Date of Graduation

7-2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Engineering (PhD)

Degree Level

Graduate

Department

Industrial Engineering

Advisor

Kelly M. Sullivan

Committee Member

Edward Pohl

Second Committee Member

Shengfan Zhang

Third Committee Member

Hugh Medal

Keywords

network reliability, node deployment, destruction algorithms, stochastic sensor, D-spectrum, Markov decision process (MDP), approximate dynamic programming (ADP)

Abstract

The long-term operation of a wireless sensor network (WSN) requires the deployment of new sensors over time to restore any loss in network coverage and communication ability resulting from sensor failures. Over the course of several deployment actions it is important to consider the cost of maintaining the WSN in addition to any desired performance measures such as coverage, connectivity, or reliability. The resulting problem formulation is approached first through a time-based deployment model in which the network is restored to a fixed size at periodic time intervals. The network destruction spectrum (D-spectrum) has been introduced to estimate reliability and is more commonly applied to a static network, rather than a dynamic network where new sensors are deployed over time. We discuss how the D-spectrum can be incorporated to estimate reliability of a time-based deployment policy and the features that allow a wide range of deployment policies to be evaluated in an efficient manner. We next focus on a myopic condition-based deployment model where the network is observed at periodic time intervals and a fixed budget is available to deploy new sensors with each observation. With a limited budget available the model must address the complexity present in a dynamic network size in addition to a dynamic network topology, and the dependence of network reliability on the deployment action. We discuss how the D-spectrum can be applied to the myopic condition-based deployment problem, illustrating the value of the D-spectrum in a variety of maintenance settings beyond the traditional static network reliability problem. From the insight of the time-based and myopic condition-based deployment models, we present a Markov decision process (MDP) model for the condition-based deployment problem that captures the benefit of an action beyond the current time period. Methodology related to approximate dynamic programming (ADP) and approximate value iteration algorithms is presented to search for high quality deployment policies. In addition to the time-based and myopic condition-based deployment models, the MDP model is one of the few addressing the repeated deployment of new sensors as well as an emphasis on network reliability. For each model we discuss the relevant problem formulation, methodology to estimate network reliability, and demonstrate the performance in a range of test instances, comparing to alternative policies or models as appropriate. We conclude with a stochastic optimization model focused on a slightly different objective to maximize expected coverage with uncertainty in where a sensor lands in the network. We discuss a heuristic solution method that seeks to determine an optimal deployment of sensors, present results for a wide range of network sizes and explore the impact of sensor failures on both the model formulation and resulting deployment policy.

Share

COinS