Realistic image compositions with deep learning priors

Image composition is an operation that incorporates an object extracted from a source image to incorporate it inside a target image. This simple operation has a lot of practical applications in the advertisement, entertainment, and movie industries. However, it is tedious for an artist to achieve realistic image compositions. Indeed, it requires different lighting, shadows, or objects present in the target image will change the target object's appearance. One of these appearance changes is new contact shadows between the target object and the target environment.

Projects allocation and advanced scheduling of concurrent construction projects

This project proposes multi-criteria decision-making and mathematical programming (optimization) techniques to design an integrated decision-making tool to assist organizations in managing this complex process and achieving more efficiency. For the first problem, allocating project managers, we develop a multi-criteria decision-making method that will offer the possibility to optimize the usage of PMs' time considering the specific constraint and the knowledge necessary for the projects.

Operational framework and application prototype for integrating DfMA and BIM

Built environments have been experiencing low productivity issues for a long time due to poor collaborative processes, inefficient information exchange, and discontinuity between design and construction. With more and more construction projects moving towards off-site construction, the concept of Design for Manufacturing and Assembly (DfMA) is gaining momentum. This strategy is expected to be adopted in the construction industry ecosystem to improve the efficiency of project delivery.

Developing an Inflammation Intensity Score based on AI analysis of blood biomarkers

With chronic inflammatory diseases such as arthritis affecting the spine, a major clinical problem is that disease can advance over years with unrecognized persisting inflammation. Recently, we discovered that elevated concentration of certain substances in blood considered together, are highly associated with persisting local inflammation. This study will use data gathered from the blood chemistry analysis of 286 spinal arthritis patients followed for up to 12 years.

Multimodal Representation Learning from raw data to detect customers emotional state in the financial industry

Currently, call centres effort in this matter is largely reactive. Someone calls in, they are upset, and agents respond accordingly. However, this approach is not always most effective, especially with difficult customers. Therefore, knowing the customers current emotional state is very important for appropriate problem solving.

A Deep Learning approach to identify and localize room assets using handheld RGB-D sensors

With the availability of low-cost edge devices equipped with color and depth sensors, such as iPhone or iPad, 3D data capturing is becoming more accessible and convenient. Asset managers and building owners seek benefiting from this potential to accurately and automatically create 3D indoor models of buildings. In particular, having 3D indoor models containing objects of interest enables several applications, such as asset inventory and maintenance management.

Optimization of the tempering heat treatment cycle of large size forgings

Large size high strength steels parts used in transport and energy applications undergo several heating and cooling cycles during their manufacturing process (casting, forging, quench, tempering). Generally, in the tempering process the parts are heated in industrial furnaces and the impact of non-uniform heating on the subsequent steps is of critical importance. A non-uniform temperature distribution may result in property variation from one end to another of the part, changes in microstructure, or even cracking.

Identification of Cost-Effective Recycling Strategies for High-Performance Thermoplastic Prepreg Production Waste

The use of composites in transportation vehicles has been increasing for many years without a proportional response in the commercialization of recycling technologies to deal with this composites waste. Most of the waste generated during manufacturing and end-of-life is currently either landfilled or incinerated. Governments around the world recognize that such practices represent an obstacle in the effort to mitigate climate change and have, therefore, started implementing legislation such as the European Waste Framework Directive to promote waste reduction and recycling.

Realistic Few-Shot Learning

The main objective of this project is to investigate, develop and evaluate state-of-the-art deep-learning algorithms for joint few-shot classification and out-of-distribution (OOD) detection. Few-shot learning deals with the challenges of limited supervision, and OOD detection attempts to identify inputs that do not belong to the set of classes seen during training. The two research problems are in line with several applications that are of high interest to the industrial partner as they tackle realistic open-set and limited-supervision scenarios.

Adaptation de l’algorithme de Rainflow pour le calcul des cycles de chargement de manière incrémentale

Les éoliennes de taille industrielle sont conçues pour une durée de vie nominale de 25 ans. Plusieurs hypothèsessont retenues lors de la conception, notamment au niveau de l’estimation des conditions environnementales etd’opération. Il est pertinent d’utiliser les données opérationnelles pour avoir une estimation propre à une éoliennedans le but d’estimer sa vie résiduelle et optimiser ainsi les opérations de maintenance ou de remplacement. Unsignal clé est la vitesse du vent qui permet de calculer les contraintes sur le rotor et les pales.