SAM-Based Approach for Automated Fabric Anisotropy Quantification in Concrete Aggregates

Area of Science:

• Materials Science
• Civil Engineering
• Computer Vision

Background:

• Accurate characterization of concrete aggregate fabric anisotropy is vital for predicting mechanical behavior and durability.
• Traditional segmentation methods struggle with noise and limited annotated data, hindering anisotropy assessment.
• Existing techniques lack robustness and automation for reliable concrete microstructure analysis.

Purpose of the Study:

• To develop an automated method for segmenting concrete aggregates using the Segment Anything Model (SAM).
• To propose a novel technique for quantifying concrete fabric anisotropy magnitude and orientation.
• To validate the proposed approach's accuracy and robustness on a custom concrete aggregate dataset.

Main Methods:

• Image preprocessing using Contrast Limited Adaptive Histogram Equalization (CLAHE) for enhanced aggregate visibility.
• Automated aggregate segmentation via the Segment Anything Model (SAM) with optimized grid point parameter (32).
• Fabric anisotropy quantification integrating computational geometry and second-order Fourier series analysis.

Main Results:

• The SAM achieved a high F1-score of 0.842 and IoU of 0.739 for aggregate segmentation.
• The proposed method demonstrated low mean absolute errors: 4.15° for orientation and 0.025 for anisotropy magnitude.
• Optimal SAM performance was confirmed with a grid point parameter set to 32.

Conclusions:

• The developed method offers a robust, accurate, and automated solution for quantifying concrete aggregate fabric anisotropy.
• This approach enhances microstructure analysis, aiding in predicting concrete performance and durability.
• The integration of SAM and computational geometry provides a powerful tool for materials science and civil engineering applications.