Combining multimodal fatigue fracture surface images for analysis with a CNN

Area of Science:

• Materials Science
• Mechanical Engineering
• Data Science

Background:

• Fatigue fracture analysis in Ti-6Al-4V is crucial for material performance.
• Traditional imaging methods like SEM provide topographical data.
• Integrating multiple data sources can enhance analytical capabilities.

Purpose of the Study:

• To investigate the utility of combining SEM, BSE, and SWLI imaging modalities for fatigue fracture surface analysis.
• To apply pre-trained convolutional neural networks (CNNs) for predicting crack growth parameters.
• To evaluate the impact of multi-modal data fusion on model performance.

Main Methods:

• Utilized Scanning Electron Microscopy (SEM), Backscattered Electron Imaging (BSE), and Scanning White Light Interference (SWLI) to capture fracture surface data.
• Employed pre-trained CNNs, leveraging color channel integration for multi-modal data fusion.
• Trained models to predict distance from load line and crack growth rate.

Main Results:

• Demonstrated that different imaging modalities contribute varying levels of importance to CNN models.
• Showed a 20% improvement in classification and a 60% improvement in regression results by combining modalities compared to SEM alone.
• Validated the effectiveness of multi-modal data integration for enhanced fracture analysis.

Conclusions:

• Combining SEM, BSE, and SWLI imaging data significantly enhances the predictive accuracy of CNNs for fatigue fracture analysis.
• The fusion of multi-modal imaging data offers a powerful approach for understanding material fatigue behavior.
• This methodology provides a pathway for more robust and accurate material characterization.