This research documents the development and evaluation of artificial neural network (ANN) models to predict the condition ratings of concrete highway bridge decks in Michigan. Historical condition assessments chronicled in the national bridge inventory (NBI) database were used to develop the ANN models. Two types of artificial neural networks, multi-layer perceptrons and ensembles of neural networks (ENNs), were developed and their performance was evaluated by comparing them against recorded field inspections and using statistical methods. The MLP and ENN models had an average predictive capability across all ratings of 83% and 85%,respectively, when allowed a variance equal to bridge inspectors. A method to extract the influence of parameters from the ANN models was implemented and the results are consistent with the expectations from engineering judgment. An approach for generalizing the neural networks for a population of bridges was developed and compared with Markov chain methods. Thus, the developed ANN models allow modeling of bridge deck deterioration at the project (i.e., a specific existing or new bridge) and system/network levels. Further, the generalized ANN degradation curves provided a more detailed degradation profile than what can be generated using Markov models. A bridge management system (BMS) that optimizes the allocation of repair and maintenance funds for a network of bridges is proposed. The BMS uses a genetic algorithm and the trained ENN models to predict bridge deck degradation. Employing the proposed BMS leads to the selection of optimal bridge repair strategies to protect valuable infrastructure assets while satisfying budgetary constraints. A program for deck degradation modeling based on trained ENN models was developed as part of this project.

Bridges play a significant role in the U.S. economy. The number of the bridges in the U.S. exceeds six hundred thousand. Almost one third of them are considered structurally deficient and will require more than $164 billion to repair or replace. Identifying the factors that affect the performance of concrete bridge decks during its service life is critical to the development of an accurate condition assessment and deterioration prediction model. Accurate bridge deck deterioration models can provide vital information for predicting short- and long-term behavior of concrete bridge decks and minimizing costly routine inspection and maintenance activities. Therefore, the main goal of this dissertation is to develop a deterioration prediction model for concrete bridge decks that is based on the National Bridge Inventory (NBI) database. To achieve the goal, five deterioration prediction models for concrete bridge decks were developed using Multinomial Logistic Regression, Decision Tree, Artificial Neural Network, k-Nearest Neighbors and Naive Bayesian machine learning techniques. Michigan bridge deck data from NBI between the years 1992 to 2015 were used for training the various prediction models. The results show that the performance of all five developed models were acceptable. However, the artificial neural network achieved the highest accuracy in the validation process. Additionally, bridge decks age, area, average daily traffic, and skew angle are found to be significant factors in the deterioration of concrete bridge decks. Furthermore, it was observed that bridge decks could stay in their condition rating more than the typical 2-year inspection interval, suggesting that inspection schedules could be extended for certain bridges that had slower deterioration rates. The contributions of this work include 1) the development of an optimized deterioration prediction model that can be used in the condition assessment process for concrete bridge decks, 2)the identification of the factors that have the most impact on concrete bridge deck deterioration,and 3) demonstrating that the inspection schedule can be longer than 2 years for bridges that do not deteriorate fast which can lead to cost and time savings. Future work can include the following: (1)developing deterioration prediction models for concrete bridge decks using deep learning techniques; (2) developing deterioration prediction models for other bridge specific elements (i.e., superstructure and substructure) using multivariant analysis; (3) developing deterioration prediction models for other (or all) U.S. states using the framework developed in this research; and (4) investigating the prospect of revising the mandated inspection interval beyond the 2-year period.

