The application of neural networks to anodic stripping voltammetry to improve trace metal analysis

This thesis describes a novel application of an artificial neural network and links together the two diverse disciplines of electroanalytical chemistry and information sciences. The artificial neural network is used to process data obtained from a Differential Pulse Anodic Stripping (DPAS) electroanalytical scan and produces as an output, predictions of lead concentration in samples where the concentration is less than 100 parts per billion. A comparative study of several post analysis processing techniques is presented, both traditional and neural. Through this it is demonstrated that by using a neural network, both the accuracy and the precision of the concentration predictions are increased by a factor of approximately two, over those obtained using a traditional, peak height calibration curve method. Statistical justification for these findings is provided Furthermore it is shown that, by post processing with a neural network, good quantitative predictions of heavy metal concentration may be made from instrument responses so poor that, if using tradition methods of calibration, the analytical scan would have had to be repeated. As part of the research the author has designed and built a complete computer controlled analytical instrument which provides output both to a graphical display and to the neural network. This instrument, which is fully described in the text, is operated via a mouse driven user interface written by the author.