A research group from the Hefei Institutes of Physical Science (HFIPS) of the Chinese Academy of Sciences recently developed a neural network-based absorbance recovery method to improve the accuracy of single path tunable diode laser absorption spectroscopy (TDLAS) measurement.

Measurement of combustion flow field temperature and component concentration distribution based on tomographic absorption spectroscopy can provide more comprehensive data for the design, monitoring and diagnosis of advanced combustion systems.

It has the advantages of high speed, high sensitivity, and strong interference immunity. However, traditional single-path measurement error of TDLAS is relatively large, affected by the distortion of absorbance, thus limiting the quantification accuracy of tomographic absorption spectroscopy (TAS).

“The focus of the problem is to solve the baseline errors that distort absorbance measurements,” said Prof. LIU Kun, one of the leading researchers of the study from HFIPS.

They found that the derivative of absorbance is more sensitive to the line-shape curvature, and the variation of curvature caused by baseline error is relatively small near the absorption peak.

Based on this, the HFIPS team, led by Prof. GAO Xiaoming and Prof. LIU Kun, designed a model to retrieve absolute absorbance profile from the derivative of the measured absorbance.

They tested this method through simulations and single-path temperature measurements, and applied it to measure the exhaust temperature and water concentration in a small diesel turbojet engine.

Results showed an error of only 0.9%, compared to thermocouple readings.

“Our results provide a valuable method for improving the accuracy of TDLAS measurements and can be easily incorporated into tomographic absorption spectroscopy,” said Prof. GAO Xiaoming.

The results of their study were published in the latest issue of FUEL.

Schematic of tomographic absorption spectroscopy measurement system. (Image by WANG Ruifeng)

Temporal variations of temperature and H₂O concentration distributions. (Image by WANG Ruifeng)