Advanced stacked modeling techniques for material porosity estimation via high-resolution computed tomography...

Area of Science:

• Materials Science
• Civil Engineering
• Computer Vision

Background:

• Traditional concrete porosity measurement is manual, time-consuming, and lacks robustness in low-resolution or noisy imaging.
• Accurate porosity assessment is critical for concrete material characterization and performance prediction.

Purpose of the Study:

• To develop an automated framework for concrete porosity estimation using deep convolutional neural networks (DeepCNN) and 2D CT scan images.
• To enhance the accuracy and efficiency of porosity measurement, especially under challenging imaging conditions.

Main Methods:

• A DeepCNN architecture with a multi-stage feature extractor and SPP-based neck was trained on augmented CT images.
• Image processing included automated ROI detection, normalization, and class-specific filtering.
• A rule-based adaptive thresholding (RBAT) strategy was employed for material classification and porosity estimation.

Main Results:

• The framework accurately estimated porosity across various concrete types (CM, GM, UHPC), with deviations within 1.3-1.5% compared to vacuum pycnometer measurements.
• The DeepCNN classifier achieved a precision-recall AUC of 1.0.
• The automated method demonstrated robustness under low resolution and noisy imaging conditions.

Conclusions:

• The proposed hybrid framework offers an accurate, automated, and computationally efficient solution for concrete porosity assessment.
• This approach is suitable for practical and industrial CT-based material characterization workflows.
• The study highlights the potential of DeepCNN for advanced non-destructive material analysis.