%0 Journal Article
%A Fehlemann, Niklas
%A Aguilera, Ana Lia Suarez
%A Sandfeld, Stefan
%A Bexter, Felix
%A Neite, Maximilian
%A Lenz, David
%A Könemann, Markus
%A Münstermann, Sebastian
%T Identification of martensite bands in dual phase steels – a deep learning object detection approach using Faster R‐CNN
%J Steel research international
%V 94
%N 7
%@ 1611-3683
%C Weinheim
%I Wiley-VCH-Verl.
%M FZJ-2023-01191
%P 2200836
%D 2023
%X Martensite banding in dual phase steels is an important research topic in the field of materials design, since it is affecting the local damage properties of the material to a large extent. Therefore, it is necessary to quantify the amount of banding and the geometrical details of the bands in a specific microstructure, for example for simulative approaches. In this study, an object detection approach is used to separate martensite bands from the rest of the microstructure and a subsequent effort is made to transfer these results to statistical quantities for the generation of representative volume elements (RVE). For this, a convolutional neural network (Faster R-CNN) was trained on manually labeled SEM-images of DP800 steel. As exact geometric definitions of martensite bands in such two-dimensional images are difficult, the influence of different band definitions was investigated. The result of the training shows generally good prediction accuracy but is strongly dependent on the chosen band definition and the underlying human bias from the labeling process. A statistical analysis using cross-validation additionally shows that reliable results can already be achieved with only small datasets of around 50 to 100 training images due to the used transfer learning approach. This is an important outcome as it eliminates the need to generate an enormously large dataset which can only be obtained from very time consuming microscopy work and manual labeling of the images.
%F PUB:(DE-HGF)16
%9 Journal Article
%U <Go to ISI:>//WOS:000943084800001
%R 10.1002/srin.202200836
%U https://juser.fz-juelich.de/record/999141