A New Nonparametric Growth Model

Abstract

This paper proposes a new nonparametric reliability growth model for the analysis of the failure rate of a system that is undergoing development test. The only restrictions on the actual, unknown failure distribution for each stage of testing is that it be continuous, have only one unknown parameter θ, and have an associated unimodal likelihood function. No assumptions regarding the parametric form of the failure rate of the development process are made, only that there is no decay in the reliability of the system during the design changes. The parameters are assumed to be ordered from one test stage to the next such that θ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> ⩾ θ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> ⩾ ... ⩾ θm. The new model performs very well based on relative error and mean square error. The model is generally superior to the popular AMSAA model, regardless of the actual underlying failure process. In addition, the results indicate a notable bias in the AMSAA model, early in the development process, regardless of the actual underlying failure process.