Why is Implementing AI So Hard in Mining?
Artificial Intelligence has transformed industries across the globe. From Advertising to Transport to Consumer Goods we have seen dramatic shifts in the way services are delivered.
The dynamic optimization enabled by AI has allowed businesses to maximize the utilization of their assets and increase return on investment. In parallel, the automation driven by these algorithms has streamlined operations to improve organizational efficiency.
Business leaders and experts all agree that the Mining industry has just as much potential as any other to benefit from AI. Mining companies all have digital roadmaps filled with high value use cases to address.
However, studies consistently indicate that Mining is not realizing as much value from AI as other industries are. Too many projects are identified, scoped, and prioritized only to fall flat during implementation.
So why is implementing AI so much harder in Mining? Here are 2 reasons:
- Mining Data is More Complex
At first glance, Mining doesn't seem so different from other industries. At least, from a modeling perspective. It can be represented as a series of operational steps, each with a certain:
- throughput capacity (e.g. 20Mtpa) and cost (e.g. $30m / year)
- set of decisions (e.g. number of trucks in a haul circuit)
- set of contextual parameters (e.g. type of rock)
This model is easy to optimize if all your data are numbers and categories. In retail, you can optimize stock levels based on historical sales, correlations, region labels etc. because much of that data already exists as numbers and categories.
In Mining however, life isn't that simple.
The decisions and contextual parameters for Mining often come in the form of geospatial, visual, and time series data.
For example, one critical contextual parameter affecting the system throughput is rock jointing.
Rock jointing can vary significantly across a site and the only practical way to measure it is to visually scan the highwall before each blast. Turning a visual inspection of a Mining face into a number or category that can be used for optimization requires a much more complex AI model trained specifically for Mining environments.
Turning a visual inspection of a Mining face into a number or category that can be used for optimization requires a much more complex AI model trained specifically for Mining environments.
Another example, an important decision made on every mine site, is "how do I design this haul road"? This is not a decision that can be represented as a simple number or category. The answer takes the form of a geospatial model synchronized to a site wide survey coordinate system. Again, turning a geospatial model into data that is useful for optimization requires specialized algorithms.
Finally, to properly connect data across different operational steps, you have to deal with time series data. The rock in the grinding circuit today may have actually been blasted 2 weeks ago and it can be almost impossible to trace it through the system. Determining the ideal processing plant settings for a specific rock type depends entirely on marrying up this geospatial and time series data.
The bottom line is:
mining data is unique so it requires unique algorithms.
2. There Aren't Many People Who Understand Both Mining and Data Science
Mines tend not to be located in the middle of technology hubs like downtown San Francisco. It might sound obvious but proximity is a key factor influencing human behavior. The people who know about Mining are not often bumping into the people who know about data science and this has created an exposure and translation gap.
Mines tend not to be located in the middle of technology hubs like downtown San Francisco.
Additionally, the complexity and uniqueness of Mining data makes it difficult to learn for data scientists and software engineers. The first challenge they face is getting access to the data (mining data is generally closely guarded); there are barely any open datasets for Mining. Next, they need to find an interpreter to teach them what the labels mean, how the numbers are related and what the actual objective is. This results in low engagement from the broader data science community. The only people with experience are the people who have been employed specifically for Mining projects.
However, even when data scientists and software developers are employed to focus on the Mining industry, it takes months to years to build the level of understanding required to deliver a solution that creates value. This is because it is not enough to simply understand the data. Developers must also have a deep understanding of the end users and how they will implement the model outputs.
It is not enough to simply understand the data. Developers must also have a deep understanding of the end users and how they will implement the model outputs.
For example, a perfect model that specifies the optimal drill pattern design for each shot is useless if it relies on data that can't be practically captured before drilling commences. AI solutions must be integrated into the day-to-day tools and processes of engineers and operators to prevent the implementation failures that are all too common.
So, considering the value that AI can generate for the mining industry, what can be done?
Miners Can Make AI Implementation Easier by Creating a More Open Tech Ecosystem
Mining companies have a lot to gain from addressing these industry-wide challenges. Fortunately, they are also in the best position to do so. Miners should apply lessons learned from other industries to build an ecosystem of data, cross-skilled data scientists, and open source libraries of code for parsing complex mining data.
Miners should apply lessons learned from other industries to build an ecosystem of data, cross-skilled data scientists, and open source libraries of code for parsing complex mining data.
Three key steps Miners should follow to create this future state:
Contribute to open data sets.
Improving the availability of data increases the ability of researchers and open source developers to create and refine tools which can then be used back on site. It may seem counterintuitive that people would develop valuable open source libraries for free but this is actually quite common in the most digitally advanced industries and these tools are creating significant value for incumbent players.
Partner and integrate before building in-house.
Creating demand for external services is vital for the health of a broader tech ecosystem. This builds a knowledge base among engineers and data scientists in outside organisations like consultancies and technology partners. It also increases commercial viability for start-ups and research projects which attracts talent into the Mining space.
To maximize value for Mining companies, internal development should not be undertaken unless the strategic advantage clearly outweighs the costs.
Develop flexible internal career pathways.
Mining companies can gain access to a wider network of expertise by allowing existing employees to explore data science through flexible work/study and part time work arrangements. Providing flexible roles for new hires can also help attract top talent from outside the industry.
It will take industry wide collaboration to build the open tech ecosystem required to make AI more effective in Mining.
We all agree that the benefits of adding AI are huge so let's work together to bring this future to life.
New technologies are rapidly changing the drilling, blasting, mining, and aggregates industries, empowering them in ways never before possible. Make sure you are taking advantage of the best tools available.
Check out our 2 Free E-books on AI applications for the drilling, blasting, and mining industries to see all the amazing advances that are available.
AI Guide for Drilling and Blasting
AI Guide for Mining
TLDR? Check out our videos instead:
Get the latest posts delivered right to your inbox