80050009

open access

Version of Record

ツウデンシキ トマク ヒョウカ ニオケル データ ノ モデル タンサク

Model Detection of Enamel Rater Value Data

Japanese (日本語)

Review of the Faculty of Maritime Sciences, Kobe University

04 <商船・理工論編>

89-100

神戸大学海事科学部

2007-07

2007-11-13

The inside of a metallic can filled with some drink is usually coated with a macro-molecular film to suppress the progress of corrosion. The degree of its coating is evaluated by so-called Enamel Rater Value (shortly, ERV) of a quick electrical stress test. ERV is nonnegative-valued, and its small value is desirable for coating. Our objective here is to examine distributional shapes of ERV data sets coming from typical manufacturing processes of aluminium cans under control. The data sets are left-censored and roughly classified into two groups according to specifications of coating. But, they have a notable common feature that each distribution of them is anti J-shaped with a very high peak near the origin and also an extremely spread tail. Four statistical distributions, including the generalized Gamma, have been adopted to fit the data sets to examine their performance. The best one is a new power normal family introduced by one of the authors (Isogai [31). The second one is the generalized Gamma, but it has a serious computational problem. Finally it is also shown that our new power normal family gives us clear information about distributional shapes of the ERV data..

紀要論文