Your behavior appears to be a little unusual. Please verify that you are not a bot.


Cutting-edge approach to achieve the perfect ton

By |  January 12, 2021
After a shot is executed, data on the fragmentation, muckpile shape, vibration and noise can automatically be integrated to allow AI to refine its models for use in the next drill-shot plan. Photo: Picsguru/iStock / Getty Images Plus/Getty Images

After a shot is executed, data on the fragmentation, muckpile shape, vibration and noise can automatically be integrated to allow AI to refine its models for use in the next drill-shot plan. Photo: Picsguru/iStock / Getty Images Plus/Getty Images

How do you get the perfect ton and solve a problem like fragmentation? And how do you take into account all of the different parameters that affect fragmentation?

Traditionally, blasters have tried empirical models such as the Kuz-Ram, but they are unable to incorporate all of the various parameters. So, while being easy to implement – and better than nothing – they are still prone to errors and accuracy issues.

Accurate fragmentation prediction models are critical to achieve good blast results and downstream cost savings. Still, due to the very complicated nature of all the parameters and their non-linear relationship, researchers are turning to AI (artificial intelligence) to try to solve the problem.

What is AI?

AI involves the programming of algorithms that mimic human thinking such as learning and problem-solving. The algorithms are programmed to operate within a data set, identify patterns in that data set, and then, from those patterns, make predictions about what would happen if new data were entered.

The algorithms can be trained to achieve a desired result by testing different combinations of variables to determine which combination will accomplish the desired result. Because they can carry out these combinations at a speed of millions per second, they are able to test every possible combination quickly, identifying the combinations that work.

Because AI only needs to identify data combinations that achieve the desired result, AI can come up with very strange ways to accomplish the task. For example, in a training exercise where AI had to make a stick figure run across an obstacle course and cross a finish line, AI created a stick figure with enormous legs that merely stepped over the entire course. Humans had to teach the AI that this was not an acceptable way to solve the challenge; that AI needed to create a figure that could run and jump instead.

AI must be trained repeatedly by humans until it is able to reliably predict outcomes and solve problems. Eventually, AI is able to eliminate the less successful ways of achieving a result. This is called machine learning, the process by which data and statistical techniques are used to help the algorithm learn how to get better at a task.

Machine learning algorithms can detect incredibly complex patterns in mountains of data, and they’re thus able to better account for nonlinear relationships. They can, for example, account for all of the parameters that influence fragmentation.

Machine learning algorithms can be applied to any kind of data, and that’s why you see an explosion of AI solutions in nearly every industry.

One type of AI

One example of machine learning technology that is prominent in mining, drilling and blasting is computer vision. Computer vision is simply training algorithms to interpret data gathered from images – in other words, identifying patterns in pixels.

These images can be gathered by not only a variety of sensors, but also numerous devices. Sensors include hyperspectral, infrared, magnetic, photographs and a number of others. The devices that collect this data include satellites, cameras and drones.

AI solutions are not limited to images. They can be applied to any kind of data, including data from smart tools such as smart drills, autonomous vehicles and seismographs.

The AI secret sauce

The power in AI and what makes it such a valuable addition to the toolbox is the feedback loop.

The algorithm is trained on a data set. The AI identifies patterns and makes predictions that are then compared to a control set of new data. The models are further refined to match the control set. Then, they are tested on more data.

With each repetition, the models get increasingly accurate and advanced. They are able to identify nuances specific to particular situations and make incredibly powerful predictions from incredibly complicated sets of data. They provide insights that would take humans weeks or months to ascertain – if humans can ascertain them at all.

Applications in your world

Data can be gathered from a number of different sources to create powerful interactive models.

AI is generally trained before users begin to upload their own data to it. Once users begin to add their own data, the model starts to customize itself to their site. The breadth of the data the users add to their AI determines how smart the AI can become and how much it can learn.

AI can be used for a limited purpose – for example, for shot design and fragmentation prediction. But without further input, such as fragmentation analysis from a post-blast drone flight, the model will not be able to verify and refine its predictions for that particular site.

AI will continually be trained in the background on general data, leading to better predictions. But the site-specific nature will be hampered.

Learning from users

In addition to blast planning and fragmentation analysis, another way to improve site-specific AI is by creating annotations for the data. Essentially, as users interact with the data, AI learns from any corrections users make.

A great example of this is muckpile prediction. In this case, the computer vision AI determines the muckpile boundaries. If users decide to adjust the boundary, then AI can learn from those adjustments.

In drilling and blasting specifically, pre-blast data like blast design parameters, smart drill logs and geotechnical data is used to create an incredibly detailed “digital twin” of a bench face that can make numerous predictions about post-blast results. 

If post-blast data such as fragmentation analysis, muckpile shape and vibration data are added, AI can verify and refine post-blast predictions. Alternatively, if users have an outcome they want to achieve, AI can recommend the pre-blast parameters based on specific site data that will best achieve the outcome.

Using the feedback loop, pre-blast and post-blast data from every subsequent blast is used to better the model of a specific site. In this way, each site has its own model to give the best possible predictions and solutions tailored to the site.

After each blast and each new data set is uploaded, the model provides better blast designs, drilling plans, fragmentation, vibration control, overpressure control, flyrock prevention, ore movement, muckpile shape prediction for equipment use and cycle time planning.

For example, if users utilize geotechnical data to supplement their 3D model, blasters will be able to take into account bedding planes, discontinuities and different rock hardness while designing a shot. Also, drillers will be forewarned about geological conditions such as dip plane and direction prior to drilling.

After drilling, boretrak or MWD (measure while drilling) data can be added to modify blasthole explosives-loading plans. A site record can also be created so AI can make predictions regarding future drilling and blasting in the area.

After a shot is executed, data on the fragmentation, muckpile shape, vibration and noise can automatically be integrated to allow AI to refine its models for use in the next drill-shot plan.

AI at scale

One of the biggest challenges to maximize AI’s ability to help the industry become more efficient is the current pre-blast and post-blast data silos.

Right now, the different units act very independently from each other at larger site operations that are often handled by contractors who don’t share data, limiting AI’s effectiveness. Without being able to utilize data from beginning to end, several opportunities for improvement are lost.

So while AI may solve the problem of rock fragmentation, overall operation optimization will require more coordination on behalf of the mine.

On the other side, one enormous opportunity is data’s universality. Insights gathered at one site with a particular geology can be used at other sites with similar geologies. What worked well in one situation could be applied to others, as there is no need for everyone to struggle with the same problems.


Blasting variables

What blasters can control

Burden
Charge length
Drill hole diameter/depth
Explosive type – strength and energy
Powder factor
Priming systems
Spacing
Stemming
Timing and sequence

What blasters can account for

Compressional stress wave velocity
Joint spacing and condition
Presence and depth of water
Rock-specific gravity
Rock strength

Source: Strayos


Ravi Sahu is CEO of Strayos.


Comments are closed