Complex-Phase Steel Microstructure Segmentation Using UNet: Analysis across Different Magnifications and Steel Types

Area of Science:

• Materials Science
• Computational Materials Science
• Image Analysis

Background:

• Accurate phase fraction quantification is crucial for understanding material properties.
• Manual annotation methods are time-consuming and prone to errors.
• Complex microstructures in alloy steels pose segmentation challenges.

Purpose of the Study:

• To develop an automated segmentation technique for phase fraction quantification in high-tensile strength alloy steel.
• To address the limitations of manual analysis in complex microstructures.
• To evaluate the model's scalability and robustness.

Main Methods:

• Leveraged the UNet convolutional neural network architecture.
• Optimized UNet performance through hyper-parameter tuning and data augmentation.
• Utilized Electron Backscatter Diffraction (EBSD) imagery.
• Employed a combined loss function for textural and structural feature capture.

Main Results:

• Achieved high accuracy in phase segmentation, demonstrated by high dice scores.
• Successfully segmented complex microstructures in alloy steel.
• Validated model scalability across varying magnifications and steel types.
• Showcased the adaptability and robustness of the automated method.

Conclusions:

• The proposed automated segmentation technique offers a precise and efficient alternative to manual methods.
• The UNet-based approach is effective for complex microstructure analysis in alloy steels.
• The model demonstrates significant potential for broader application in materials science research and industry.