| The design and implementation of a framework for the automated analysis of samples by inductively coupled plasma atomic emission spectroscopy (ICP-AES) is presented. Various components of this framework have been explored and validated. After initial work was undertaken to determine a suitable pattern recognition technique for sample classification, approaches for detecting the presence of a problematic matrix, and for selecting analysis conditions and calibration methodologies were developed. For the small universe of 14 elements used, the system performed well when matrix effects were due to concomitants in the examined universe. The application of generalized regression neural networks (GRNN) to compensate for spectral interferences and matrix effects and for correcting long-term drift was also examined. In post-processing the analysis results, the size of the GRNN training sets were found to be too small to permit accurate correction of matrix effects though accurate drift correction was possible. |
