Super-Resolution Residual U-Net Model for the Reconstruction of Limited-Data Tunable Diode Laser Absorption...

Area of Science:

• Combustion diagnostics
• Laser-based measurement techniques
• Computational fluid dynamics

Background:

• High-resolution temperature distribution reconstruction is vital for precise combustion diagnosis.
• Tunable Diode Laser Absorption Tomography (TDLAT) is effective but limited by line-of-sight measurements and data acquisition.
• Insufficient data in extreme environments hinders TDLAT's diagnostic capabilities.

Purpose of the Study:

• To develop a software-based super-resolution reconstruction method for TDLAT.
• To enhance the resolution of temperature distribution reconstruction from limited TDLAT data.
• To improve the accuracy and efficiency of combustion diagnosis using TDLAT.

Main Methods:

• Proposed a super-resolution reconstruction method utilizing a super-resolution residual U-Net (SRResUNet).
• Employed residual networks and U-Net architectures to extract deep features from limited TDLAT data.
• Conducted simulation studies to optimize model parameters and assess performance.

Main Results:

• The SRResUNet model effectively improved reconstruction accuracy with super-resolution.
• Achieved good anti-noise performance in temperature distribution reconstruction.
• Demonstrated reconstruction errors of approximately 5.3%, 7.4%, and 9.7% for 2x, 4x, and 8x super-resolution, respectively.

Conclusions:

• The SRResUNet method successfully enhances TDLAT resolution for temperature distribution reconstruction.
• This deep learning approach offers a viable solution for overcoming data limitations in TDLAT.
• Indicates potential for other deep learning methods, like ESRGANs, in limited-data TDLAT applications.