Publications

Aspect-based sentiment classification has become a popular topic in natural language processing. Exploiting dependency syntactic information with graph neural networks has recently become a popular trend. Despite their success, methods that rely heavily on a dependency tree face major challenges. This concerns the alignment of aspects and their word sentiments due to the richness of the language and the fact that a dependency tree might produce noisy signals from unrelated associations. This paper introduces a Dual-Relational Graph Attention Network (DRGAT) that fully exploits syntactic structural information and then the sentiment-aware context (e.g., phrase segmentation and hierarchical structure) of the constituent tree of a sentence. Additional constituency and dependency attention mechanisms provide comprehensive syntactic information across words, thereby enabling an accurate connection between aspect words and corresponding sentiment words. Considering that the original parsed constituency tree may have a large depth, this could lead to words being far apart increasing the computational overhead. The constituency tree of each sentence is dynamically reconstructed by determining the importance of each relational node. Extensive experimental results on six English datasets demonstrated that fully exploiting syntactic information can achieve excellent sentiment classification results.

This research introduces an experimental framework based on 3D acoustic and psycho-acoustic sensors supplemented with ambisonics and sound morphological analysis, whose objective is to study urban soundscapes. A questionnaire that highlights the differences between what has been measured and what has been perceiveSd by humans complements the quantitative approach with a qualitative evaluation. The comparison of the measurements with the questionnaire provides a global vision of the perception of these soundscapes, as well as differences and similarities. The approach is experimented within the historical center of the Tunisian city of Sidi Bou Saïd, demonstrating that from a range of complementary protocols, a soundscape environment can be qualified. This framework provides an additional dimension to urban planning studies.

This paper evaluates the impact of urban renewal programs (PRU) on creating a safe environment. People’s perceptions of PRU and its impact on security were collected through questionnaires and sketch maps. This study applies a space syntax approach to analyse the spatial and structural properties of successive urban space evolutions and compare them with people’s perceptions of security. The findings reveal a strong correlation between areas with high levels of social housing concentration, poverty rates and insecurity. Surprisingly, visually controlled areas are vulnerable and attractive locations for drug trafficking. Density indicators have a minor influence on the perception of insecurity.

Le concept de jumeau numérique offre une multitude d'applications et d'utilisations dans divers domaines. Tout d'abord, il permet de simuler et de tester des scénarios virtuels avant de les implémenter dans le monde réel. Cela peut être particulièrement utile dans le domaine de la conception et du prototypage, où les ingénieurs peuvent explorer différentes configurations et identifier les problèmes potentiels avant même de commencer la production physique. De plus, les jumeaux numériques sont largement utilisés dans la fabrication prédictive. En utilisant des modèles avancés alimentés par des données en temps réel, ils peuvent prédire les performances futures des équipements industriels, anticiper les pannes et optimiser les paramètres de fonctionnement pour améliorer l'efficacité opérationnelle. Le projet JENII (Jumeaux d’Enseignement Numériques Immersifs et Interactifs), mené par l’ENSAM en collaboration avec le CESI, le CNAM et le CEA, et financé par l’ANR, vise à développer des jumeaux numériques pour la formation, particulièrement pour les procédés comme la forge et la fonderie, entre autres. JENII vise à révolutionner l’éducation, en créant une suite de jumeaux numériques pour des cas d’applications d’ingénierie. En intégrant des environnements virtuels avec des interactions en temps réel, ce projet offre aux apprenants une expérience d'apprentissage innovante qui complète les méthodes traditionnelles d'enseignement.

Locating and removing landmines and other ERW (Explosive Remnants of War) is dangerous, hazardous, and time-consuming. It requires implementing multilevel on-site surveys: general non-technical surveys to mark the areas affected and technical surveys to determine the perimeter of related minefields. This paper introduces a landmine location-based prediction model, combining military experience with machine-learning techniques and spatiotemporal data, by introducing a new approach for area selection and adding military-based features for context modelling and model training. Besides predicting landmine’s location areas, this model classifies the affected regions by priority and difficulty of clearance, in such a way as to minimise the long time needed by surveys and reduce the danger related to that task, thus providing the clearance organisations with a good resource allocation for their operations. We applied several machine learning techniques that combine Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBOOST), taking into consideration the imbalanced data problem and tweaking for the best performance and accuracy. The experimental results show that the model has the potential to provide reliable predictions and valuable services for demining operations on the field.

