Neural network for baseline-free combustion diagnosis based on TDLAS

High-pressure absorption spectroscopy plays a pivotal role in measuring temperature and pressure distributions within industrial reaction processes and engine combustion. However, achieving precise pressure measurements in high-temperature and high-pressure environments through absorption spectroscopy remains a challenge in scientific investigations. This is mainly due to the significant uncertainty of the spectroscopic parameters in existing databases (such as HITRAN and BT2) under high-temperature and high-pressure environments. Additionally, baselines are unstable in high-pressure environments, making reliable measurements difficult. In this paper, a novel neural network model is proposed to address these challenges, called the Spectral Inversion Neural Network (SINN). SINN offers the unique capability of concurrently retrieving temperature, pressure, and H₂O concentration in high-temperature and high-pressure environments. Unlike conventional methods, SINN enables end-to-end inversion of raw spectra without necessitating baseline correction, incorporating considerations for temperature, pressure, and concentration. The model accommodates adjustments based on experimental spectra to mitigate inaccuracies in spectral parameters. With minimal pre-processing requirements for input data, rapid inversion capabilities, and heightened averaging accuracy, SINN emerges as a viable approach for real-time measurements in harsh environments.