Artificial Neural Networks are being used increasingly in many fields of industry and commerce. Their ability to “learn by example” and to generalize this knowledge so as to give correct predictions of previously unseen data make extremely attractive. There is great potential for applying this technology to corrosion problems as in principle, artificial nueral networks can ‘learn’ the behaviour of materials in a range of corrosive environments and thereby predict behavior. ln practice data in the
Iiterature is derived from differing sources, it is often of poor quality and important parameters are sometimes omitted. Also, care has been exercised when applying artificial nueral networks to ensure that their operational range is bounded and that some measure of output quality is given to the user. This paper will discuss particular problems and limitations of applying neural networks to corrosion data and suggest ways in which these limitations can be overcome.
Keywords: neural networks, error bounds, factors of confidence, corrosion data