Photoelastic Stress Field Recovery Using Deep Convolutional Neural Network.

Area of Science:

• Computational Imaging
• Materials Science
• Data Science

Background:

• Photoelastic fringe patterns are crucial for stress field analysis in loaded objects.
• Traditional methods struggle with complex fringe distributions due to specimen geometry and experimental setups.
• Deep convolutional neural networks (CNNs) show promise in solving inverse problems within computational imaging.

Purpose of the Study:

• To develop a deep convolutional neural network (CNN) capable of accurately decoding stress distribution from complex photoelastic fringe images.
• To address limitations of traditional stress analysis methods when dealing with intricate fringe patterns.
• To validate the proposed CNN model's performance across various experimental configurations.

Main Methods:

• An encoder-decoder based deep convolutional neural network architecture was designed.
• The network was trained and validated using a synthetic dataset of photoelastic fringe patterns.
• Performance was quantitatively evaluated using metrics such as Mean Squared Error (MSE), Structural Similarity Index Measure (SSIM), and Peak Signal-to-Noise Ratio (PSNR).

Main Results:

• The proposed CNN accurately recovers stress distribution from complex photoelastic fringe images.
• The network demonstrates robustness across different experimental configurations.
• The stress recovery network achieved an average Structural Similarity Index Measure (SSIM) of over 0.99 on the synthetic dataset.

Conclusions:

• The developed deep convolutional neural network effectively solves the inverse problem of stress field recovery from photoelastic fringe patterns.
• This approach offers a significant advancement for stress analysis in computational imaging, particularly for complex scenarios.
• The high performance metrics indicate the method's reliability and potential for practical applications.