known as RACE21 expected to return over $1 billion in EBITDA improvements by 2021.

The Challenge: Improving mining equipment uptime and availability

Keeping their fleet of haul trucks and heavy equipment running is vital to Teck’s mining and mineral development operations. To ensure equipment uptime and minimize operational disruptions, the company’s Maintenance and Operations team needed to be able to effectively predict engine failure, before it occurred. If they could do this, they could eliminate costly, unplanned downtime.

Anytime a haul truck experienced a critical engine failure, this removed the asset from operation for a minimum of four days. If these failures could be eliminated, Teck could achieve significantly improved truck availability while realizing millions in cost savings by reducing maintenance cost per production hour in mining operations.

The Result

Teck’s haul trucks are no send performance and general engine and truck health data into an alarm management system. Teck personnel and SMEs can use this historical data to determine which features of the engine are most indicative of potential failures through statistical analysis.

Of the 10 predictive models, ClearObject has developed a model of engine fuel pump failure, general anomaly detection models, as well as engine performance monitoring. Enhancements to each model are being made during an ongoing project phase for Model Improvement & Sustainability.

Going Forward

Teck and ClearObject will continue developing the maintenance intelligence solution as a part of the company’s overall RACE21 plan to save more than $1B by the end of 2021. Officials at Teck are set to accomplish their objective through operational efficiency, driven largely by the Teck maintenance intelligence tool and 10 predictive models to reduce the number of catastrophic events to their mining haul trucks.

Frequently asked questions

Why does Teck need to predict engine failures?

A critical engine failure takes a haul truck out of operation for a minimum of four days. Eliminating these failures improves truck availability and reduces maintenance cost per production hour.

What do the models detect?

The 10 models include an engine fuel pump failure model, general anomaly detection models and engine performance monitoring.

How do the models improve over time?

ClearObject provides ongoing model improvement and sustainment, using in-service feedback from technicians to reduce false notifications and make predictions more accurate.