IMAGINE being able to have your best production day every day.
This is now a reality thanks to developments in big data acquisition and modelling, machine learning, prediction and simulation software and a dramatic increase in the accuracy delivered by all these tools.
Welcome to the new era of digital twinning an entire mine.
Typically, mine optimisation was conducted in silos – drilling, blasting, crushing, recovery, milling etc. – looking at individual decision choices and set points within a plant to achieve the best results in that specific centre.
However, by utilising historical and current data from the entire mine value chain, modelling the whole process and then relating it back to the individual geology blocks, it is possible to create a unique, new paradigm for prediction, simulation and optimisation for all mining processes.
Petra has spent many years developing cutting edge digital twin models.
The end goal was to optimise processes such as extraction, comminution and recovery for different geology and this is what their unique MAXTA software delivers – and then some. Changing geology is responsible for much of the uncertainty in traditional modelling, so Petra (which is Greek for rock) made it their mission to re-write the manual and give engineers and geologists alike, greater detail and accuracy in their predictions and simulations.
How to optimise for geology
The challenge has always been how to optimise for the different geology that comes through each mining process. Digital twin models are more widely known in a manufacturing context where inputs can be measured and the behaviour of outputs can be defined by physics.
In mining however, inputs are primarily rock, whose behaviour is difficult to measure at any scale.
Indices exist to measure certain characteristics such as blasting and geotechnics, while processing is based upon laboratory scale mineralogical analyses, and pilot scale tests.
But the advent of ‘big data’, cloud computing and machine learning now provides the possibility to learn from historical performance and no longer rely on static models.
It is now possible to use a mine’s actual performance in conjunction with machine learning (actually learning from the data) to determine what that particular model should be.
Historically, explicit models were used, where data is input into a known function and then outputs are derived.
However, outputs from functions are typically prone to errors in the region of 10-40pc because of the uncertainty associated with how rocks behave in different geological conditions.
With machine learning, there are large scale inputs – essentially all relevant data from the whole mining history.
This especially includes geological data such as the block model, drilling, weather and hydrology – all overlaid.
Then, the downstream process performance data is used (outputs from plant, crushers, mills and other value drivers).
Machine learning then takes the inputs, compares them to the various performance outputs and works out the actual function. It automates the process of defining a function that explains the relationships between inputs and performance outputs in the mine.
This can be done at each point along the value chain.
This is the power of Petra’s MAXTA software platform.
The hard part is the ability to mine this data and work out the attributes of the ore as it arrives at each point of the value chain.
Petra commenced four-and-a-half years of ore tracking research and development with a case study from Telfer mine at IMARC in 2015.
Since then the company has worked with iron ore (haematite and magnetite deposits), copper-gold porphyry, epithermal gold silver and others to map data such as SAG mill throughput and recovery back to the pit – also applying all available big data from the mine to the mill model.
This process has now evolved, and is far more complex, incorporating blended stockpiles, crusher data and conveyor systems.
The end result is the ability to know what ore is arriving at any particular point in the chain.
When drilling, weather and geological data is added, it forms a solid basis for building a whole-of-mine model.
Prediction
Digital twin models feedback historical and current data performance data to predict given outputs from geological block model and other data.
Plant performance and throughput can be put back into the geological model via the block model and be used to derive more accurate scheduling and mine optimisation.
This is the realm of MAXTAGeomet package.
Predictive modelling can be used, for example, to see how geology will play forward with a particular schedule of blocks to be mined.
MAXTAProcess software maps how this will play out over a week, month or year.
Simulation
The digital twin software is well known for its ability to run complex simulations and deliver highly accurate results.
It allows engineers and geologists to ask questions of the model – most often in relation to drilling and blasting.
MAXTADrill &Blast has different drill and blast design options or levers that engineers can utilise to test the effects of variables such as drill patterns, explosive types, blast hole diameters and more.
These can be observed in relation to dig rates, crusher throughput and mill output.
Interesting insights are often uncovered through these simulations such as certain inputs improving the performance of one crusher but impeding the performance of another.
Hence it is possible to run simulations until both are optimised for maximum total output.
Exploration applications
A natural progression for the use of this big data is in brownfield exploration.
