Journal of The Japanese Society for Quality Control

In this paper, we make comparisons between the eigenvector method and the logarithmic least squares method, which are used to analyze a paired comparisons data. Examining the relations between two methods, for examples, the closeness of two solutions and the measurement of consistency, we show that the superiority of the eigenvector method argued by Saaty does not necessarily hold. In conclusion, we recommend the logarithmic least squares metdod which is more informative than the eigenvector method.

View full abstract

In this paper, interaction patterns in L_<18> orthogonal array are classified as the confounding patterns between interaction and main effect ,and the degree of confounding is evaluated quantitatively for each pattern. An equivalence relationship is defined among the forms of confounding patterns to classify into several confounding patterns. To evaluate the degree of confounding quantitatively, the ratio of sums of squares between interaction and main effect are used. For instance, these results suggest that when there are only factors with three levels and the number of factors is six at most, these factors should be assigned from the third column to reduce the confoundings as least as possible.

View full abstract

Show abstractHide abstract

I out of M ; G system, which remains in a marking state if there are at least I components among M identical components available, is one of the systems which are used in order to attain the high reliability by segregating the failed component from the system and masking the failure of component. In this paper, we have studied the analysis of reliability data of I out of M ; G system taking the probability of succeeding in segregating the failed component and reconfigureing the system, p, into consideration .Maximum likelihood methods are applied to derive the estimates of both p and the parameters of life time distribution of the component and their properties are investigated mainly in the case of large sample. The results show that the maximum likelihood estimates are asymptotically normally distributed and in the case of exponential distribution, the estimate of θ, which is mean time between failures, is unbiased. In the last, the model is extended from the identical to the different system configuration, where I and M may be distinct, and the estimates of the parameters are also studied.

View full abstract

Download PDF (595K)

This paper discusses a scheme of automated software design review, utilizing a production system. The paper presents a way in which (real world) check list items can be re-written into a set of rules in a generic form. The rule is comprised of pre-conditional, conditional and conclusive parts. Any design specification, prepared by using such integrated software engineering environments as SEA/I, can be reviewed by rule-based reasoning mechanism. The review is undertaken to see whether the design conforms to a set of rule, and if violation found, diagnoses are given. The experiments show that the automated design review provides such advantages as stabilized quality control, especially being usefull for preventing quality degradation which occurs in an evolutional software development.

View full abstract

Register with J-STAGE for free!