観測期間が異なるフィールド寿命データの解析法

抄録

In field studies of industrial products, data on usage times are obtainable only from repair requests made by owners during some specified warranty period. Usage times for products which do not fail during the warranty period are generally not recorded. This makes estimation of the lifetime distribution function F(x) in usage time very difficult for products for which usage times differ from calendar times (e.g., automobiles, copy machines, etc.). The author proposed some methods which deal with the data when observational periods measured by calendar time are the same for all products. However, in real situations, observational periods are sometimes different for each product because of the variety of product usage starting points. In such cases, Saito's method^<5)> is currently being used in many companies. In this article the author proposes a consistent estimator of F(x) and compares it with Saito's method. Situations where the proposed method becomes better than Saito's method are presented .Some of them are when the percentage of products which have failed during an observational period is small and when the number of products which have been sold is large.