Computer Vision in Mining Operations: From Process Monitoring to Drone Imagery Analysis

From roads and cell phones to wind farms and medical equipment, mining is the backbone of modern civilization, giving us the essential minerals needed for building infrastructure and advancing technologies. Essential to modern economies and daily life, mining processes must operate as efficiently as possible.

Modern processing plants are designed and engineered to use traditional instrumentation—such as level sensors, flow meters, and pressure gauges—to provide real-time data feeds. But in a handful of instances—particularly in plants built before the rise of Industry 4.0—the condition and performance of parts of the process can’t be measured using traditional sensors.

Instead, subject matter experts measure them using the “eye test.” These instances usually occur when there's a need to assess a wide area (such as a conveyor belt) to classify the conditions, but a precise measurement isn't critical—often, a straightforward “good” or “bad” rating will suffice.

At BCG X, the tech and design unit of BCG, we recently helped a client leverage computer vision—a field of AI that allows computers to understand and describe images accurately— to reduce the continuous visual monitoring required of operators. This, in turn, frees operators to focus on other value-adding tasks, like maintenance of the circuit. Through our work, we’ve found in these applications that deploying relatively cheap, general-purpose surveillance cameras is enough to deliver value, providing a cost-effective alternative to measurement-specific instrumentation designed for precision. In some circumstances, it may even come “free” if surveillance cameras have already been installed and AI can leverage these video feeds.

Since drone imagery is not new in mining applications—already used for pit progress tracking and stockpile inventory monitoring—it also provides another rich data source that can be used for “free” to complement the more established use cases of mine-specific monitoring and event prevention analytics.

In this blog post, we’ll expand on two specific use cases where we have seen success with the application of computer vision in mining operations: monitoring the condition of vibrating screens using surveillance cameras—and monitoring the operating landscape across a great expanse of land leveraging existing drone flights.

Use Case #1: Monitoring the Condition of Vibrating Screens Using Surveillance Cameras

Vibrating screens sort and separate minerals based on size, using mechanical vibrations to move materials over screens. Small particles fall through the screen while larger particles move down the length of the screen. These screens can become “blinded,” reducing the effectiveness of the separation process and thereby reducing the grade (i.e., quality) of the desired mineral or mineral concentrate.

In our work, we found that the screen's condition can be classified based on the shape of particles on the screen and the location of these particles across the screen's width. For example, the ideal working condition is a near-uniform distribution of round particles across the width of the screen. Any other observed conditions would constitute an issue with the screen—perhaps poor water pressure or a clogged screen needing cleaning.

A computer vision algorithm was developed to continuously monitor the condition of the vibrating screen, classify its condition as “good” or “bad,” and raise an alarm if the bad condition persists over a specified period. A vital part of this solution involved defining and implementing custom metrics that would aid the classifier in distinguishing between different operating conditions. We deployed the system in multiple client mineral separation plants and found it to improve the operating discipline of maintaining this piece of critical equipment, resulting in better product quality.

The figure above demonstrates a range of operating conditions (top row) that can be detected by the algorithm, using the custom real-time metrics (bottom row) to perform the classification.

 
Use Case #2: Monitoring the Mining Operating Landscape Using Drone Imagery

The purpose of any computer vision application is to identify key features in the field of view and convert them into valuable business insights. The task is similar to the classic “Spot the Difference" game, where players scrutinize two nearly identical images to catch subtle discrepancies. But these entertaining puzzles consist of just two small, static images. The human eye cannot continuously monitor and detect subtle but significant changes at the scale of an entire mining operation.

Imagine you have a system that can automatically pick up fine details that even human eyes can miss, even at a large scale. By leveraging drone data and the technical AI foundation of the classic game, we can ideate and support a multitude of use cases. In fact, we identified more than 20 high-value use cases at our client’s mining operation.

Above is an example of our foundational algorithm taking on its own "Spot the Difference” and being able to solve it for a split second.

With this foundation, one can extend the logic to enable use cases such as monitoring integrity of berms and road condition, tracking portable equipment (e.g. pumps, pipes, screeners, etc.) moves, tracking inventory of spares and work-in-progress material across the mine site and many more. In our client’s instance, for example, among other use cases we focused on building and maintaining a database of pipe inventory (and associated conditions) across the mine site. Leveraging the object detection foundations—and combining them with open-source Optical Character Recognition (OCR)—we demonstrated the ability to connect each length of pipe around the site and tie it to a unique ID in the pipe database. The combination of “spot the difference” AI and OCR enabled the database to be built from scratch—and any changes in pipe condition, pipe location, or pipe usage to be captured as part of regular drone flights across the site.

The Future of Computer Vision in Mining

Computer vision can help the mining industry in many ways beyond traditional process monitoring. From our experience, several use cases can be built immediately on top of existing surveillance camera feeds or drone flight imagery—without the need to inject capital for expensive hardware.

The alignment between computer vision and drone technology opens up many new opportunities in the mining industry. Providing a candid view of operations and tirelessly delivering deep analytical insights can enhance operational efficiency while promoting safety and sustainable operations.

As these technologies continue to advance, their potential to revolutionize the mining industry further is boundless. Mining companies that embrace these innovations stand to gain a competitive edge, ensuring their operations are not only profitable—but also aligned with best practices for environmental stewardship and worker safety.