Data Mining of General Corrosion of Most Commonly Used Alloys Using Kohonen Mapping

This study was developed to provide an optimum method for designing metal alloys. Information on general corrosion of the most commonly used alloys, namely nickel-, iron-, aluminum-, stainless steel-, copper- and carbon-based alloys, was obtained from the National Institute of Standards and Technology (NIST). Parameters such as pH, temperature, conductivity, and composition of the electrolyte used in each experiment (alloy in contact with an environment), along with alloy composition (UNS) and corrosion rates, were collected. The parameters in the collected data were the electrolyte characteristics, the alloy composition, and the general corrosion rates. The data consisted of over 4000 samples. The next task was to cluster the data by similarities of the parameters. A web-based, publicly available Kohonen mapping software was used to perform the clustering analysis; the two dimensional Kohonen map was chosen. The map is composed of a number of cells on a plane of two dimensions and each cell stores a sample prototype representing that cell. Kohonen maps have the ability to preserve the topological properties of the data; i.e. samples with similar “high corrosion rates” will cluster together in a given cell of the map, while samples with “low corrosion rates” will cluster in a different cell, far apart from the cell storing the “high corrosion rates” samples. Once the Kohonen map is trained to cluster the samples by their corrosion rates, each one of the parameters of the samples (representing the electrolyte characteristics, the alloy composition and the general corrosion rates) is drawn as a function of the (X,Y) map position of the cell where the samples were stored. In this paper, we present the results obtained with nickel alloy data for which 1369 samples (or independent experiments) were collected.