Servizio Comunitario di Informazione in materia di Ricerca e Sviluppo - CORDIS

FP5

VESPISM Sintesi della relazione

Project ID: G5RD-CT-2000-00315
Finanziato nell'ambito di: FP5-GROWTH
Paese: Germany

Software package

Based on the so-called phase-field theory a state-of-the-art software package for simulation of microstuctural evolution in alloys have been developed. This software package, i.e. MICRESS, can now simulate in 3D, phase transformations in alloys including e.g. solidification, recrystallization and grain growth.

The initial code existing at the start of the project has migrated to Fortran 90, an improved nucleation model has been incorporated, account has been made for limited interface mobility due to both solute drag and particle pinning, a recrystallisation model has been added and a model added to account for effects from stress and strain fields.

Furthermore, MICRESS has been linked to Thermo-Calc using the so-called TQ-Interface. The TQ-Interface is a product in itself and it has been integrated into other software outside this project. The interface establishes a standard for using thermodynamic data in application software. For linking the TQ-Interface to MICRESS several specific modifications and improvements had to be made. This link to Thermo-Calc provides the MICRESS user with easy access to assess and proper thermodynamic and kinetic information, to incorporate and base a particular simulation on.

A key factor for the success of the integration of real thermodynamics in the phase field software is the speed of calculation. A complex simulation in 3D may take from several hours to weeks even on a fast computer. Adding thermodynamic calculations inside this may make the simulation 100-1000 times slower. Even if the computer speed is expected to improve in the future the goal has been to make simulations possible also on standard computers. This means a technique had to be developed to improve the speed and for this purpose a neural network was adopted. Instead of calculating the thermodynamic properties in each grid-point and each time step the neural network is ¿trained¿ during an initial stage, and during the main part of the simulation the network will provide interpolated values with the possibility to improve these using the full thermodynamic description.

Informazioni correlate

Contatto

Georg J. SCHMITZ
Tel.: +49-241-8098014
Fax: +49-241-38578
E-mail