This article presents a two-step method to enhance metal-forming predictions by integrating Virtual Reality (VR) into Digital Twin models, focusing on single-blow cold copper upsetting operations. The process begins with developing a real-time predictive surrogate model that considers actual process parameters, acting as a crucial link between conventional numerical simulations and immediate decision-making. Subsequently, the surrogate model is integrated into a realistic VR environment, aligned with the experimental forging setup. The study underscores the need and potential advantages of real-time digital twins in the forging field, emphasizing the bridging capability between numerical simulations and instant decision-making through predictive modeling and immersive virtual environments.

Biomass pyrolysis is a thermochemical process used for renewable products and energies. However, there are still issues that need to be addressed for process modeling and optimization. This study focuses on the relationship between heating rate, shrinkage, and products from flax fibers using thermogravimetric analysis (TGA), microscopic observation, and pyrolysis-gas chromatography/mass spectrometry (Py-GC/MS). TGA confirms sequential evaporation of water then decomposition of hemicellulose, cellulose, and lignin. Observations from the micro-reactor show that flax fibers undergo shrinkage within the temperature range of 335 to 370 °C, depending on the heating rate. Pyrolysis products were analyzed using Py-GC/MS at four different final temperatures from 350 to 500 °C, revealing the presence of anhydrosugars, furans, ketones, phenols, esters, alcohols, aldehydes, and acids. The results indicate a correlation between temperature and the increase in furans and ketones. The analysis suggests that furans and ketones are associated with shrinkage.

This work presents an innovative non destructive control process in order to guide for the end of life (re-use or recycling) of permanents magnets (PM) rotors of electric motors. The process is based on the measurement of the external field produced by the PM rotors. A Finite Element model of the rotor and its environment has been used to simulate a process of classification of the geometry of PM rotors, crucial information for the disassembling of the PM. Firstly, using a finite element model, we were able to investigate the field distribution outside the rotor of the magnetic flux density for different PM rotors designs. From this information, we can set up a classification methodology, based on the Central Voronoi Tessellation (CVT) method, which can help to identify how the PM are inserted within the rotor. This information can be very useful to choose PM disassembling process that should be applied.

L'augmentation des produits en fin de vie et des déchets de produits de consommation pose des défis majeurs en termes de gestion des ressources, notamment des matériaux critiques et des composants stratégiques. Dans le contexte de la transition électrique, ces enjeux sont exacerbés, mettant en lumière la nécessité d'une gestion efficace des ressources et d'une économie circulaire. Cette proposition avance une vision du développement de solutions adaptées de l’industrie 4.0, dans le but de promouvoir une Industrie 4.0 Circulaire. L’objectif est de présenter les technologies et leurs enjeux dans le cadre d’une Économie Circulaire 4.0, afin de favoriser la réparation ou la réutilisation de parties (modules ou composants) encore fonctionnelles à l’étape de fin de vie des produits (souvent en fin d’usage), dans le but de préserver les ressources abiotiques, de réduire les impacts environnementaux et, de favoriser le basculement des activités économiques vers des filières régénératives locales.

We present an adaptive physics-informed deep homogenization neural network (DHN) approach to formulate a full-field micromechanics model for elastic and thermoelastic periodic arrays with different microstructures. The unit cell solution is approximated by fully connected multilayers via minimizing a loss function formulated in terms of the sum of residuals from the stress equilibrium and heat conduction partial differential equations (PDEs), together with interfacial traction-free or adiabatic boundary conditions. In comparison, periodicity boundary conditions are directly satisfied by introducing a network layer with sinusoidal functions. Fully trainable weights are applied on all collocation points, which are simultaneously trained alongside the network weights. Hence, the network automatically assigns higher weights to the collocation points in the vicinity of the interface (particularly challenging regions of the unit cell solution) in the loss function. This compels the neural networks to enhance their performance at these specific points. The accuracy of adaptive DHN is verified against the finite element and the elasticity solution respectively for elliptical and circular cylindrical pores/fibers. The advantage of the adaptive DHN over the original DHN technique is justified by considering locally irregular porous architecture where pore–pore interaction makes training the network particularly slow and hard to optimize.