MAXTA provides the opportunity to better predict the behaviour of ore in the plant and even the grade, so it can also be a useful tool for exploration modelling along strike or down dip.
As with all forward prediction models, digital twin provides the ability to learn from past data and that is also true for exploration planning.
Reconciliation
Blind testing by users of MAXTAGeomet software has returned month-by-month accuracy of between 0.5-1.9pc accuracy when analysing tonnage of mill throughput.
When this is compared to conventional empirical static testing accuracy levels of between 10-20pc error – the difference in accuracy is staggering.
A predictive model capable of delivering results accurate to less than 2pc is a powerful tool and allows mining companies to reduce uncertainty surrounding their projects and articulate a lower risk opportunity to potential investors, or financiers.
This increases their ability to raise funds in a highly competitive market.
Optimisation
An optional optimisation layer is built in to MAXTA which goes over the top of the machine learning level.
Effectively, the optimisation level asks the model ‘how do you get the best performance for particular geological conditions?’
Primarily used for processing plant use, the same principle is also applied to drill and blast.
Again, through simulation, the model is able to suggest the optimal design based on historical performance.
Education and acceptance
Providing accurate models and simulations in this manner is a big change for a lot of people, according to Penny.
“It is an area we have been working on over the last two years in particular,” Penny said.
“How do you present optimal decision support in a way that engages the majority of engineers and geologists?
“Tech guys already understand the advantages but in order for the benefits to be scalable, there is a need to engage all users.
“We need to make it part of the normal workflow and enable people trust it.
“Engineers and geologists on site, who are required to deliver KPIs need faith in the tool.”
To further enhance user uptake and confidence in MAXTAInterp was added.
This includes machine learning ‘glass boxes’ where users can see the relationships between inputs and how they interact.
Most mine digital twins have between 40 and 80 inputs, so when users are able to literally see what is going on inside the MAXTA models, they are able to ‘sense check’ these interrelationships and ask themselves, ‘Do I believe that this is correct?’
After asking questions and seeing results that agree with their own experience, users can then trust the system in an operational environment and begin to use it to its full potential.
A real-world example of this acceptance and confidence was experienced by PanAust.
MAXTA was used to analyse and model their tailings grade, which is a good proxy for recovery rates.
The modelled data had good agreement with historical data and performance and clearly reproduced areas in which they had experienced difficulty with recovery.
This resulted in engineers and geologists alike, placing a high degree of confidence in the MAXTA models .
Implementation
Implementing the MAXTA on a new site typically follows a few distinct steps.
Firstly, the mining team identify a particular business challenge and the Petra team then determine if it is suitable for MAXTA.
Then, the site sponsors the next phase of work to be done, which often involves a business analyst or improvement specialist who captures the required data and supplies it to Petra to build into a model.
When the first model is complete, Petra submits it for stakeholder feedback.
This meeting normally becomes an engagement workshop where all parties check that the right areas have been modelled.
“This often turns into an iteration of the initial development process as stakeholders tend to redefine their priorities slightly,” Penny said.
“It allows for a sharper project focus and the model is adjusted accordingly.”
Users have several options when it comes to how they will engage with MAXTA.
Naturally, it is available as a stand-alone web application, but it can also be integrated with a number of industry leading software suites.
Outputs can be directed into OSIsoft PI, whom Petra has a technology partnership with.
Maptek Vulcan software also has a technology partnership with Petra and their mine planning software is also able to accept inputs from MAXTA.
In this way, users can access MAXTA from a platform they are already familiar with, in addition to the web app if necessary.
Feedback loop
The way in which MAXTA develops a model based on historical and current data, reconciles the model and then provides, predictive modelling, simulations and even suggests optimal designs – often with an accuracy in the range of 2-15pc depending upon the fidelity (Monthly <2 pc error, hourly +/-10-15pc accuracy) – sets it apart in the mining world.
It is difficult to see how a mining operation could achieve this level of detail and fidelity with conventional software and ‘good old-fashioned know-how’; methods that typically return error factors in excess of 20-40pc.
Mines often struggle to execute to plan due to the uncertainty in their geology, but with MAXTA’s ability to machine learn from data, mines can now get their best day, every day.
That is money in the bank.
More information: Petra Data Science
07 3310 8774
[email protected]
www.petradatascience.com.au
