ミズヤマ ハジメ   MIZUYAMA Hajime
  水山 元
   所属   青山学院大学  理工学部 経営システム工学科
   職種   教授
言語種別 日本語
発行・発表の年月 2005
形態種別 学術雑誌
査読 査読有り
標題 ロジスティック回帰援用型多段階品質情報推移モデルの提案 : 多段階品質情報推移モデルに基づく品質データマイニングの研究(第1報)
執筆形態 共同
掲載誌名 日本経営工学会論文誌
巻・号・頁 129-138頁
著者・共著者 淺田克暢, 山田賢太郎
概要 Detecting causal factors of chronic quality defects is very important, and often the most critical step of improving manufacturing quality in a multi-stage production system. Since advanced information technology has enriched manufacturing databases, it is now the time to ask how to utilize databases to streamline the process of this causal factor detection. However, applying conventional multivariate statistical analysis methods or modern data mining approaches simply to a database does not always provide sufficient knowledge for revealing the key factors of chronic defects and how they cause the defects. Thus, the authors propose a novel framework for more sophisticated exploratory quality data analysis in order to support detection of the causal factors. The proposed data analysis framework is named the "multi-stage quality information model (MSQIM)", which, if possible, should establish some hypotheses on the causal factors and/or the defect-causing mechanisms, and should at least identify which process steps within the production system further efforts of causal factor detection should be focused on. MSQIM first divides a manufacturing database into several segments, each of which corresponds to a certain process step within the production system. It then traces how the amount of information on the resultant manufacturing quality varies along with the process steps so as to identify the relevant process steps that require further focus. The varying pattern of the quality information is also studied in a qualitative way so that it assists in hypothesis generation. This paper mainly discusses how to implement MSQIM based on logistic regression. It also demonstrates how the proposed approach works through an industrial example.