|
STUDY & REALISATION
SOFRESUD offers its know-how in several domains:
Embedded Electronics / Mechanical Design
SOFRESUD proposes study and realisation services for hardware and software embedded systems:
Hardware skills: Computer-aided design (CAD) center (SolidWorks), allowing developing small mechanical systems / DSP workshop allowing developing micro cards
Software skills: Driver development near material / Computer (RS422 and 485) and networks (Ethernet, FDDI) / Algorithm / Real time process) / MMI
I/O card samples developed by SOFRESUD
External protection for low temperatures
AUDIO-Sonar box
The Technical-Operational Simulation
For its R&D studies, SOFRESUD relies on its multidisciplinary engineers team in order to validate theoretical performances by numerical simulations. The numerical simulation helps to define and validate:
before the beginning of the prototype development phase.
The technical-operational simulations of DOG and TAU products were made with MATLAB software.
DOG: Theoretical results of gap position of a slow target and a DOG simulator in several phases
TAU: Therotical impacts of a flurry of 15 shells fired by a small calibre gun controlled by the QPD
Data Mining / ATLAS Software
SOFRESUD has developed the statistical modeling engine, ATLAS, in order to solve all modeling forecasting problems.
Based on statistical regularisation theories, the engine implements a derivation of the principle of structural risk modelling as described by V. Vapnik. The main idea is to minimise not only the error made on learned data - as standard algorithms will do - but also to take into account the confidence interval which may be delimited with the help of Vapnik's equations. The algorithm uses a particular technique, owner of controlled injection of noise into the data, which is clear to the user, and allows him to generate templates whose first quality is resiliency. ATLAS templates are able to manage noisy data, related variables or missing values, so that they are particularly efficient to represent most real problems.
ATLAS key points: Some robust module generation Where usual technics content themselves with getting an excellent fit with learning data, ATLAS makes an optimal balance giving greater place to the minimization of the generalization called error which is made when non learning data are processed. Some linear phenomena modeling ATLAS builds polynomial models whose crossed terms take non linearities and interactions between explanatory variables of the studied phenomenon into account. Phenomenon analysis ability The resulting models are given under an explicit form. Thus they can be used immediately for the comprehension on what could be the influence on the studied phenomenon of each variable or combination of variables. Computing time performance ATLAS builds a model based on polynomial functions, with outstanding execution time, then enable to process several millions recordings and to create real-time applications.
ATLAS was developed under contract ANVAR and ATLAS is the subject of an international patent: "Controlled Capacity Modeling Tool".
SOFRESUD has developed a Client / Server application, ATLAS-Server, integrating ATLAS technology.
|
| Study & Realisation |
 |
