Deep Transfer Learning for Ni-Based Superalloys Microstructure Recognition on γ' Phase.

Area of Science:

• Materials Science
• Metallurgy
• Computer Science

Background:

• Ni-based superalloys are critical for high-temperature applications in aviation and energy.
• Microstructure analysis of these alloys is essential for design and development but traditionally requires expert knowledge.
• Current deep learning methods for microstructure analysis struggle with generalization to new datasets.

Purpose of the Study:

• To develop a deep transfer learning method for automated microstructure recognition of the γ' phase in Ni-based superalloys.
• To address the generalization limitations of deep learning models in materials science.
• To reduce the reliance on extensive labeled datasets for microstructure analysis.

Main Methods:

• A deep transfer learning approach was employed for microstructure recognition.
• Two datasets (W-900 and W-1000) of Ni-based superalloys were prepared and annotated.
• The method's effectiveness was evaluated by transferring knowledge between the two datasets.

Main Results:

• The proposed method achieved state-of-the-art segmentation accuracy with as few as 3-5 labeled images.
• Fast convergence was observed during knowledge transfer between datasets.
• A user-friendly software tool was developed to aid materials experts.

Conclusions:

• Deep transfer learning offers an effective solution for rapid and accurate Ni-based superalloy microstructure analysis.
• The developed method significantly improves efficiency for materials experts.
• This facilitates the accelerated design of new Ni-based and multicomponent alloys.