Generating Input Data for Microstructure Modelling: A Deep Learning Approach Using Generative Adversarial Networks

Abstract

For the generation of representative volume elements a statistical description of the relevant parameters is necessary. These parameters usually describe the geometric structure of a single grain. Commonly, parameters like area, aspect ratio, and slope of the grain axis relative to the rolling direction are applied. However, usually simple distribution functions like log normal or gamma distribution are used. Yet, these do not take the interdependencies between the microstructural parameters into account. To fully describe any metallic microstructure though, these interdependencies between the singular parameters need to be accounted for. To accomplish this representation, a machine learning approach was applied in this study. By implementing a Wasserstein generative adversarial network, the distribution, as well as the interdependencies could accurately be described. A validation scheme was applied to verify the excellent match between microstructure input data and synthetically generated output data.

DRAGen – A deep learning supported RVE generator framework for complex microstructure models

“…This section provides a short overview on the two very prominent RVE generators Dream.3D and Neper as well as a short review of the work of Pütz et al [26] to give a better understanding of the data types DRAGen v.1.0 is now able to process. Also, the algorithms used in the main body of DRAGen v.0.1 and introduced in Henrich et al [25] are briefly discussed.…”

“…If done so, the final data set will contain combinations of grain characteristics that never appear in the underlying experimental data. [26] …”

“…Motivated by this, a previous study by Henrich et al [25] presented a basic structure for microstructure modeling with complex features. Additionally, Pütz et al [26] published an algorithm that uses Artificial Neural Networks (ANNs) in order to reproduce the characteristics of any microstructural feature of interest. A further development of this method was published by Fehlemann et al [27] .…”

Identification of Martensite Bands in Dual‐Phase Steels: A Deep Learning Object Detection Approach Using Faster Region‐Based‐Convolutional Neural Network

“…In a previous study, the pronounced banding was found responsible for this scatter in the fracture strain. [ 6 ] Figure 1 a shows a panoramic SEM image of the microstructure with a height of 320 μm and a width of 1153 μm. The data used in this study was originally generated for another study, where the effect of local strains and stress triaxiality on the void nucleation and evolution of this DP800 was investigated.…”

“…[ 3 ] The specific microstructure has a big influence on basic parameters such as the toughness or the yield stress but even more on local properties on the microscale like the damage initiation and accumulation [ 4 ] or the fatigue properties. [ 5 ] For well‐known steels such as dual‐phase (DP) steels, the relations between microstructure and damage properties have already been studied in a vast number of publications, see, for example, Pütz et al [ 6 ] for investigations about the damage tolerance of DP800 and DP1000, Tasan et al [ 7 ] regarding simulative experimental studies about stress and strain partioning in DP800, or Heibel et al [ 8 ] for studies on the edge crack sensitivity of different DP and complex phase steels. Experimentally, however, it is almost impossible to quantify the specific effects of individual microstructure parameters (e.g., grain size or phase ratio), since the modification of a particular parameter normally also modifies some or all of the other parameters.…”