Hardware Aware Acceleration For Deep Neural Networks

The result of this project (which will be demonstrated by a use case) can make health equipments to be used outside of hospitals. This is achieved by reducing the computation cost of running Deep Learning models by 3rd party tools and use our accelerator solution to run the size reduced and optimized model. This greatly helps to lower the barrier for using costly equipments and make them more affordable and reachable to people in need of these equipments.

Développement d’un outil d’aide à la décision pour la sélection de sites pour des entreprises manufacturières en Amérique du Nord

Les dépenses d’investissement en capitaux d’une entreprise manufacturière représentent un secteur important du développement industriel : les acquisitions, le développement de marchés, la recherche, ou l’ajout de lignes de production dans une entreprise transnationale, s’accompagnent souvent, en effet, de flux de capitaux et de main d’oeuvre importants. À partir de différentes sources de données démographiques et de marché, publiques et privées, le projet vise à développer un outil d’aide à la décision pour la sélection de sites pour des entreprises manufacturières.

Fusion de SLAM et GPS pour véhicule militaire autonome

L’entreprise Rheinmetall Canada développe un véhicule militaire autonome qui a la faculté de se déplacer dans des environnements jugés dangereux et accidentés, sans l’intervention d’un pilote. Ce véhicule peut accomplir des missions telles que la protection des soldats et le ravitaillement des troupes.

Generation of correlation hypotheses between Adverse Events (AEs) and NamedEntity Recognition (NER) of drugs in social media and scientific journals usinga machine learning approach

Pharmacovigilance (PV) has evolved and grown more complex over the past 5 to 10 years due to increasing data volumes, evolving regulations, influence of emerging markets and the emerging social media and innovative technological advances.
Fast detection of Adverse Drug Reactions (ADRs) could allow the pharmaceutical industry to anticipate and then to control more efficiently eventual risks associated to taking some medications.

Improving energy system planning solutions by accounting for inherent uncertainties through robust optimization - Year two

More and more distributed energy resources (smart loads, self-generation, electric vehicles, etc.) are installed directly at the customers. This causes fluctuations in the distribution network that can reverse the power flow or increase the cold pick-up effect. The infrastructures in place have not been designed for this new reality and they must be adapted accordingly, and ideally, at minimum costs.

Study of magneto-elastic torque meter for power steering system

Permanent magnets and magnetic field sensors are widely used in the automotive industry for the detection of the rotational speed of a gear or the torque measurement of gearbox shafts, vehicle axles, and electric motors. For instance, the action of a driving wheel can modify a magnetic pattern encoded in a transmission shaft. Magnetic sensors positioned around the shaft can detect the stress applied on the driving wheel. This is what we call a magneto-elastic torque meter. It provides the signal for the operation of the power steering system.

Investigation of the potential of static liquefaction of tailings by taking into account the evolution of the hydro-geotechnical properties during and after their deposition - Year two

Mines generate large quantity of tailings. In most cases, they are transported by pipes and deposited in tailings ponds and confined by dams. To limit the footprint and land area of tailings pond, the dams have to be uplifted progressively with the increase in the tailings level. Several methods exist to uplift the tailings dams. Our partner is particularly interested by the upstream dam construction and a critical concern is how to evaluate the maximum height of the uplift to avoid any static liquefaction. Several numerical models exist to this end.

Study on the hydro-geotechnical properties and establishment of a numerical model for waste rocks - Year two

Mines produce large amount of waste rocks, mostly disposed on ground surface in form of pile. In underground mines, waste rocks are increasingly used to construct barricades to retain mining backfill in the stopes. Waste rocks can also be used as inclusions to accelerate the drainage and consolidation of tailings. To properly evaluate the stability of these infrastructures, numerical models are needed. However, the existing numerical models suffer from two major limitations.

Mode Split Prediction for Rotating Disks with Flexible Stator Coupling

Hydroelectric turbines vibrate when they operate, due to their rotation in water. Those vibrations, if not studied and looked after, may damage the runners, resulting in a decrease of production and operating range. Therefore, these turbines should be numerically modeled and submitted to simulations. However, as they are complex objects and it would thus take a long time to model them, it is possible to represent them with simple geometric forms, such as cylinders or disks, that still describe well the physical phenomena involved.

Méthodes de prédiction pour la collecte d’huile usagée

Le projet de recherche porte sur la prédiction probabiliste d’une demande intermittente.
On cherche à prédire la distribution statistique du niveau de remplissage d’un réservoir en fonction du temps, à l’aide des données des derniers services. Ces prédictions permettront d’optimiser les fenêtres de temps des prochains services. De ce fait, on augmentera le taux de remplissage moyen des réservoirs collectés ainsi que la profitabilité des routes.