Energetic upscaling strategy for grain growth. II: Probabilistic macroscopic model identified by Bayesian techniques

Exploiting Machine Learning in Multiscale Modelling of Materials

“…The conventional approach of the mean-field homogenization is being replaced with the Bayesian inference through application of ANN as well as NARMAX approach for the determination of feature set of the input-output database. The Bayesian approach has already being increasingly applied to single-scale problems such as alloy design [84], grain growth [85], as well as property predictions [86][87][88]. However, the application of such a approach for multiscale materials modelling problems has been limited [32].…”

Parametric Study of Directed Energy Deposition Of Duplex Stainless Steels To Optimize Ferrite-Austenite Phase Ratio And Residual Stresses

“…(i) First we use a thermal analysis of DED [11,12] based on analytical solutions, which can deal with relatively complex geometry, and takes account of latent heat release during solidification. This simulation of the temperature field history has already been validated experimentally by infrared measurements using pyrometers [11] and an infrared camera [13], and was used to predict grain growth [14,15,16] during thermal cycling. (ii) In addition, we use a fast model of diffusion of alloying elements based on analytical solutions to take into account phase transitions in DSS [17], which has recently been developed and coupled to [11] to predict ferrite-austenite phase ratio in thin-walled structures, with satisfying comparison against experimental measurements performed by electron backscatter diffraction techniques (EBSD).…”

Very fast simulation of growth competition between columnar dendritic grains during melt pool solidification

“…To overcome this difficulty, this contribution aims at developing a very fast numerical approach including growth competition of columnar dendritic grains, and equiaxed grains nucleated from the melt. The proposed upscaling strategy to reduce computation time is similar to the one developed for grain growth during annealing in [18][19][20]. It relies on the development of an algorithm based on 1) solidification maps to determine the region where CET occurs [3], 2) the dendritic preferred growth direction, 3) the application of the Walton and Chalmers criterion [21] relying on the concept of favorably oriented grain (FOG) to determine which grain survives the competition [8,[22][23][24], and 4) a solidification front propagation rule along the thermal gradient direction.…”