List institute develops algorithms and software tools for a great variety of data analysis and processing. These data can come from measuring equipment for biology, food industry, process control and others or from sensor networks placed in buildings, industrial equipment or vehicles and so on… These data are often difficult to interpret because of their important volume, complexity or variety. Our researchers make them exploitable through advanced raw signal processing methods and automated learning statistical models allowing developing results analysis or decision making support tools.
We collaborate with industrial companies from varied sectors like energy, health, transport, security…
Among our academic partners
CSTB (French scientific centre for building technics), INP Grenoble (GipsaLab et G2ELab, France), INSA Rouen (LITIS, France), INRIA, INRA, CEA Life science division (DSV, France)
- Ability to process complex or heterogeneous data such as mass spectrometry, sensor networks (temperatures, pressure, speed, electromagnetic signals, movement…), large-scale data volume management
- Generic know-how capitalised through algorithmic bricks and reusable software platforms
- Information extraction and structuring for operational use
Time patterns extraction and dictionary learning
A multi-sensor monitoring equipment data basis is being analysed in order to detect patterns that become repetitive independently from their temporal position or intensity such as a peak on a curve. These patterns are automatically inventoried in a 10 to 20 elements dictionary that will be submitted to an expert or automatically validated according to their relevance and integrated to the monitoring. This “parsimonious decomposition ” methodological approach uses only the observed data and is much faster than a physical model design.
Equipment monitoring or industrial process, biological data analysis
Massive data visualisation
Massive complex data visualisation can concern for example data collected every second for more than a year by several hundreds of temperature data, electricity and water consumption, etc. Visualising them is a huge concern and technical issue that’s why dimension reduction methods exist such as 2D or 3D non linear projection. We develop these methods that preserve or highlight the data multidimensional structure. They are used to design reliable and efficient interactive complex data mining tools.
Containers analysis by neutron interrogation, building energetic assessment analysis, industrial systems real use characterisation
- european project EURITRACK/ERITRACK
- european project E-Dash
- BrainVis project(Labex DigiCosme Paris-Saclay)
- S. Lespinats, M. Aupetit, “CheckViz: Sanity check and topological clues for linear and non-linear mappings”, Computer Graphics Forum30, 113-125 (2011)
- M. Aupetit, « Approches topologiques pour l'analyse exploratoire de données et l'aide à la décision », Thèse d’HDR, Univ. Paris-Sud, Orsay (2012)
- P. Blanchart, M. Depecker-Quechon, “A non-linear semantic-preserving projection approach to visualize multivariate periodical time-series”, IEEE Transactions on Neural Networks25 , 1053-1070 (2014)
e-learning on status or behavioural statistical models
Statistical models e-learning functions are integrated into equipment monitoring tools in order to follow their operations or status and to predict their ageing when we do not have a sharp physical modelling. These monitoring tools have been developed and validated to follow on-line charge status and electric vehicles’ batteries’ ageing in real use in collaboration with CEA LITEN institute.
Electric batteries, fuel cell, water distribution networks
- programme interne avec le CEA LITEN
- european project SW4EU – Smart Water for Europe
- A. Barré, F. Suard, M. Gérard, M. Montaru, D. Riu, “Statistical analysis to understanding and predicting battery degradations in real-life electric vehicle use”, Journal of Power Sources245, 846-856 (2014)
- A. Barré, F. Suard, M. Gérard, D. Riu, “Electric vehicles performance estimation through a patterns extraction and classification methodology”, Journal of Power Sources273, 670-679 (2015)
"Interpreting heterogeneous and massive data from analysers or distributed systems"
Data analysis and systems’ intelligence Chief of laboratory
+33 1 69 08 84 39