Reducing Environmental Risks in Mining Using Machine Learning and Data Fusion to Improve Tailings Analysis

Project reference: MN73

Application deadline: 10th April 2023

How to apply

After mine closure, tailings (mining waste) can contain elements like arsenic, lead or copper, and bear the risk of ground water pollution, soil contamination and other environmental damage. These lead to serious impacts on human health, environmental quality and disrupts socioeconomic development. To mitigate these dangers worldwide, tailings need to be analysed and potentially reprocessed – a highly topical issue that accompanies the mining industry today, and in the coming decades.

The classical way of tailings investigation is based on samples from holes drilled into the tailing’s bodies; these samples are then analysed in the laboratory. This is an expensive and complex process. J&C Bachmann, has realized a push probe technique that analyses materials in the tailings without extraction. This technique drastically reduces the cost of tailings analysis and makes it more viable for governments and companies to progress on the issue. J&C Bachmann will provide real-world measurement data from measurement campaigns as well as lab analyses.

Detailed project description:

Mining produces tailings which are rejects or resultant waste stream after performing the process of separating valuable fraction from the uneconomic fraction of an ore. It was widespread practice that these tailings were not monitored. After mine closure, tailings have often been left to their own. Consequently, elements like Arsenic, Lead or Copper are washed out over time and bear the risk of ground water pollution, soil contamination and other environmental damage.

To mitigate these dangers worldwide, tailings need to be analysed and potentially reprocessed – a highly topical issue that accompanies the mining industry today and in the coming decades. The United Nation and business communities developed first international standards for tailings management in 2020 after a deadly tailing disaster in Brazil. Besides the need to avoid dangers to the society, it is often financially viable to pursue the reprocessing as valuable minerals can be extracted from the tailings with modern and clean extraction methods.

The fast and accurate measurement of tailings for determination of mineralogies is a basic requirement for the successful accomplishment of this task. The classical way of tailings investigation is based on samples from holes drilled into the tailing’s bodies; these samples are then analysed in the laboratory. This is an expensive and complex process. The company involved in this project, J&C Bachmann, has realized a push probe technique that analyses materials in the tailings without extraction. This technique drastically reduces the cost of tailings analysis and makes it more viable for governments and companies to progress on the issue and thus speeds up action where required.

The push probe is equipped with spectrometers for X-ray and infrared radiation. X-ray and infrared spectra complement each other ideally. The dimensionality of the data is high given that the complexity related to the analysis. We expect that data fusion and machine learning techniques will allow a broad band of mineralogies to be quantified in high accuracy. J&C Bachmann will provide real-world measurement data from measurement campaigns as well as lab analyses. The high data quantity requirements for machine learning are planned to be fulfilled by simulation models with standard industry tools, and inter- and extrapolation techniques.

Research objectives:

• Simulate new spectrometer information using industry-standard simulation tools (e.g., Geant4).

• Develop and train machine learning-based classification models on real-world and simulated data from spectrometers to determine and quantify mineralogies.

• Fuse spectrometer information to a data set and train an advanced machine learning model on the fused data to determine and quantify mineralogies.

• Compare the standard-methodology for mineralogy determination with the results from machine learning models.

• Evaluate the dependency of the model on specific parameters of the tailing data used.

Research questions:

• Which of the tailings can be reliably determined and quantified using X-ray and infrared spectra and how accurately?

• How can the available data be multiplied using simulation models and extrapolation techniques to generate significant amount of data required for developing advanced machine learning models?

• Does fusion of the sensor data of measurement instruments allow a more comprehensive and reliable determination of the mineralogies than the simultaneous and independent use of instruments?

• How can the information be transferred into a mineralogic model of a tailing to allow neutralization of poisonous elements in the tailing with minimal ecological impact?

• How to transfer and adapt machine-learning models developed for specific kinds of tailings (e.g., coal tailings) to other commodities (e.g., copper tailings)?