Prédiction dynamique de durées de trajets de camions de transport de minerai dans les mines souterraines

Characterized by its challenging operational conditions and multifactorial versatility, the underground mining environment greatly complicates the task of predicting ore haul truck travel times. While this task is necessary for precise planning of the sequence of operations for the upcoming shift, conventional techniques are often restricted to simplistic statistical measures, which are sorely lacking in precision, or to human expertise. Naturally, this expertise is confined to a given mining site or even to specific routes only, its availability is limited, and the quality of its performance is difficult to quantify. In stark contrast to these techniques, our novel, robust, and sophisticated approach relies on data from emerging underground vehi-cle identification beacon systems. It combines a complex data preparation algorithm that generates multiple operational variables from this data and an integrated machine learning model capable of leveraging the observed time series to make its travel time predictions. Regardless of the underground mining site studied, provided it has a beacon detection system, our succession of models can predict ore haul truck travel times on all predominant routes with increased reliability, even when said beacon system exhibits serious failures. Even more uniquely, given the current literature, the proposed comprehensive model does not require other data sources for proper training, and the prediction variable values are all known in advance of each shift by mining operation planners. Our approach has been tested and vali-dated not only on the mining site where it was developed but also on another dissimilar site whose detection system had a major fault, demonstrating its robustness and generalizability to other mining contexts. Critical improvements to existing beacon detection systems were proposed, and numerous research avenues were suggested in light of the vast amount of work that remains to be done to perfect prediction techniques. With all these contributions, this thesis paves the way for substantial improvements in planning and operational efficiency in underground mines, marking a significant advance in the field of mining engineering.