Modern mining is a high-stress environment for its people and equipment. Operations must identify profitable material, extract it efficiently, and then recover the product at the lowest cost possible. Any setback increases costs and delays.
Mostly Reactive Instead of Proactive
Traditionally, our industry has been reactive to hurdles as they appear. Staff continually plan for all contingencies but are inevitably faced with daily changes to their plans, changes that ultimately result in lower productivity and higher costs. These variations often happen because it is hard to obtain actionable information—rapidly—in a naturally irregular operation.
Limited Data Points and Slow Data Processing
Mining gets its most reliable data from long-term studies and models, short-term blasthole and face sampling, plant sampling, and…not much else. Most everything else is inferred. For example, ore processing relies on geo-metallurgical information. Unfortunately, this data is scarce because:
- Bulk sampling is expensive,
- There are few places where sampling can be undertaken effectively, and
- The turnaround time is very long.
In many cases, material handling is controlled by an equation using data from a metallurgical campaign performed months before. This scenario creates long uncertainty gaps in the operation as material is transported and processed.
“A critical root cause of lost productivity and efficiency results from not accurately understanding what material is being processed at any given point.”
AN INNOVATIVE SOLUTION
Continuously capture a comprehensive dataset of images using inexpensive off-the-shelf cameras positioned in many places throughout the mine. These images are matched with mining data and analyzed by cloud-based, machine learning algorithms to improve confidence in geo-metallurgical properties prior to material processing.
Abundant Data Points and Near Real-time Data Processing
Imago Chroma can help remove the fog of uncertainty. It uses inexpensive, off-the-shelf visible light and hyper-spectral cameras to record every step of material handling. Imagery is taken everywhere:
- as a shovel digs a face,
- as a truck is loaded,
- as the same truck dumps at a stockpile,
- as material is dumped into a crusher,
- or even as it moves through different processing stages in the plant.
“All this imagery contributes to performance predictions produced by deep-learning algorithms.”
Mine plans can be re-evaluated more frequently using these predicted performances. Plans are then adjusted to remove the high/low loads on equipment and avoid stressful environments that lead to downtime, lost production, and wasted resources.
IMPACT & VALUE
What It’s Not: Expensive sensors + machine learning solving isolated problems within a mine operation
It is becoming a commonplace practice to attach sensors to each piece of equipment or activity. These sensors are specially-built and feed information into improving that single task. For example:
- detecting particle sizes on a conveyor, or
- XRF geochemical analyses of a face.
These types of sensors are usually capital intensive, require significant calibration, and are difficult to install. An abundance of these sensors, combined with machine learning, will NOT revolutionize mining.
What It Is: Cheap, off-the-shelf cameras providing insights at lots of locations at a mine
Imago Chroma uses commodity cameras to capture material imagery at many locations. Current off-the-shelf cameras are of a high quality, disposable nature, and easily set up. Installing many of them is simple and has very low impact on an active mine operation.
Multiple, Varied Locations
Predictions should not be built from a single location or process. Instead, they should leverage the relationships between imagery and hauling, crushing, and other processing statistics. Our key innovation is how Imago links unrelated activities together cheaply:
- How might the crusher be affected as material dries on a ROM stockpile?
- Does this stockpile contain a rock type that affects plant recoveries?
- Is ore feed from this bench likely to be attacked efficiently during beneficiation?
IMPLEMENTATION & SCALABILITY
Optimism vs Reality
Imago understands that it is optimistic to capture imagery across the entire mining chain, combine it with so many different production statistics, AND find insights across related tasks.
Therefore, we have taken a considered approach to implementing Imago Chroma. There have been incremental steps to both prove the technological ideas and confirm their viability at operating mines.
- Built cloud imagery infrastructure,
- Introduced core tray photography management as a testbed product,
- Researched machine learning techniques to relate imagery across tasks,
- And prototyped the imagery capture in many different locations (i.e. down blastholes, in laboratories, from plant samples, etc.)
In addition to the core tray photography, two new Imago products are currently coming to market:
- Predictive ore delineation based on photographs of chip samples, and
- Predicting blast characteristics using downhole imagery.
We reduce overall implementation risk by delivering independent—but integrated—products. Each product combines with the others to increase our understanding of how mine performance is affected by different material types.
Not Just a Startup, but also a Growing Business
Imago is already a growing business. We were founded 12 months ago and already have a presence in many of the world’s largest mining companies.
“Over 90 million images are currently stored in Imago, with thousands being uploaded into it on a daily basis.”
There is currently a strong plan in place to expand Imago and its associated products. However, this expansion is constrained by too many ideas, a fast-moving environment, and financial limitations.
#DisruptMining can bring Imago Chroma to market sooner. We want an opportunity to capture imagery across an entire operation and correlate it with the mine’s ore processing performance. Both technical and financial support from GoldCorp would help with this endeavor.