Real-time Classification of Aerosol Single Particles Chemistry Composition

To quickly process the laser desorption/ionization mass spectra which were generated by aerosol laser time-of-flight mass spectrometer (ALTOFMS) during its working and extract the valuable information, an adaptive resonance theory-based neural network, ART-2a algorithm, was successfully applied to the real-time classification of single particles of NaCl, the mixture of NaCl and CaCl(2) aerosol, melamine and the atmospheric aerosol. Experimental results showed that when vigilance factor was 0.1 and learning rate was 0.05, ART-2a algorithm could successfully reveal the aerosol particles categories. Besides the successful probability was near 100%, the centroid mass spectra for the single particle classes were obtained, which could represent the characteristic of single particle classes remarkably. The number of NaCl single particle classes as a function of vigilance factor was also discussed. The result showed that the number of NaCl particle classes was larger when vigilance factor reached 0.8; furthermore the classification was much more precise. The mass spectra acquisition and control software using ART-2a can basically meet the requirements of real-time classification of atmospheric aerosol single particles.