Open source software for numerical computation

Skip to main content

Use of Scilab & Xcos for Design of Experiments and Optimizations using chained proprietary simulation components

By Stéphane JIMENEZ, ArcelorMittal

ArcelorMittal R&D has elected Scilab as the core engine to develop a generic simulation platform for the experimentation, the fitting and the optimization of in house developed numerical models.

The full solution, which includes a business repository that will host the model components and act as an ATOMS server for the simulation platform, relies on two main paradigms:

  • Manage and share knowledge inside R&D segment through executable model components (not source codes) including the quantity of heterogeneous legacy models.
  • Provide the researchers with an application allowing to graphically setup simulation workflow of chained models, and the associated design of experimentation, without necessarily having to write any line of code.

The power and wealth of Scilab simulation engine has no more to be demonstrated, as well as its ability to easily link libraries developed in foreign code like C++, C, FORTRAN, Java, Python etc.

This was enough to retain Scilab as the favorite candidate for our simulation platform, and further to seriously consider it as an open source replacement for Matlab and iSight. Then, Xcos has represented an ideal basis to rapidly develop the simulation workflow graphical editor.

We are currently prototyping the solution through main business cases.
We have developed our Xcos toolbox and blocks allowing to graphically manipulate the identified legacy FORTRAN models in the palette. We have developed dedicated components as Scilab macro to implement the most used algorithms in design of experimentation (typically sensibility analysis, model fitting or optimization).
Next step will be to consolidate or rewrite in C the current developments and to integrate all into a packaged application.