During a recent CIM Webinar, Steven Putt offered a solution for a common obstacle in mine to mill tracking; loss of information in the coarse ore stockpile prior to the grinding circuit. There is a wealth of information in the data that sits unprocessed and unconnected at mine sites. In the presentation he explained how we use Machine Learning to reconcile timing differences from dumping ore into a primary crusher, to sitting in a stockpile, and to when the ore goes through the rest of the mill.
At the end of the presentation, some interesting questions were asked about the stockpile's viability to blend optimally. We have those questions and our answers below.
What is the consistency of the delay through the stockpile?
The amount of time it takes the material to pass through the stockpile varies, and is dependent on the stockpile's size and feed rather than draw rates. This is what makes reconciliation so difficult; we need to dynamically calculate the delay time. There is enough robust data within a mill's database to run dynamic time warping, a machine learning method, to compute the delays as they change. We don't need the timing of the delay to be consistent; we need the data to be recorded consistently so we can find the patterns of the delays from stage to stage. Running the data through machine learning will learn the rhythms of the stockpile and filter out inconsistencies.
Does the stockpile behave like a "first-in, first-out" process or is it more complicated?
Great question. For the most part, yes. There are trends in how certain rocks fall down the stockpile. The more data we can analyze from your mine site, the more clarity we will have to offer your teams. Our next step is to integrate stockpile camera feeds and image processing, results from mills that have them set up, to help see the flow of the pile. Contact us if your experiencing this particular issue at your site and you want to let MinePortal find out more.
Is this a demonstration that stockpiles are poor blenders? Isn't the purpose of blending in the stockpile to smooth out peaks and reduce variability?
Yes, the idea behind a coarse ore stockpile (COS) is to reduce material hardness and grade variations in the mill. However, there is enough variation in material properties between the processing stages to allow rock types to be traced into and through the mill, even when blended in the stockpile. By analyzing this data we offer teams a higher understanding of how different ore properties and their different ratios of blends behave in the stockpile and through the mill.
Production teams can alter their plans and help the mill achieve a steady state throughput with a constant feedback loop of these insights.
What are the data points collected for stockpile determination?
A few, but not all, are assay data, powder factor, fleet management, primary crusher amps, mill feed response, sag response, and ball mill response.
How is the data reported?
We use MinePortal to funnel analysis into the cloud so you can receive answers in PowerBI, Excel, another software in your workflow, or your MinePortal login.
How does the mine site use it?
With this data you can optimally plan, blast, and blend ore for steady state throughput, increased production, and reduced recycle rates. The below graph is an example of how one mine site's planner uses MinePortal's analyzations of current behaviors and predictions for future behaviors in Excel to determine the risks and opportunities of different blends.
Common data obstacles that hold mine sites back from doing mine to mill tracking.
Below are DataCloud's solutions to those obstacles.