Predicting Atmospheric Corrosion Rates of Coppers in Taiwan by Artificial Neural Networks
| Autor: | Ya-Ping Chiu, 邱雅苹 |
|---|---|
| Rok vydání: | 2014 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 102 Due to the effects of certain environmental factors, deterioration and destruction of metal materials and products have been observed as “atmospheric corrosion”. Since Taiwan belongs to typical sea-island climate and massive industrial air pollutants are emitted, the atmospheric environment becomes highly corrosive which has significant impacts on metal materials. Therefore, further in-depth exploration of atmospheric corrosion of metals is an important and necessary issue. In view of that the copper is widely used in various fields, this study employs the data of atmosphere corrosion investigation of metals in Taiwan collected by Harbor and Marine Technology Center from September, 2009 to September, 2012. Two categories, namely general industrial zones and coastal industrial zones, are investigated. Environmental factors, such as average relative humidity, deposition rate of chloride, deposition rate of sulfur dioxide, and so on are used as input variables of artificial neural network (ANN), associated with multiple linear regression (MLR) and principal component analysis (PCA), to predict atmospheric corrosion rates and corrosivity category of coppers. The performance of different models will also be compared, and the results may provide useful information for reference of design and maintenance of copper objects in Taiwan. The result shows that sulfur dioxide deposition is the most significant factor impacting copper corrosion rates in general industrial zones. However, for coastal industrial zones, both sulfur dioxide deposition and chloride deposition are significant factors. The results reveal that among the different models utilized in this study, the winter and annual corrosion rates predicted by ANN have the most accurate performance. However, the prediction of PCA-ANN has generally the worst performance. For the corrosion predictions of C5 and C5+ levels, all of the models have better performance for the winter and annual corrosions than other seasons. But for C3 and C4 levels, none of the models can obtain accurate corrosion predictions. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|