AI based digital twin for optimizing mineral recovery


Outotec and Labra AI built an AI tool for predicting the most important parameters of a mineral concentrator process to enable optimization of mineral recovery

Outotec and the artificial intelligence company Labra AI built a series of deep learning models which predict the most important parameters in mineral processing. The tool enables creation of scenarios and optimization of those scenarios to maximize economical value in mineral processing. The AI tool was made possible by Outotec’s process knowledge as well as Labra AI’s skills in developing complex artificial intelligence models.

What if you could predict your business’ most important parameters? Or create future scenarios of these parameters quickly based on your external inputs? And finally, what if you could create hundreds of these scenarios and optimize for value?

Most would agree that would be very valuable indeed. This was the objective Outotec and Labra AI took: Develop a deep learning software solution which can make fast and accurate predictions on important mineral processing parameters, used to create scenarios for maximizing output value.

A key factor in predictive models is the trade-off between model detail and processing speed. An increase in the other means the other can be decreased. Traditional physical simulation models are often slow and reliant on sequential processing. They calculate every interaction in a process explicitly in every time step. A converse approach could be a look-up table, which is built entirely on memory where predictions are pre-calculated for every possible occurrence. Mind you, the number of possible occurrences in a system increases exponentially with every input, which means a look-up table approach doesn’t scale well to larger problems with complex and data-intensive processes. And the calculations to fill the table would take considerable time, too.

Neural networks connected in series are a powerful tool for time series prediction.

The grinding mill is an essential part of mineral processing.

The answer, as organizations are figuring out, is deep learning - a balance of memory and processing power - while delivering accurate predictions. The weights of a neural network are stored in memory, while calculations can be executed concurrently as a result of the neural network architecture and modern GPUs. The deep learning approach was used in this problem to achieve accurate predictions learned from data and make predictions fast enough to create scenarios.

Outotec and Labra AI worked on the project cooperatively with each bringing their own strengths. Outotec brought its world-class knowledge of mineral processes and process modeling expertise, and Labra AI delivered in the areas of deep learning, physical modeling and software development. The collective effort resulted in a highly accurate and fast AI tool for predicting the most important parameters in mineral processing.

About Labra.AI

Labra AI builds stellar AI products and custom solutions - the kind of work that takes your business to the next level and maximizes the value of your data.

Martti Pankakoski, CEO,
martti.pankakoski(a)labra.ai

About Outotec

Outotec develops leading technologies and services for the sustainable use of Earth’s natural resources. Our 4,000 top experts are driven by each customer’s unique challenges across the world. Outotec's comprehensive offering creates the best value for our customers in the mining, metal, and chemical industries. Outotec shares are listed on NASDAQ Helsinki.

www.outotec.com