4 Presentations

• 
  03:30 PM - 03:55 PM

  Stochastic Mathematical Programming Framework for Waste Management in Industrial Mining Complexes

  • Victor Freire Guimaraes, presenter, McGill - COSMO Laboratory
  • Dimitrakopoulos Roussos, Université McGill

  Effective waste rock management is crucial for long-term mine production planning. Ignoring the role of potentially acid-generating (PAG) waste rock can lead to significant treatment costs to avoid acid rock drainage (ARD). Encapsulation of PAG material can prevent or mitigate ARD by limiting exposure. Traditional practices do not optimize production schedules while addressing this. This work integrates waste management and progressive reclamation into a simultaneous stochastic optimization framework that applies multi-neighborhood simulated annealing to incorporate non-linearities in the objective function and solutions in a reasonable time frame. Uncertainties in acid generation are addressed through geostatistical simulations of the rock’s geochemical properties. A case study at a copper-gold mining complex shows successful encapsulation with minimal financial impact.

• 
  03:55 PM - 04:20 PM

  Méthode non paramétrique de détection de signaux faibles pour la maintenance anticipée, appliquée à une flotte de véhicules miniers

  • Martin Cuenot, presenter, Polytechnique Montréal
  • Bruno Agard, Polytechnique Montréal, CIRRELT
  • Michel Gamache, Polytechnique Montréal
  • Souheil-Antoine Tahan, École de Technologie Supérieure

  Dans le contexte industriel minier, la disponibilité opérationnelle des équipements est essentielle. Il ne s’agit pas seulement de réaliser des maintenances après défaillance, mais d’anticiper les dysfonctionnements. Les approches classiques estimant la durée de vie restante, via des modèles mathématiques d’inférence statistique et/ou d’apprentissage machine, nécessitent un historique de pannes assez exhaustif. Cet historique n’est malheureusement pas toujours disponible ou fiable. Par conséquent, une autre manière consiste à détecter en amont des signaux faibles indiquant une altération dans le fonctionnement ou une défaillance potentielle.
  Notre approche vise à surveiller des indicateurs spécifiques du comportement de la machine (ex. productivité spécifique, consommation spécifique) et à détecter des observations atypiques ou tout comportement différent de l’historique récent de la flotte de la machine étudiée.
  Pour cela, les métriques comportementales sont normalisées par une transformation iso-probabiliste non paramétrique vers une distribution normale, puis décorrélées à l’aide d’une décomposition de Cholesky. Cette projection dans un espace latent permet de mettre en évidence des divergences de comportement. Plusieurs groupes d’observations (un grand nombre dans le cas général ou un sous-ensemble d’observations au contexte similaire) peuvent être utilisés afin de créer cet espace et pour l’exploiter dans la détection des divergences (globales et contextuellement atypiques). Nous appliquons cette méthode dans le cas de camions miniers effectuant des trajets pour remonter du minerai afin de détecter des signaux faibles, potentiellement indicateurs de défaillances. Nous montrons la pertinence de la méthode et son potentiel pour un déploiement dans un contexte industriel.

• 
  04:20 PM - 04:45 PM

  STOCHASTIC SHORT-TERM PRODUCTION PLANS INTEGRATED TO FLEET SIZE OPTIMIZATION, SHOVEL ALLOCATION AND TRUCK DISPATCHING IN OPEN PIT MINES

  • Lee Zamalloa, presenter, McGill University - COSMO Stochastic Mine Planning Laboratory
  • Roussos Dimitrakopoulos, COSMO Stochastic Mine Planning Laboratory, Université McGill

  The estimation of the fleet resources in an open pit mine is a complex problem commonly discussed at the early stages of the mining operation. It must consider short- and long- term production plans, mine geometry, safety policies, communications constraints, and many other factors subject to different degrees of uncertainty and circular analysis. Therefore, the fleet size (number, equipment capacity, models, etc.) is usually first evaluated under generalist scenarios, to be later adjusted based on more realistic operational parameters. This preliminary fleet is then used by the strategic planners to calculate the short-term production scheduling, shovel allocations and truck matching, setting up the base framework for the future truck dispatching system.

  However, most of the values used during the planning stage are averages or best estimates, and the output calculations may not result in exact matches once the short-term production plans are put into execution. Thereby, once in operation, the allocated fleet may be excessive or insufficient for the production plan to be met. While data such as metal content (grade) or rock properties from the ore blocks can be reconciliated once the Run-Of-Mine material is delivered at the processing facility, the size of the fleet cannot as it was planned based on imperfect information from the long-term planning stage. Once verified, the variations in the material quality and quantity may result in unexpected shovel movements and exceptions for truck rematching/rerouting. All these changes add an unexpected increment in the transportation cost, and potential adjustments in the fleet plans for the following periods, creating a snowball effect with unpredictable outcomes. The element of variability affecting the fleet plan is not restricted to material properties, but also to equipment availability, communications, haul roads, power lines, and other sources of uncertainty, creating multiples scenarios on which to decide.

  This research project integrates elements of uncertainty into the short-term fleet management plans. The variabilities are incorporated as distribution models to the stochastic mathematical optimization, offering a broader view on the potential changes in the shovel allocations and truck dispatching. By adding the factor of probability to the fleet plans, the dispatch operator can make more informed decisions based on the likelihood of the fleet resources to be allocated in different working areas for optimum results. The new fleet plan will then reduce the risk of equipment misallocations and rerouting.

• 
  04:45 PM - 05:10 PM

  Improving the performance of production scheduling for open pit mines using cutting planes with additional variables

  • Venkat Akhil Ankem, presenter, GERAD, Polytechnique Montreal
  • Guy Desaulniers, GERAD - Polytechnique Montréal
  • Michel Gamache, Polytechnique Montréal
  • Vincent Raymond, RioTinto

  The open-pit mine production scheduling problem (OPMPSP) is a fundamental optimization challenge in mining engineering, characterized by large-scale integer programming models that require substantial computational resources. Our study considers the discretized ore body representation known as the block model, to strategically determine the schedule of block excavation (extraction time, quantity, and processing decisions) over a planning horizon while adhering to the operational constraints and maximizing the net present value of the profit. The problem is modeled with a mixed-integer linear programming formulation with all the constraints considered in real-life industrial applications. We propose a methodology to improve the computational efficiency of OPMPSP by incorporating cutting planes that utilize additional decision variables employed in real-time industrial applications. Including the additional decision variables into the cutting planes and removing the redundant cuts improves the overall solution time compared to the traditional cutting planes, thus enhancing the performance of existing cutting planes and further strengthening the formulation. We also introduce a new set of cutting planes based on blocks that require multiple time periods to deplete, which have been largely overlooked in the academic literature. These new cutting planes further improve the solution time when integrated with the model. Through extensive computational experiments on real-time mine instances, we demonstrate that the proposed methodology significantly reduces solution times while maintaining optimality, making it a valuable contribution to large-scale mine planning optimization. All the models are also integrated with a rolling-horizon heuristic to assess their effectiveness at solving larger instances, and their compatibility with other heuristics. Thus the study contributes valuable insights and practical approaches to address the complexities inherent in large-scale OPMPSPs.