As the backbone of the US transportation system, bridges are also its most visible part. There are over 600,000 bridges across all US states ensuring network continuity. In order to optimize such activities and use the available monies most effectively, a solid understanding of the parameters that affect the performance of concrete bridge decks is critical. The National Bridge Inventory (NBI), perhaps the single-most comprehensive source of bridge information, gathers data on more than 600,000 bridges in all fifty states, the District of Columbia, and the Commonwealth of Puerto Rico. Recently there has been a growing interest in analyzing the NBI database. The NBI uses visual inspection, a commonly practiced damage detection method, to rate bridge decks. Focusing on concrete highway bridge deck performance, the present study developed a nationwide database based on NBI data and other critical parameters, such as bridge age, deck area, climatic regions, and distance from seawater. Additionally, two new performance parameters were computed from the available concrete bridge deck condition ratings (CR): Time-in-condition rating (TICR) and deterioration rate (DR). Following the aggregation of all these parameters to form a nationwide database, filtering and processing were performed. Approaches to dealing with inconsistencies and missing data are proposed as well. After developing the nationwide database this research presents network-level, one-way statistical relationships to get a better understanding of the parameters. ☐ Next, a data mining technique on the nationwide database was used to analyze the data. Data mining is a discovery procedure to explore and visualize useful but less-than-obvious information or patterns embedded in large collections of data. Given the amount and variety of parameter types in a large data set such as that of the nationwide database, using traditional clustering techniques for discovery is impractical. As a consequence, this research has applied a novel data discovery tool called two-step cluster analysis to visualize associations between concrete bridge deck design parameters and bridge deck condition ratings. Two-step cluster analysis is a powerful knowledge discovery tool that can handle categorical and interval data simultaneously and is capable of reducing dimensions for large data sets. The two-step cluster analysis is a useful tool for bridge owners and agencies to visualize general trends in their concrete bridge deck condition data and support them in their decision-making processes to effectively allocate constrained funds for maintenance, repair, and design of bridge decks. ☐ Understanding the attributes of bridge deck performance is central to asset management. This research attempts to characterize how various environmental and structural parameters affect bridge deck performance by employing a binary logistic regression. The logistic model shows the relationship between a dependent variable (lowest vs. highest bridge deck deterioration) and the relative importance of a number of independent variables selected for this study (predictor variables). Observations of extreme bridge deck deterioration taken from the nationwide database were used in the model. Bridge deck deterioration was computed as the decrease in CR over time. Maintenance responsibility fulfillment, functional classification of inventory route, design and construction type, average daily truck traffic, climatic regions, and distance to seawater, were all used as independent variables. Our application of a binary logistic regression model for bridge deck deterioration provides practical insight regarding how certain parameters influence bridge deck performance. ☐ A leading factor in structural decline of highway bridges is the deterioration of concrete decks. Thus, a method to forecast bridge deck performance is vital for transportation agencies to allocate future repair and rehabilitation funds. The objective of this study was the development of a nationwide CR deterioration model based on the nationwide database through the use of a Bayesian statistical approach that predicts probability of CR decrease. In addition to CR data, the impact of other governing factors on CR decrease are shown in the paper, such as average daily truck traffic (ADTT), maintenance responsibility fulfillment, deck structure type, and regional climate effect. One singular advantage of this method is that it can be continually updated as additional NBI information becomes available. Moreover, the results of this model can be used as prior data in future Bayesian studies. The results presented in this study, by providing a better idea of how US concrete bridge decks are performing based on the NBI data, are intended to furnish a progressive bridge management system. ☐ Results yielded by each of the analysis above will encourage future researchers to add other crucial parameters not contained in the nationwide database such as structural design characteristics (e.g., minimum deck thickness), construction practices (e.g., curing practices), specifications (e.g., water-to-cement ratio), and other notable factors (e.g., application of deicing salts). Furthermore, analyze the nationwide database in various statistical application areas leading to more accurate understating of the factors affecting bridge deck deterioration and enhanced deck deterioration prediction models.

" TRB's second Strategic Highway Research Program (SHRP 2) Report S2-R06A-RR-1: Nondestructive Testing to Identify Concrete Bridge Deck Deterioration identifies nondestructive testing technologies for detecting and characterizing common forms of deterioration in concrete bridge decks.The report also documents the validation of promising technologies, and grades and ranks the technologies based on results of the validations.The main product of this project will be an electronic repository for practitioners, known as the NDToolbox, which will provide information regarding recommended technologies for the detection of a particular deterioration. " -- publisher's description.

