Gaussian process regression for direct laser absorption spectroscopy in complex combustion environments.

Area of Science:

• Combustion diagnostics
• Spectroscopy
• Machine learning applications

Background:

• Tunable diode laser absorption spectroscopy (TDLAS) is a key technique in combustion research.
• Existing post-processing methods struggle with speed and accuracy for large, blended spectral lines.
• Accurate inference of thermodynamic properties from spectral data is crucial for combustion analysis.

Purpose of the Study:

• To evaluate Gaussian process regression (GPR) for processing direct absorption spectroscopy data in combustion.
• To infer gas properties directly from absorbance spectra using GPR.
• To compare GPR performance against traditional line profile fitting methods.

Main Methods:

• Trained parallelly-composed single-output GPR models and multi-output GPR models using the linear model of coregionalization (LMC).
• Utilized simulated spectral data based on a defined test matrix.
• Focused on inferring multiple thermodynamic properties simultaneously from absorbance spectra.

Main Results:

• GPR models demonstrated feasibility for accurate inference of multiple gas properties.
• The approach effectively handles wide spectral ranges with numerous blended spectral lines.
• Single-output GPR models showed sufficient accuracy and efficiency for temperature and concentration inference.

Conclusions:

• GPR is a viable and effective alternative to traditional line profile fitting for TDLAS data.
• The developed GPR models offer improved accuracy and speed for combustion diagnostics.
• Further validation is recommended, but GPR shows strong potential for real-time combustion analysis.