Date of Award
Doctor of Philosophy (PhD)
Dr. Brandon Ross, Committee Chair
Dr. Amin Khademi
Dr. Mashrur Chowdhury
Dr. Weichiang Pang
The quality of highway bridge infrastructure in United States is of major concern. One in every four bridges in the US is deficient. This research applied Artificial Intelligence, Systems Dynamics and linear modeling techniques to investigate the causes and effects of bridge deterioration and to forecast bridge infrastructure condition and improvement costs. The main contribution of the research is the development and demonstration of these methods within the context of highway bridges. These methods provide bridge designers and policy makers new tools for maintaining, improving, and delivering high quality bridge infrastructure. To start with, a comprehensive review of the current state of bridge deficiency in US was conducted. Through extensive data mining of the National Bridge Inventory (NBI), the causes and trends in bridge deficiency were identified. This exercise addressed questions such as: What is the current extent of bridge deficiency? Is deficiency getting better or worse? What are the biggest problems causing deficiencies? It was observed that though the general condition of bridges is improving, additional work needs to be done in fixing bridge deficiency and bridge functionally obsolescence in particular. Subsequent to the review of bridge deficiency, four distinct but related modeling studies were conducted. These phases are: 1) Capacity Obsolescence/Sustainability assessment, 2) Causal Loop Diagram (CLD) and linear modeling for bridge improvement costs, 3) Artificial Neural Network (ANN) model for bridge condition ratings and bridge variable effects, 4) Non-linear auto regression (NARX) model for bridge inventory condition prediction. In the first phase, a conceptual model was developed to minimize capacity obsolescence, one face of functional obsolescence. A framework was developed to minimize bridge capacity obsolescence while optimizing the use of embodied energy over the service life of bridges. The research demonstrated how design phase consideration of bridge obsolescence can contribute to sustainability of bridge infrastructure. As a novel approach for studying bridge improvement costs, the second phase used a Causal Loop Diagram (CLD), a tool used in the field of System Dynamics. Using a CLD, the causes and effects for bridge deterioration were qualitatively described. A segment of the qualitative relationships described through the CLD were then analyzed quantitatively for the South Carolina bridge inventory. The quantitative model was based on linear modeling and was developed and validated using NBI data. The model was then applied to estimate future bridge inventory sufficiency ratings and improvement costs under possible funding scenarios. For effective mitigation of bridge deficiency, it is important to identify the effects of different variables on bridge conditions and forecast bridge condition. In the third phase of modeling, Artificial Neural Networks (ANN) models were used to study the effects of bridge variables on bridge deck and superstructure condition ratings. The models considered prestressed concrete bridges in South Eastern United States. Simulations based on Full Factorial Design (FFD) were conducted using the developed ANN models. The simulations highlighted the effects of skew, span and age on bridge condition ratings. Given sufficient source data, the approach can be broadly applied to consider other bridge types and design variables. In the last phase, time based ANN learning algorithms were used to forecast bridge condition ratings and bridge improvement costs. Non Linear Auto Regression with Exogenous Inputs (NARX) model was developed using NBI data for South Carolina bridges over the last decade. The study estimated bridge condition ratings as a function of bridge geometry, age, structural, traffic attributes and bridge improvement spending. This doctoral research contributed to the development of multiple qualitative and mathematical models for forecasting bridge inventory condition and improvement costs by applying ANN, CLD, and linear regression techniques. While the conclusions of these studies are bound by the scope of the data and methodical constraints of the research, the methods can be more generally applied to aid in better bridge management policies and contribute to sustainable bridge infrastructure in United States.
Jonnalagadda, Srimaruthi, "Artificial Neural Networks, Non Linear Auto Regression Networks (NARX) and Causal Loop Diagram Approaches for Modelling Bridge Infrastructure Conditions" (2016). All Dissertations. 1725.