Multiphase Reconstruction of Heterogeneous Materials Using Machine Learning and Quality of Connection Function

Area of Science:

• Materials Science and Engineering
• Computational Materials Science
• Digital Image Processing

Background:

• Accurate 3D microstructure reconstruction is crucial for understanding material properties.
• Challenges exist in achieving precise phase volume and structure-property linkages, especially with limited data.
• Existing methods often struggle with complex microstructures and point cloud data integration.

Purpose of the Study:

• To develop an optimized method for high-quality 3D microstructure reconstruction.
• To improve phase representation accuracy and compatibility with point cloud data.
• To enable reliable structure-property linkages and finite element analysis.

Main Methods:

• Utilized convolutional occupancy networks and point cloud data from inner microstructure layers.
• Implemented a Quality of Connection Function (QCF) repetition loop for model weight optimization.
• Reconstructed 3D representations from 2D serial images of isotropic and anisotropic materials.

Main Results:

• Successfully reconstructed 3D microstructures with precise phase representation and volume accuracy.
• Demonstrated efficacy compared to screened Poisson surface reconstruction and local implicit grid methods.
• The optimized model minimized errors between statistical properties and the reconstructed model.

Conclusions:

• The developed method provides a robust solution for 3D microstructure reconstruction.
• The approach is suitable for various material types, including multi-phase and anisotropic structures.
• This advancement facilitates accurate structure-property analysis and finite element modeling.