The miner conducted a study to see if AI technology could be successfully applied to its concentration processes at its Aitik copper mine in northern Sweden.
Johannes Sikström, development engineer and project manager for system technology at Boliden, said his division developed dynamic simulations of the mine's processes.
"These simulations can be used in the same way as a game, where we define what is a win and what is a loss. In the case of self-learning algorithms - so-called deep learning or reinforcement learning - the challenge is the great quantity of data necessary for the algorithm to learn enough about the system for it to make effective decisions," he explained.
In previous projects, Boliden primarily researched machine-learning techniques that analyse data without allowing the algorithm itself to influence it. The aim of the AI project was to allow the algorithm to ‘self-learn'.
Following initial studies into suitable tools, together with Anders Hedlund at data analytics company BI Nordic, the project led to a degree project involving ABB and Boliden. Max Åstrand from ABB was appointed supervisor, while his colleague Mattias Hallén took on the job.
"We directed our attention to the grinding process in Aitik, where we have a well-developed simulation model. We wanted to see if AI was able to do better than our existing control strategy," said Sikström.
"Mattias did a fantastic job setting up the architecture and getting the various environments to ‘play ball' with each other. We were then able to test various algorithms and different goal functions."
The first part of the project was to test a ‘Q-learning' reinforcement algorithm; these algorithms are used to find the best action to take in a given situation. The aim was to control the mill's load within a given range. According to Boliden, after around 40 attempts, the algorithm taught itself to do so; however, it solved the task in a way that wouldn't work in practice.
"In the next step, we investigated the ability of the algorithm to optimise a gain instead of optimising a process variable. The goal function for the gain was created as a theoretical model using metal prices, grinding and throughput, for example," the company explained.
As a result, Sikström said the AI algorithm was able to produce a greater gain.
"So-called wall time was around 80 hours before AI had learned to run the process profitably, in this case equivalent to a plant operating time of more than 300 years. The study highlights the value of simulations, and the AI technology shows exciting development opportunities for Boliden's future process control," he added.
During the study, the AI algorithm was able to produce a better result than the miner's current control method; however, Sikström believes that further studies are necessary before the miner can approach a viable production solution.
"Several technical details need to be resolved, and it is important to use accurate simulation models and well-defined goal functions. Because an algorithm is only able to solve the problems formulated for it, process know-how and experience are at least as important in this type of development as classic process control," he concluded.