| Photoacoustic spectroscopy (PAS) is an important branch of modern laser spectroscopy with high sensitivity, good selectivity, and fast response. Spectral signal analysis of multi-component gas has been a challenging problem in the spectral analysis field, and is also an important part of the PAS for trace gas detection and the resulting data processing. At room temperature and pressure and in the infrared absorption band, the intensive overlaps of measured gas absorption lines due to pressure broadening lead to the resolution of the measured PAS signal as a puzzle of the spectral data processing. The PAS absorption of small molecular gas at its feature wavelength is a relatively sharp rovibrational transition line, thus the probability density function (PDF) of the spectral data takes on a super-Gaussian distribution. The distribution characteristics of this observation sample and the interaction behavior of feature components hidden in the observed sample data make an extraction of multi-component photoacoustic (PA) feature absorption signals by using independent component analysis (ICA) technology based on blind source separation (BSS) become possible. In this work, in order to meet multi-component gas PAS detection, ICA technique is used for the separation of multi-component in gas PAS, so its main contents include the following aspects:In connection with the characteristics of multi-component gas PAS, an analytical and processing program based on ICA technique is proposed for the collected PAS data, the ICA basic principles, the ICA application methods, as well as the PDF distribution characteristic of PAS data are studied, which provides a theoretical basis for ICA technology being applied to PAS measurement to accurately identify and distinguish the trace-gas content in the multi-component gas mixture.By an analysis of the characteristics of multi-component gas PAS profile distribution and spectral line data in the near-infrared bands, an idea of using BSS model based on an overcomplete-ICA-basis sample to separate overlapping feature components is given. Through a numerical analysis to the measured sample data, an BSS model with the overcomplete-ICA-basis sample is built up, a weight truncation constraint equation is derived, and a five-point sampling transformation of ICA basis sample based on the constraint equation is proposed. On this basis, by use of the FastICA algorithm based on the constraints of weight-truncation equation and negative-entropy approximation, the detection and analysis of ammonia in human breath gas as well as discharge feature gases in transformer oil under room temperature and pressure are made:in experiment of the breath gas detection and analysis, after the measured PAS data within0.1114nm interval near1522.4400nm selected, the concentrations of NH3, H2O and CO2were simultaneously analyzed; In the detection and analysis experiment of discharging feature gases in dissolved gas-in-oil, after the PAS data within0.15nm interval at around1530.3709nm sampled, the concentration analyses of CH4, H2O, C2H2and CO2at five different wavelengths were simultaneously made, and a separable and effective superposition distance of overlapping absorption peaks based on wavelength and with hidden in the PAS was determined at between0.0011nm and0.0174nm; Through a purification of C2H2calibration gas by using ICA to remove H2O background components hidden in the measured PAS data, the detection sensitivity limit of C2H2gas PAS based second harmonic modulation was determined at0.71ppbv.In this paper, the proposed ICA technology, against strong background noise, can extract the trace-gas PAS feature components, and its application, in such aspects as improving analysis accuracy and signal-to-noise ratio, finding potential feature components, as well as broadening the probing laser source with a fixed band to make a simultaneous detection of multiple gases in a small wavelength interval, can be further popularized. |
PDF Full Text Request