Expectation-maximization algorithm to develop a normal distribution NHPP reliability growth model

Abstract

One of the most important goals of any organization for a product that goes into the design or manufacturing process is to improve its reliability. Reliability growth is defined as improving a product’s criteria (input parameters) during operation using changes in design or production. This paper presents a hardware-centric approach for the reliability growth of systems following normal distribution based on the Non-Homogeneous Poisson Process (NHPP). To reach this goal, the reliability growth modeling equations for the NHPP with the assumption of the normal distribution for failure data are extracted and obtained. Then, the maximum likelihood estimation technique based on the Expectation-Maximization (EM) Algorithm has been used to estimate the reliability model parameters with the given failure data. Results show that the proposed approach is more accurate than the traditional reliability growth models. Finally, the developed model has been implemented in an application aerospace system with empirical failure data and simulates the process of growth or deterioration of the system with high precision.