著者

吉野 睦

出版者

一般社団法人日本品質管理学会

雑誌

品質 (ISSN:03868230)

巻号頁・発行日

vol.44, no.3, pp.286-293, 2014-07

---

Recently, it is required to treat the big-data also in the manufacturing industry. They are laboratory data, inspection data, diagnosis data, and so on. But, unfortunately, they are not able to compute with conventional multivariate analysis. The causes are an increase of power of a test and the curse of dimensionality. Then, the new statistical method called "data-driven analysis" is applied as solution. On the other hand, conventional SQC method corresponds to "event-driven analysis". The difference between both is a difference of the conditional probability used as a criterion of judgment. Event-driven analysis uses P(O | T), and data-driven analysis uses P(T | O), where T is theory and O is observation. In in-house statistics education, the educational content that makes this difference clear was desired. This report reviews how this subject was tackled by several examples in DENSO.