Bridges are key elements in the US transportation system. There are more than six hundred thousand bridges on the highway system in the United States. Approximately one third of these bridges are in need of maintenance and will cost more than $120 billion to rehabilitate or repair. Several factors affect the performance of bridges over their life spans. Identifying these factors and accurately assessing the condition of bridges are critical in the development of an effective maintenance program. While there are several methods available for condition assessment, selecting the best technique remains a challenge. Therefore, developing an accurate and reliable model for concrete bridge deck deterioration is a key step towards improving the overall bridge condition assessment process. Consequently, the main goal of this dissertation is to develop an improved bridge deck deterioration prediction model that is based on the National Bridge Inventory (NBI) database. To achieve the goal, deterministic and stochastic approaches have been investigated to model the condition of bridge decks. While the literatures have typically proposed the Markov chain method as the best technique for the condition assessment of bridges, this dissertation reveals that some probability distribution functions, such as Lognormal and Weibull, could be better prediction models for concrete bridge decks under certain condition ratings. A new universal framework for optimizing the performance of prediction of concrete bridge deck condition was developed for this study. The framework is based on a nonlinear regression model that combines the Markov chain method with a state-specific probability distribution function. In this dissertation, it was observed that on average, bridge decks could stay much longer in their condition ratings than the typical 2-year inspection interval, suggesting that inspection schedules might be extended beyond 2 years for bridges in certain condition rating ranges. The results also showed that the best statistical model varied from one state to another and there was no universal statistical prediction model that can be developed for all states. The new framework was implemented on Michigan data and demonstrated that the prediction error in the combined model was less than each of the two models (i.e. Markov and Lognormal). The results also showed that average daily traffic, age, deck area, structure type, skew angle, and environmental factors have significant impact on the deterioration of concrete bridge decks. The contributions of the work presented in this dissertation include: 1) the identification of the significant factors that impact concrete bridge deck deterioration; 2) the development of a universal deterioration prediction framework that can be uniquely tailored for each state’s data; and 3) supporting the possibility of extending inspection schedules beyond the typical 2-year cycles. Future work may involve: 1) evaluating each of the factors that impact the deterioration rates in more depth by refining the investigation ranges; 2) investigating the possibility of revising the regular bridge deck inspection intervals beyond the 2-year cycles; and 3) developing deterioration prediction models for other bridge elements (i.e. superstructure and substructure) using the framework developed in this dissertation.

This report summarizes a research project aimed at developing degradation models for bridge decks in the state of Michigan based on durability mechanics. A probabilistic framework to implement local-level mechanistic-based models for predicting the chloride-induced corrosion of the RC deck was developed. The methodology is a two-level strategy: a three-phase corrosion process was modeled at a local (unit cell) level to predict the time of surface cracking while a Monte Carlo simulation (MCS) approach was implemented on a representative number of cells to predict global (bridge deck) level degradation by estimating cumulative damage of a complete deck. The predicted damage severity and extent over the deck domain was mapped to the structural condition rating scale prescribed by the National Bridge Inventory (NBI). The influence of multiple effects was investigated by implementing a carbonation induced corrosion deterministic model. By utilizing realistic and site-specific model inputs, the statistics-based framework is capable of estimating the service states of RC decks for comparison with field data at the project level. Predicted results showed that different surface cracking time can be identified by the local deterministic model due to the variation of material and environmental properties based on probability distributions. Bridges from different regions in Michigan were used to validate the prediction model and the results show a good match between observed and predicted bridge condition ratings. A parametric study was carried out to calibrate the influence of key material properties and environmental parameters on service life prediction and facilitate use of the model. A computer program with a user-friendly interface was developed for degradation modeling due to chloride induced corrosion.