[ Information ] [ Publications ] [Signal processing codes] [ Signal & Image Links ]
[ Main blog: A fortunate hive ] [ Blog: Information CLAde ] [ Personal links ]
[ SIVA Conferences ] [ Other conference links ] [ Journal rankings ]
[ Tutorial on 2D wavelets ] [ WITS: Where is the starlet? ]
If you cannot find anything more, look for something else (Bridget Fountain)
Si vous ne trouvez plus rien, cherchez autre chose (Brigitte Fontaine)
Google
 
Web www.laurent-duval.eu lcd.siva.free.fr

Updates

Internship position: Robustness study of learning models under data sampling: application to time series/signals prediction on process engineering data

Stage : Étude de la robustesse à l’échantillonnage de modèles d’apprentissage : application à la prédiction de séries temporelles sur des données procédés



Context

Process engineering deals with the application of physical chemistry to industry. Its main goal is to transform materials. It aims at designing and managing the functioning of a process with different chemical or physical transformations. This domain relies on the regular monitoring of different sensors, parameters, and physical quantities (temperature, pressure, concentration, etc.) to optimize operating points and product quality as a function of source materials. Being able to construct predictive or explicative statistical models, relating acquired or computed data to thought properties, is becoming an important stake for process efficiency. Approaches combining functional knowledge and data science methods have been developed recently, to improve model performance and robustness. Challenges are numerous: data availability, uncertainties, physical model validity, ability of off-line experimental models to work in on-line industrial conditions, etc. The proposed subject deals with the latter issue. Industrial-type data often only have available measurements acquired at a relatively low frequency rate (typically day or week), with little control of the exact timing. Meanwhile, pilot-unit experimental data may be obtained at higher rates (hour, minute) with precise time control.

Objectives

The candidate will be hosted by the Process Experiments Division and will work in close collaboration with the Digital Science and Technology Division. She/he will analyze data related to a given process, in continuity with another internship completed in 2021. She/he will study the impact of different sampling schemes on the reliability and robustness of already developed imputation and prediction models. The heart of the work will consist in developing a workflow allowing the reduction of the initial sampling rate of pilot-unit data, to better mimic that of industrial measurement. Doing so, we are interested in precisely quantifying the efficiency loss and the resulting uncertainties, and the necessary adaptation of the prediction models. We expect the possibly of providing “best practice/good sampling” recommendations, with respect to temporal dependencies of measures variables.