A hybrid calibration-free/artificial neural networks approach to the quantitative analysis of LIBS spectra

A 'hybrid' method is proposed for the quantitative analysis of materials by LIBS, combining the precision of the calibration-free LIBS (CF-LIBS) algorithm with the quickness of artificial neural networks. The method allows the precise determination of the samples' composition even in the presence of relatively large laser fluctuations and matrix effects. To show the strength and robustness of this approach, a number of synthetic LIBS spectra of Cu-Ni binary alloys with different composition were computer-simulated, in correspondence of different plasma temperatures, electron number densities and ablated mass. The CFLIBS/ANN approach here proposed demonstrated to be capable, after appropriate training, of 'learning' the basic physical relations between the experimentally measured line intensities and the plasma parameters. Because of that the composition of the sample can be correctly determined, as in CF-LIBS measurements, but in a much shorter time.