Published: Jan 1997
| ||Format||Pages||Price|| |
|PDF (240K)||14||$25||  ADD TO CART|
|Complete Source PDF (7.9M)||14||$109||  ADD TO CART|
Stainless steels and other nickel-containing alloys are the principal materials used for key corrosion-resistant equipment in the chemical process industries. Since 1982, the National Institute of Standards and Technology has accumulated considerable data from thousands of field corrosion tests. Classification methods from a diversity of technical disciplines were applied to data documenting the corrosion behavior of three commonly used stainless steel alloys in industrial acetic acid environments. These data are characterized by statistical uncertainty; there is no completely accurate solution obtainable for the problem investigated. The classification methods examined include linear discriminants, nearest neighbor methods, polynomial networks, and decision trees. Extensive resampling techniques were used to estimate the true error rates to facilitate a comparative analysis. For this highly dimensional problem, the nearest neighbor method performed the best, given the feature selection determined by the decision trees and polynomial networks, which performed second and third best, respectively. The transparency of the decision tree approach, along with its feature selection abilities and commendable performance, make this classification method most attractive overall.
computer learning, corrosion, decision tree, empirical modeling, expert system, linear discriminant, nearest neighbor, polynomial network, stainless steel
Physical scientist, National Institute of Standards and Technology, Gaithersburg, MD
Professor, Katholieke Universiteit Leuven, Leuven,