| dc.description.abstract | Cyber-physical systems (CPSs), a term coined in 2006 refers to the integration of computation with physical processes. Particularly, modern critical infrastructures are examples of CPSs, like smart electric power grids, intelligent water distribution networks, and intelligent transportation systems. CPSs provide critical services that have great impact in nation’s economy, security, and health. Therefore, reliability is a primary metric. Nevertheless, the study of complex CPSs reliability demands understanding the joint dynamics of physical processes, hardware, software, and networks. In the present research, a series of studies is proposed to contribute to the challenging reliability analysis of CPSs by considering the reliability of physical components, hardware/software interactions, and overall reliability of CPSs modeled as networks. First, emerging technologies such as flexible electronics combined with data analytics and artificial intelligence, are now part of modern CPSs. In the present work, accelerated degradation testing (ADT) design and data analysis is considered for flexible hybrid electronic (FHE) devices, which can be part of the physical components or sensors of a CPS. Second, an important aspect of CPS is the interaction between hardware and software. Most of the existing work assume independency between hardware and software. In this work, a probabilistic approach is proposed to model such interactions using a Markov model and Monte Carlo simulation. Third, networks have been widely used to model CPSs reliability because they both have interconnected components. Estimating the network reliability by using traditional artificial neural networks (ANNs) has emerged as a promissory alternative to classical exact NP-hard algorithms; however, modern machine learning techniques have not been fully studied as reliability estimators for networks. This dissertation proposes the use of advanced deep learning (DL) techniques such as convolutional neural networks (CNNs) and deep neural networks (DNNs) for all-terminal network reliability estimation problem. DL techniques provide higher accuracy in reliability prediction as well as the possibility to dispense with computationally expensive inputs such the reliability upper bound. In addition, most of the previous works assume binary states for the components of networks, whereas the present work incorporates a Bayesian method to consider degradation for network reliability estimation and updating of parameters. | en_US |
