Software / Remote Monitoring / Big Data Analytics / Predictive Maintenance / Asset Management System

Nobody likes an ore loser: Predictive analytics helps keep a refinery's milling equipment moving

In this installment of What Works, learn how predictive analytics is helping a soda ash producer bump up its grind.

Ciner Wyoming in Wyoming’s Green River basin makes soda ash (sodium carbonate), a substance used in the production of glass, chemicals, and soap and in other industrial applications. It does this by mining and refining a natural substance called Trona. Ciner uses continuous drum miners to mine the Trona, which then is calcined and dissolved to separate the desired soda ash from insoluble impurities. Insoluble impurities go into a machine called the Vertimill, which grinds them further to extract any trapped soda ash.

The challenge for Ciner is that the grade of the ore it mines varies, and when a lot of low-grade ore is processed, a lot of insoluble impurities go into the Vertimill – and the Vertimill can handle only so much stuff being fed into it at one time. If it gets overloaded and goes offline, the consequences are serious: 60% of the plant’s production rate is lost when the Vertimill is down.

Lab analysis of the ore – to determine whether it’s “good” or “bad” – can only be performed after it’s processed. There’s no real-time way to assess the ore itself to get an indication of the volume of impurities that will wind up in the Vertimill. But Ciner needed some way to predict when the Vertimill is at risk of being overloaded.

The solution? An analytics tool that could filter, so to speak, through 10 streams of process data generated as ore makes its way from the ore bin through a calciner, conveyors, and agitators. If patterns could be discerned from buildups of bad ore before they reached the Vertimill, then the Ciner team could intervene whenever these patterns were detected in order to adjust the flow of material into the Vertimill. This would help ensure not only that the Vertimill didn’t get overloaded and go offline but also that operators didn’t act overly cautiously, reducing flow into the Vertimill when it wasn’t necessary to do so and slowing production.

Ciner already was collecting process data, but making sense of all of it – picking out which indicators were significant in light of other indicators, etc. – proved impossible for technicians and engineers to do on their own.

“Analytics are a lot more difficult than I think people give them credit for – running them and looking for signals and weeding out the bad stuff,” says Jolene Baker, SMART plant lead at Ciner. As Baker and Ciner’s Scott Schemmel describe it, a chance meeting with Crick Waters, a senior vice president at pattern recognition software provider Falkonry at an OSIsoft user group event in Salt Lake City, provided the introduction to the data-crunching tool that Ciner sought.

Baker mentioned to Waters that the Ciner team had been working with data scientists and machine learning tools from third-party organizations in the hope of addressing its unplanned downtime with the Vertimill. Waters relays that in that first meeting, Baker said: “This looks like the kind of thing we could do ourselves. Is this tool designed for my process engineers so that they can solve these problems without even a third party?”

“I said yes,” Waters says. “It connects directly into your PI server; you can choose the signals that are relevant to your particular process or your assets, and your process engineers or your maintenance people really are the ones that interpret the results and make use of them.”
Notes Baker: “I’ve done analytics, (and it’s surprising) just how much more your eyes can be opened to what’s actually going on” with the help of pattern recognition software.

The pattern recognition software from Falkonry that Ciner used to augment its expanded use of its OSIsoft platform, which until 2015 been used only as a data historian, allows users to see “little things that you never would have caught before,” she adds.

In a two-month trial, the Falkonry software was used to crunch Ciner’s data streams, grouping and color-coding similar operating conditions. The time periods for “bad ore” events and times when there was an inadequate grinding media charge were identified and defined to help show what operating conditions for different pieces of equipment looked like leading up to known problems with the Vertimill. The software then was able to develop prediction models for these undesired events; the Ciner team ran numerous tests of the models to validate them.

Thanks to the models, operators now have better visibility into the ore-refining process in real time and can better identify, for example, the implications for the Vertimill of an alarm that sounds in the calciner. Insolubles from bad ore build up hours before the Vertimill is affected, and because such buildups now are identified earlier, there is time to take corrective rather than reactive action. When new events do occur at some point along the pipelines, the models update automatically. The business implication is decreased production losses and cost savings from avoided reactive work.

“It’s showing you so much more than something you just created as an analytical interface ... You cover the first-order, second-order, third-order problem just looking at the patterns, and that’s quick, and then all of a sudden you’re investigating the fourth-, fifth-, sixth-, seventh-, eighth-order problems,” says Baker. “Your view of what’s happening just gets so much larger.”

For operators, says Schemmel, it’s an experience of, “Hey, I see new patterns; that’s interesting; I never would have correlated those things.” And with decision science built into the software, there’s no steep learning curve with the software for operators, he says. “You don’t have to train that operator or that process engineer on our or any other machine learning language. You can just give them the tool,” Schemmel says.

Were operators on board with using a new analytical software tool to inform their work rather than than relying on their own equipment expertise and institutional knowledge to decide when and how to respond to red flags in the refining process? Actually, yes, says Baker.

“I think they’ve been excited about it since the beginning,” she says. “(This is) a very complicated thing to look at. We’ve tried for years – they want the insight.”

Baker adds: “They’re on board because it helps their job ... If you give them the easy way to do it, they’re going to do it. If you give them something like, fill out this piece of paper and this piece of paper and then look at this – you can’t do all that stuff. You have to think of the solution they want and then go that route.”