Research on a spatiotemporal prediction method for two-dimensional temperature fields based on TDLAS array sensors...

Area of Science:

• Combustion diagnostics
• Optical spectroscopy
• Machine learning applications

Background:

• Traditional tunable diode laser absorption spectroscopy (TDLAS) offers limited spatial resolution for combustion fields.
• Existing combustion diagnostic tools face challenges with dynamic response delays.

Purpose of the Study:

• To develop a spatio-temporal predictive diagnostic method for 2D combustion fields.
• To improve the accuracy and speed of flame temperature measurements.
• To leverage deep learning for enhanced combustion diagnostics.

Main Methods:

• Utilized 2D array TDLAS direct imaging with a 64-pixel detector for parallel O2 absorption measurements.
• Developed a SwinLSTM deep learning model to capture spatial and temporal temperature field dynamics.
• Employed a "point-surface integration" strategy for multidimensional validation using thermocouples and thermal imagers.

Main Results:

• Achieved a maximum relative error of 3.75% in single-point quantitative inversion.
• Accurately reconstructed the 2D flame temperature field and its macroscopic structure.
• SwinLSTM-D model achieved high performance in prediction tasks (SSIM: 0.961, PSNR: 38.625 dB), outperforming other deep learning models.

Conclusions:

• The proposed method accurately reconstructs 2D flame temperature fields.
• The deep learning model effectively predicts the spatiotemporal evolution of flame temperatures.
• This research offers novel approaches for combustion measurement and diagnostic technologies.