Development and application of covariate based reliability models: utilizing constrained maximum likelihood optimization

In this research, a difficult yet practical problem of modeling failures as functions of stress profiles was addressed. Failures, both system and component based, can in many cases be explained in terms of the stresses experienced. These stresses are crucial in understanding the reliability of the component or system. If the underlying stresses can be determined, it becomes possible to create reliability models that incorporate them. In many cases reliability models can be made independent of stresses or in terms of a single stress. In this scenario the process of building the respective reliability model is not complex. These simple scenarios that have a limited number of stresses do not necessarily demand a rigid algorithm. When creating a reliability model for a single, or perhaps a system that experiences two stresses, trial-and-error is sufficient. Problems arise however when a system undergoes an excessive number of different stresses. These stresses all impact the system differently, and thus they must be modeled accordingly. Such a trial-and-error method would not be practical or appropriate. The primary goal of this research is to develop algorithms that can systemically approach these situations. In the case of the research, the system under study experiences a variable load profile. This algorithm aims to create an approach that can accurately capture the system reliability, while factoring in the system stresses. US Navy NAVAIR, in recent time, has had an increased interest in studying system and component reliability. This is in part due to the large amount of resources that corrective actions and preventative maintenance require. These failures however, are based on the current system stress profile. Currently, the Navy has plans for a changing stress profile, as there will be a change in air wing composition. This changing stress profile is predicted to negatively impact system reliability. It is possible however, to create predictive models using the current and past failure data. This modeling approach utilizes a two parameter Weibull distribution to account for a changing stress profile. Inputs into this model are the anticipated composition of the naval air wing.M.S.Includes bibliographical referencesby Robert Kosak