Accurate depth-resolved temperature profiling via thermal-radiation spectroscopy: numerical methods vs machine...

Area of Science:

• Materials Science
• Spectroscopy
• Computational Physics

Background:

• Accurate temperature measurement is crucial for understanding material properties and behavior.
• Depth-resolved temperature profiling provides critical insights into thermal transport and phase transitions within materials.
• Traditional methods for thermal analysis often lack the precision required for complex material structures.

Purpose of the Study:

• To present and compare three distinct computational approaches for retrieving depth-resolved temperature distributions from thermal radiation spectra.
• To validate these methods using both synthetic and experimental thermal-radiation data.
• To identify the most effective technique for accurate temperature profiling in diverse material systems.

Main Methods:

• Nonlinear equation solver (commercial software)
• Custom-built nonlinear equation solver
• Deep Neural Network (DNN) model
• Validation using synthetic datasets with varied temperature profiles and noise levels
• Assessment using experimental spectra from a heated fused-silica window

Main Results:

• The deep neural network (DNN) model demonstrated superior performance compared to conventional numerical techniques.
• The DNN approach consistently provided accurate depth-resolved temperature profiles across synthetic and experimental datasets.
• Both custom and commercial nonlinear solvers showed limitations in accuracy and robustness.

Conclusions:

• Deep neural network (DNN) models offer a robust and highly accurate solution for depth-resolved temperature profiling from thermal radiation spectra.
• The DNN-based method surpasses traditional numerical solvers in performance for materials analysis.
• This work advances the capability for precise thermal characterization in materials science and engineering.