About this Abstract |
Meeting |
MS&T23: Materials Science & Technology
|
Symposium
|
Materials Informatics for Images and Multi-dimensional Datasets
|
Presentation Title |
Phase Segmentation of Steel Microstructures via Semi Supervised Deep Learning |
Author(s) |
Nikhil Chaurasia, Shikhar Krishn Jha, Sandeep Sangal |
On-Site Speaker (Planned) |
Nikhil Chaurasia |
Abstract Scope |
An effective training approach for segmenting ferrite–pearlite microstructures using Convolutional Neural Network (CNN) UNET machine learning architecture (a semantic classifier) is presented in this work. CNN-driven image segmentation needs a large number of labelled training microstructures, which are typically unavailable. An innovative solution to the aforementioned issue is offered. First, polycrystalline templates are generated by running a 3D nucleation and growth model whose kinetics adhere to Avrami's. After that, clipped pictures of pearlite and ferrite (from actual microstructures) are randomly positioned on the individual grains of the polycrystalline templates, resulting in synthetic microstructures with varying fractions and scales of the two phase. Using a limited number of cropped images, a few thousand synthetic microstructures were produced. When evaluated on actual ferrite-pearlite microstructures, the accuracy of the UNET trained on the synthetic training set was around 98%, which demonstrates the strength of the above approach. |