Theses and Dissertations from UMD
Permanent URI for this communityhttp://hdl.handle.net/1903/2
New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a give thesis/dissertation in DRUM
More information is available at Theses and Dissertations at University of Maryland Libraries.
Browse
2 results
Search Results
Item Dynamic Bayesian Network Data Updating Approaches for Enabling Causal Prognostics and Health Management of Complex Engineering Systems(2022) Lewis, Austin Drisko; Groth, Katrina M; Reliability Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Complex engineering systems (CESes), such as nuclear power plants or manufacturing plants, are critical to a wide range of industries and utilities; as such, it is important to be able to monitor their system health and make informed decisions on maintenance and risk management practices. However, currently available system-level monitoring approaches either ignore complex dependencies in their probabilistic risk assessments (PRA) or are prognostics and health management (PHM) techniques intended for simpler systems. The gap in CES health management needs to be closed through the development of techniques and models built from a systematic integration of PHM and PRA (SIPPRA) approach that considers a system's causal factors and operational context when generating health assessments. The following dissertation describes a concentrated study that addresses one of the challenges facing SIPPRA: how to appropriately discretize a CES's operational timeline derived from multiple data streams to create discrete time-series data for use as model inputs over meaningful time periods. This research studies how different time scales and discretization approaches impact the performance of dynamic Bayesian Networks (DBNs), models that are increasingly used for causal-based inferences and system-level assessments, specifically built for SIPPRA health management. The impact of this research offers new insight into how to construct such DBNs to better support system-level health management for CESes.Item A DATA-INFORMED MODEL OF PERFORMANCE SHAPING FACTORS FOR USE IN HUMAN RELIABILITY ANALYSIS(2009) Groth, Katrina M.; Mosleh, Ali; Mechanical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Many Human Reliability Analysis (HRA) models use Performance Shaping Factors (PSFs) to incorporate human elements into system safety analysis and to calculate the Human Error Probability (HEP). Current HRA methods rely on different sets of PSFs that range from a few to over 50 PSFs, with varying degrees of interdependency among the PSFs. This interdependency is observed in almost every set of PSFs, yet few HRA methods offer a way to account for dependency among PSFs. The methods that do address interdependencies generally do so by varying different multipliers in linear or log-linear formulas. These relationships could be more accurately represented in a causal model of PSF interdependencies. This dissertation introduces a methodology to produce a Bayesian Belief Network (BBN) of interactions among PSFs. The dissertation also presents a set of fundamental guidelines for the creation of a PSF set, a hierarchy of PSFs developed specifically for causal modeling, and a set of models developed using currently available data. The models, methodology, and PSF set were developed using nuclear power plant data available from two sources: information collected by the University of Maryland for the Information-Decision-Action model [1] and data from the Human Events Repository and Analysis (HERA) database [2] , currently under development by the United States Nuclear Regulatory Commission. Creation of the methodology, the PSF hierarchy, and the models was an iterative process that incorporated information from available data, current HRA methods, and expert workshops. The fundamental guidelines are the result of insights gathered during the process of developing the methodology; these guidelines were applied to the final PSF hierarchy. The PSF hierarchy reduces overlap among the PSFs so that patterns of dependency observed in the data can be attribute to PSF interdependencies instead of overlapping definitions. It includes multiple levels of generic PSFs that can be expanded or collapsed for different applications. The model development methodology employs correlation and factor analysis to systematically collapse the PSF hierarchy and form the model structure. Factor analysis is also used to identify Error Contexts (ECs) – specific PSF combinations that together produce an increased probability of human error (versus the net effect of the PSFs acting alone). Three models were created to demonstrate how the methodology can be used provide different types of data-informed insights. By employing Bayes' Theorem, the resulting model can be used to replace linear calculations for HEPs used in Probabilistic Risk Assessment. When additional data becomes available, the methodology can be used to produce updated causal models to further refine HEP values.