| Condition-Based Maintenance (CBM) and Prognostics and Health Management (PHM) technologies aim at improving the availability, reliability, maintainability, and safety of systems through the development of fault diagnostic and failure prognostic algorithms. In complex engineering systems, such as aircraft, power plants, etc., the prognostic activities have been limited to the component-level, primarily due to the complexity of large-scale engineering systems. However, the output of these prognostic algorithms can be practically useful for the system managers, operators, or maintenance personnel, only if it helps them in making decisions, which are based on system-level parameters. Therefore, there is an emerging need to build health assessment methodologies at the system-level.;Fault diagnosis/prognosis research at the system-level has been approached via data-driven and model-based methods. Data-driven methods do not provide any insight into the effects of system parameters on fault growth, thus offer little assistance in planning future operations or optimizing maintenance schedules. Model-based methods have been focused primarily on component-level prognostic models. Some researchers have borrowed concepts from the artificial intelligence domain and developed Model-Based Reasoning (MBR) methods but these are qualitative approaches and not useful for most of the practical application domains.;This thesis presents a methodology that is based on a hierarchical architecture consisting of three layers; system-level, subsystem-level, and component-level. Based on this hierarchy, four types of variables are defined, i.e., system-level variables, subsystem-level variables, load variables, and stress factors. These variables are interconnected with each other through four types of models, i.e., failure mechanism model, load-stress model, subsystem-level model, and system-level metamodel. At the lowest level (component-level), damage accumulation models that do not require anteceding diagnostic activities, are used to estimate the damage in critical components. At the highest level of the hierarchy, response surface metamodeling methods are used to build a system-level health assessment model. These methods originated from the statistical theory of design of experiments (DOE), and focus on planning experiments to reduce the number of total observations by locating data points in "meaningful" regions. Therefore, a large dimensional input space can be modeled using a small number of data points. The methodology is then tested, using a turbofan engine simulator developed by NASA. The Response Surface Model (RSM) results are compared with those of Artificial Neural Networks (ANNs), which are another type of metamodeling methods. For a given amount of data, RSM results are found to be almost as accurate as the ANN results. In contrast to ANN, which is a black box type of model, RSM is represented as a polynomial. The model representation in a polynomial form is particularly useful in this application since the coefficient values represent the relative effect of input variables on the system-level health, thereby, facilitating the decision making process at the system-level.;A brief overview of the main contributions of this thesis is given below. • Development of a hierarchical framework for using component-level prognostic models in assessing system-level health. • Development of a methodology for capturing and representing the coupling between subsystems' degradation. • Development of a methodology for modeling the effects of system's usage pattern on system-level health. • Application of these methodologies to a complex system for demonstration purposes. |
