Neural Network Based Quantitative Prediction of Corrosion Phenomena in Cars and Their Anti-Corrosion Design

点击翻译按钮获取中文摘要

In this research, we have developed a method to quantitatively predict long-term corrosion occurring m cars, and additionally the method to perform anti-corrosion design. As the key technology, a multilayered neural network is employed to analyze a number of case studies of practical corrosion in cars. The process of the developed method is as follows.(1)At first, a number of case studies of corrosion of cars are given to a multilayered neural network, which is trained. Performance of the trained network in accuracy of learning and training and its network topology are investigated in detail.(2)Sensitivity studies are performed using the trained network. Here, various kinds of corrosion parameters are considered for the sensitivity study. Influential parameters for corrosion phenomena are then selected.(3)The case studies are again learned using the compacted neural network with the selected parameters. Its performances are again investigated in detail.(4)Using the trained compacted neural network, solutions for anticorrosion design of cars are effectively searched and visualized as a multidimensional design window.