Microstructure segmentation with deep learning encoders pre-trained on a large microscopy dataset

Stuckner, Joshua; Harder, Bryan J.; Smith, Timothy M.

doi:10.1038/s41524-022-00878-5

Cited by 36 publications

(10 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Thus, depending on the respective application, a compromise must be found, especially since the creation of a training dataset is the most time-consuming part of the implementation. Furthermore, the model could be optimized by pre-training the encoder part of the UNet specifically on optical microstructural images (Stuckner et al, 2022). In this way the weights of the encoder can be optimized for extracting only the most relevant features from microscopic images, instead of using the ImageNet weights, as in this showcase.…”

Section: Discussionmentioning

confidence: 99%

Efficient reconstruction of prior austenite grains in steel from etched light optical micrographs using deep learning and annotations from correlative microscopy

Bachmann

Müller²,

Britz³

et al. 2022

Front. Mater.

View full text Add to dashboard Cite

The high-temperature austenite phase is the initial state of practically all technologically relevant hot forming and heat treatment operations in steel processing. The phenomena occurring in austenite, such as recrystallization or grain growth, can have a decisive influence on the subsequent properties of the material. After the hot forming or heat treatment process, however, the austenite transforms into other microstructural constituents and information on the prior austenite morphology are no longer directly accessible. There are established methods available for reconstructing former austenite grain boundaries via metallographic etching or electron backscatter diffraction (EBSD) which both exhibit shortcomings. While etching is often difficult to reproduce and strongly depend on the investigated steel’s alloying concept, EBSD acquisition and reconstruction is rather time-consuming. But in fact, though, light optical micrographs of steels contrasted with conventional Nital etchant also contain information about the former austenite grains. However, relevant features are not directly apparent or accessible with conventional segmentation approaches. This work presents a deep learning (DL) segmentation of prior austenite grains (PAG) from Nital etched light optical micrographs. The basis for successful segmentation is a correlative characterization from EBSD, light and scanning electron microscopy to specify the ground truth required for supervised learning. The DL model shows good and robust segmentation results. While the intersection over union of 70% does not fully reflect the model performance due to the inherent uncertainty in PAG estimation, a mean error of 6.1% in mean grain size derived from the segmentation clearly shows the high quality of the result.

show abstract

Section: Discussionmentioning

confidence: 99%

Efficient reconstruction of prior austenite grains in steel from etched light optical micrographs using deep learning and annotations from correlative microscopy

Bachmann

Müller²,

Britz³

et al. 2022

Front. Mater.

View full text Add to dashboard Cite

show abstract

“…A straightforward representation is pixel or voxel data for which convolutional neural networks (CNN) often provide appropriate inductive bias. Meanwhile, these models are well established in materials science, especially for tasks such as segmentation and prediction of damage and crystallographic phases [2][3][4][5][6] . However, for predicting microstructural fatigue damage sites, where high-dimensional data is necessary to capture the interactions between various influence factors, pixel/voxel-based approaches are often infeasible due to their pronounced computational demand.…”

Section: Introductionmentioning

confidence: 99%

Materials fatigue prediction using graph neural networks on microstructure representations

Thomas

Durmaz

Alam

et al. 2023

Preprint

View full text Add to dashboard Cite

The local prediction of fatigue damage within polycrystals in a high-cycle fatigue setting is a long-lasting and challenging task. It requires identifying grains tending to accumulate plastic deformation under cyclic loading. We address this task by transcribing ferritic steel microtexture and damage maps from experiments into a microstructure graph. Here, grains constitute graph nodes connected by edges whenever grains share a common boundary. Fatigue loading causes some grains to develop slip markings, which can evolve into microcracks and lead to failure. This data set enables applying graph neural network variants on the task of binary grain-wise damage classification. The objective is to identify suitable data representations and models with an appropriate inductive bias to learn the underlying damage formation causes. Here, graph convolutional networks yielded the best performance with a balanced accuracy of 0.71 and a F\textsubscript{1}-score of 0.34, outperforming phenomenological crystal plasticity (+68\%) and conventional machine learning (+17\%) models by large margins.

show abstract

Section: Introductionmentioning

confidence: 99%

Materials fatigue prediction using graph neural networks on microstructure representations

Thomas

Durmaz

Alam

et al. 2023

Preprint

View full text Add to dashboard Cite

The local prediction of fatigue damage within polycrystals in a high-cycle fatigue setting is a long-lasting and challenging task. It requires identifying grains tending to accumulate plastic deformation under cyclic loading. We address this task by transcribing ferritic steel microtexture and damage maps from experiments into a microstructure graph. Here, grains constitute graph nodes connected by edges whenever grains share a common boundary. Fatigue loading causes some grains to develop slip markings, which can evolve into microcracks and lead to failure. This data set enables applying graph neural network variants on the task of binary grain-wise damage classification. The objective is to identify suitable data representations and models with an appropriate inductive bias to learn the underlying damage formation causes. Here, graph convolutional networks yielded the best performance with a balanced accuracy of 0.71 and an F1-score of 0.34, outperforming phenomenological crystal plasticity (+68%) and conventional machine learning (+17%) models by large margins. Further, we present an interpretability analysis that highlights the grains along with features that are considered important by the graph model for the prediction of fatigue damage initiation, thus demonstrating the potential of such techniques to reveal underlying mechanisms and microstructural driving forces in critical grain ensembles.

show abstract

Microstructure segmentation with deep learning encoders pre-trained on a large microscopy dataset

Cited by 36 publications

References 48 publications

Efficient reconstruction of prior austenite grains in steel from etched light optical micrographs using deep learning and annotations from correlative microscopy

Efficient reconstruction of prior austenite grains in steel from etched light optical micrographs using deep learning and annotations from correlative microscopy

Materials fatigue prediction using graph neural networks on microstructure representations

Materials fatigue prediction using graph neural networks on microstructure representations

Contact Info

Product

Resources

About