Machine Learning as a Method for Retrieving Pressure Values by Analyzing Spectral Line Parameters: The Hydrochloric Acid Case

This study proposes a noninvasive machine learning approach to infer pressure by analyzing the infrared spectral lines of the HCl molecule. High-resolution spectra were simulated using the HITRAN database across various pressures (15-900 mbar), temperatures (273-373 K), and optical paths (1-10.5 cm). Voigt profile parameters (amplitude, center, height, and Gaussian/Lorentzian widths) were extracted from these spectral lines and used to train six ML models. The ExtraTrees algorithm demonstrated superior performance, achieving an RMSE of 23.95 mbar on synthetic data. Validation with experimental spectra (78-790 mbar, 293 K) revealed strong agreement at lower pressures, with errors below 5% (e.g., 2.62% at 78 mbar). The hybrid methodology, which combines simulated training with experimental validation, circumvents the need for direct sensor exposure to corrosive environments and offers a reliable alternative for pressure retrieval.