A neural network approach to qualitative analysis in inductively coupled plasma-atomic emission spectroscopy (ICP-AES)

Abstract

The use of a simple neural network, the Bidirectional Associative Memory (BAM), is described for the qualitative and semiquantitative analysis in ICP-AES. A computer program was developed linking the BAM to the solution of an overdetermined linear equation system by a least squares fit (LSF) using 480 prominent lines of 68 chemical elements. The BAM carries out the qualitative analysis and in the following LSF, the concentrations of the previously found elements are estimated. First, the assessment of the algorithm was performed by analyzing only synthetic spectra without noise, resulting in a recall of 100% for all elements covered. The investigation of spectra affected with noise of various degrees yielded a recall of about 96–97% and an RSD of 3% for spectra of twenty elements at the 10% noise level. A first test of the algorithm with experimental spectra has also given satisfactory results.