PC-SRGAN: Physically Consistent Super-Resolution Generative Adversarial Network for General Transient Simulations.

Area of Science:

• Scientific Machine Learning
• Image Super-Resolution
• Generative Adversarial Networks

Background:

• Machine learning, especially Generative Adversarial Networks (GANs), has transformed Super-Resolution (SR).
• Current SR methods often produce images lacking physical meaningfulness, limiting their scientific application.
• Interpretable scientific simulations require physically consistent high-resolution images.

Purpose of the Study:

• To develop a physically consistent Super-Resolution Generative Adversarial Network (PC-SRGAN) for scientific applications.
• To enhance image resolution while maintaining physical meaningfulness and interpretability.
• To improve the accuracy and efficiency of scientific machine learning models.

Main Methods:

• Development of PC-SRGAN, a novel approach integrating physical consistency into GAN-based SR.
• Incorporation of numerically justified time integrators and advanced quality metrics.
• Evaluation of PC-SRGAN performance against conventional SR methods, including with limited training data.

Main Results:

• PC-SRGAN significantly improves Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) compared to conventional SR methods.
• Achieves comparable performance to SRGAN using only 13% of the training data.
• Demonstrates physical consistency, making it suitable for time-dependent simulations.

Conclusions:

• PC-SRGAN advances scientific machine learning by providing physically meaningful super-resolution.
• The method enhances accuracy, efficiency, and process understanding in scientific research.
• PC-SRGAN offers a reliable and causal machine learning model for scientific domains, with code publicly available.