Carbon Dioxide Concentration Estimation in Nonuniform Temperature Fields Based on Single-Pass Tunable Diode Laser...

Area of Science:

• Spectroscopy
• Machine Learning
• Chemical Sensing

Background:

• Traditional tunable diode laser absorption spectroscopy (TDLAS) for CO2 concentration measurement relies on ratios of spectral lines with different temperature dependencies.
• Temperature gradients in flow fields can cause significant deviations in CO2 concentration measurements, and analytical compensation is challenging.

Purpose of the Study:

• To develop a novel technique for accurate CO2 concentration prediction in the presence of temperature gradients using a single laser absorption spectrum and machine learning.
• To address the limitations of traditional TDLAS methods in non-uniform temperature environments.

Main Methods:

• Utilizing the entire absorption feature, considering variations in shape and peak intensities with temperature and concentration.
• Developing a data-driven predictive model trained on simulated data (digital twin concept) to identify and compensate for temperature field effects.
• Validating the model's performance using experimental data.

Main Results:

• The developed machine learning model accurately predicts CO2 concentrations in flow fields with temperature gradients.
• The model demonstrated superior predictive performance compared to the conventional two-line method in all experimental test cases.
• Gradient-weighted regression activation mapping confirmed the model's use of both peak intensities and spectral shape changes for prediction.

Conclusions:

• The proposed machine learning technique offers a robust and accurate solution for CO2 concentration measurement under challenging temperature gradient conditions.
• This approach overcomes the limitations of traditional TDLAS by leveraging spectral feature analysis and data-driven modeling.