Predict to prevent: Transforming mining with machine learning

By | 3 minute read | October 1, 2019

mining machine

Over the past few decades, the mining industry has been mired in a productivity slump of sorts. On the whole, production efficiency is down and costs are up. Mining companies have naturally looked for ways to turn this around, and digitalization has been one of the chief approaches these companies have followed.

It’s understandable why. Mining companies have a lot of data at their disposal. Sensors are seemingly everywhere in their underground operations. But thus far it has been very hard for mining companies to capitalize on all their data because of the difficulty in making sense of it all.

So what’s the most important data for mining companies? The short answer: assets. Mining is one of the most asset-intensive businesses there is. At every point in the extraction chain— drilling, cutting, crushing, screening and removing ore-bearing rock—heavy equipment is critical. And it takes a beating. When equipment breaks down, requiring unscheduled maintenance, production takes a hit, costs rise and a critical measure of capital efficiency in mining—overall equipment effectiveness (OEE)—goes down.

Predict to prevent

One of the big challenges for mining companies has been keeping this equipment up and running. The elusive vision they’ve pursued: minimizing unscheduled maintenance by optimizing scheduled maintenance.

As a company that serves mining companies around the world, we at Sandvik Mining and Rock Technology have been at the front lines of this effort. We saw that our customers were clamoring for a way to improve the operational productivity of their equipment. They wanted a way to use all the data at their disposal to more accurately predict when equipment failures were most likely, which in turn would enable them to optimize their approaches to scheduled maintenance.

Our strategy was to partner with IBM to enhance OptiMine®, our information and process management solution. With IBM, we jointly developed an analytics and predictive maintenance solution known as OptiMine Analytics. Running on IBM Cloud, our solution combines the IBM Watson IoT and IBM Maximo Asset Management solutions to analyze a huge repository of data captured from sensors deployed across our customers’ operations. The vast majority of these sensors are attached to fixed equipment, such as conveyor belts, and mobile equipment, such as trucks and front-end loaders. Each of these sensors transmits data wirelessly to a centralized data “lake,” where it’s indexed and stored.

Smarter decisions in the control room

What makes OptiMine stand apart is the use of machine learning algorithms to analyze the equipment sensor data at a component level. The idea is as basic as it is powerful: if you analyze a large enough dataset on the maintenance and failure pattern for a particular component, you’ll be able to make an accurate prediction of when that component—say, part of an engine, a transmission or brakes—is likely to fail. The central insight these models produce—a prediction of each component’s lifetime—is tremendously powerful because it gives operators the critical element they need to optimize scheduled maintenance practices across their entire operations for all their equipment.

In everyday practice, this optimization is most visible in the mine’s control room, the nerve center for operational decisions. OptiMine dashboards provide control room staff with a real-time view of equipment status. Most importantly, they point out potential trouble spots so that staff can take immediate corrective action—even during the current shift—if things aren’t going in the right direction. Rather than the deal with the bottleneck caused by a broken piece of equipment, the mine operator has the ability to make a smooth transition with little to no disruption in the production flow.

A fast path to value

To the customers who use our OptiMine Analytics IoT solution, the most prominent benefit has been an ability to take a more predictive—and thereby more proactive—approach to keeping their assets up and running. By combining the speed and power of cloud-based analytics with transparency across their operation, mine operators have been able to act on their insights in a way that directly impacts their production efficiency. Indeed, companies using the solution have been able to reduce their mine production downtime by as much as 30 percent, resulting in reductions in cost-per-ton ore production of up to 50 percent.

Here’s one of the most important things about cloud-based analytics: Our customers are getting an immediate bang for the buck and they’re showing their high rate of license renewal. There’s no better measure of success than a customer coming back because they’re getting value.

Listen to Patrick Murphy discuss how Sandvik is using machine learning to transform mining asset management: