Neural-network-based multi-spectral thermometry and emissivity reconstruction in cavity high-temperature environments

Area of Science:

• Physics
• Materials Science
• Computer Science

Background:

• Accurate radiation thermometry is crucial for high-temperature processes.
• Challenges include unknown emissivity and complex multi-reflection effects in cavities.
• Conventional methods struggle with low-emissivity materials.

Purpose of the Study:

• To develop a neural-network-assisted framework for accurate radiation thermometry.
• To overcome limitations of unknown emissivity and multi-reflection effects.
• To enable reliable thermometry in challenging materials like alloys and ceramics.

Main Methods:

• Combines Monte Carlo ray-tracing with deep learning.
• Utilizes physics-informed training data with diffuse/specular reflections.
• Employs alternating neural networks for decoupled temperature and emissivity prediction.
• Incorporates full multi-reflection modeling.

Main Results:

• Achieved 0.7% temperature error (9 K) and 0.05-0.1 emissivity error in the 2-16 µm spectral range.
• Outperformed first-order methods by 5%-27% in emissivity reconstruction.
• Maintained <1% error with only 10 spectral channels and tolerated 1% intensity noise.

Conclusions:

• The developed framework provides accurate radiation thermometry in high-temperature cavities.
• It effectively addresses challenges from unknown emissivity and multi-reflection effects.
• Enables reliable thermometry for low-emissivity materials where traditional methods fail.