On the Rocks

by Deirdre Kelly

photography by horst Herget

The digitalization of the design and maintenance of underground excavations is helping to mitigate some of the risks associated with underground mining, finds a new study out of York University. Data driven approaches, such as machine learning algorithms and equipment automation, enable rock engineering professionals to predict underground stress changes and complex rock deformation as mining progresses with greater accuracy, increasing the potential for safer and more sustainable mining practices.

“Machine learning presents an opportunity to extract more nuanced information from available data, and therefore make more refined engineering decisions,” says Josephine Morgenroth (PhD ’22), a Lassonde School of Engineering rocks mechanics engineer and researcher who studies the opportunities for using machine learning in Canada’s mining industry.

Josephine Morgenroth seated on large rocky area

Her peer-reviewed journal paper looks at the application of machine learning at Garson Mine, a 112-year-old underground nickel mine near Sudbury, Ont. The mine site is known for its challenging ground conditions and high stress environment. A series of seismic events have occurred at Garson Mine in the past, necessitating the need to gain a more informed understanding of the seismically active structures located underground. 

A manually calibrated and complex finite difference numerical model was introduced to assess seismic risk. This model is used to inform mine operations and scheduling to lessen some of the risks associated with Garson’s rock bursts (spontaneous failure of rock that can occur in mines).

But the manual model calibration is tedious and time consuming, leading Morgenroth, as part of her research, to propose a Long-Short Term Memory (LSTM) network to assist in finite difference model calibration, a means to forecast stresses in the mine. 

The significance of this approach is that it frees up rock engineers to validate model inputs and interpret outputs, instead of manually calibrating the finite difference model.

Two LSTM networks were developed for Garson Mine, which resulted from an analysis performed to determine the optimal algorithm architecture to adequately predict stress changes. Research is ongoing, supported by the Natural Sciences and Engineering Research Council of Canada. Says Morgenroth, “More accurate forecasts of changes in stress conditions will allow earlier intervention and reaction to challenging stress environments, leading to increased safety of excavations and mine personnel.” ■

Visit Link

Up Next

Inclusive Health Care

A York scholar confronts bias in AI-assisted medical research

Read More