Abstract
Verifying the reliability of mass-produced products such as consumer electronics and home electrical appliances by collecting data on problems in the early stages of use in the market is as important as reliability evaluations at the development stage. The reliability data is used to decide the proper provisioning quantities of components for maintenance. However, for economic reasons, most of the data collected is restricted to the total sales and the total number of failures reported on a product for a given month. This is because the data is collected via a sales and service network that is widely distributed around the country. Therefore, for maintenance service, it becomes important to be able to make as accurate an estimate of the reliability of relevant components in the product as possible given the limited data that is available. This paper shows that the unknown parameters of the conditional probabilities of product failures in each sales lot can be non-parametrically estimated using the EM algorithm after creating a model from this limited data. In addition, the accuracy of this method is evaluated by comparison with the conventional method of estimating the conditional probabilities when the number of failures of the product in each sales lot in a certain Period is given.
