A novel combination of experimental design and artificial neural networks as an analytical tool for improving performance in thermospray flame furnace atomic absorption spectrometry

Abstract

In this work, we present the combined effect of artificial neural networks (ANN) and experimental design as a suitable analytical tool for improving the performance of thermospray flame furnace atomic absorption spectrometry (TS-FFAAS) using Mg as leading case.To this end, mixtures of different amounts of methanol, ethanol, and i-propanol in water were assayed as carriers at different flow rates and different flame stoichiometries (air/acetylene ratios). Different levels of these variables determined the experimental domain, consisting in a cube which was divided into eight identical cubical regions that allowed increase in the number of available experimental points. A Box–Behnken design (BBD) was employed in each one of the regions. The name Multiple Box–Behnken design (MBBD) was given to this new approach. Then, the features of ANN were exploited to find the optimum conditions for conducting Mg determination by TS-FFAAS.The prediction capability of ANN was examined and compared to the least-squares (LS) fitting when applied to the response surface method (RSM).The suitability of the new approach and the implications on TS-FFAAS analytical performance are discussed.