The study of modeling methods of material corrosion data in soil environment based on PCA and ANN

FU dongmei ¹, LI Xiaogang^2,3,Fu Zhenze¹ Gao Jin^2,3, Lu lin^2,3

1、School of Information and Engineering, University of Science and Technology Beijing,Beijing 100083,China

2、Materials Science and Technology School, University of Science and Technology Beijing, Beijing 100083,China

3、Beijing Key Laboratory for Corrosion, Erosion and Surface Technology , Beijing 100083，China

Abstract

Material soil corrosion is a complicated phenomenon. Restricted by experiment costs and environmental conditions, soil corrosion data was quantitatively limited. Also, those data were affected by many factors, which made soil corrosion data to be a typical small-sample and high-dimensional data. In the past, curvefitting, statistic analyzing and pattern recognition model were often used to deal with this kind of corrosion data. However, these traditional methods have some defects in respect to small-sample and high-dimensional data, for example: curvefitting is not appropriate to high-dimensional data; statistic analyzing demands sufficient sample data; and pattern recognition model requires the known model and adequate data. Aiming at these limitations, a series of data analysis methods were developed in recent years.

In this study, a novel modeling method, referring to the small-sample and high-dimensional data, was postulated and applied to analyze the soil corrosion data of carbon steel. Firstly, with the application of ill-conditioned data filtration, correlation analysis (CA) and principal component analysis (PCA), the low-dimensional data was obtained, which preserved more than 93% information of the original data. By this means, the data dimension reduced from 13 to five. Further, the prediction model of carbon steel corrosion was established with ANN method, which is proved to have better prediction precision. Generally, the prediction model offered by this study is also feasible to predict the material corrosion in other environments.

Keywords: CA, PCA, ANN, Prediction model

* Contract Email: g.jin@163.com

**Acknowledgements

The authors are grateful for the financial support from the National R&D Infrastructure and Facility Development Program of China (registration number: 2005DKA10400). And from the National natural science fund project, No: 50499336-10