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

Filters

2008 - 200912010 - 20131

Settings

Subject: search.filters.subject.Artificial Neural Network

Search Results

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    SENSITIVITY EVALUATION OF THE MECHANISTIC-EMPIRICAL PAVEMENT DESIGN GUIDE (MEPDG) FOR FLEXIBLE PAVEMENT PERFORMANCE PREDICTION
    (2013) Li, Rui; Schwartz, Charles W.; Civil Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    The AASHTO Mechanistic-Empirical Pavement Design Guide (MEPDG) provides pavement analysis and performance predictions under various feasible design scenarios. The MEPDG performance predictions for the anticipated climatic and traffic conditions will depend on the values of the input parameters that characterize the pavement materials, layers, design features, and condition. This research focuses on comprehensive sensitivity analyses of flexible pavement performance predictions to MEPDG design inputs under five climatic conditions and three traffic levels. Design inputs evaluated in the analyses include traffic volume, layer thicknesses, material properties, groundwater depth, and others. Correlations among design inputs were considered where appropriate. Both local One-At-a-Time (OAT) sensitivity analyses and global sensitivity analyses (GSA) were performed. Two response surface modeling (RSM) approaches, multivariate linear regressions and artificial neural networks (ANN), were developed to model the GSA results for evaluation of MEPDG input sensitivities across the entire problem domain. The ANN-based RSMs were particularly effective in providing robust and accurate representations of the complex relationships between MEPDG inputs and distress outputs. The design limit normalized sensitivity index (NSI) adopted in this study provides practical interpretation of sensitivity by relating a given percentage change in a MEPDG input to the corresponding percentage change in predicted distress/service life relative to its design limit value. The design inputs most consistently in the highest sensitivity categories across all distresses were the hot mix asphalt (HMA) dynamic modulus master curve, HMA thickness, surface shortwave absorptivity, and HMA Poisson's ratio. Longitudinal and alligator fatigue cracking were also very sensitive to granular base thickness and resilient modulus and subgrade resilient modulus. Additional findings are also provided for each specific pavement type.
  • Thumbnail Image
    Item
    Predicting Water Table Fluctuations Using Artificial Neural Network
    (2008-11-17) Wu, Chung-Yu; Shirmohammadi, Adel; Fischell Department of Bioengineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    Correctly forecasting groundwater level fluctuations can assist water resource managers and engineers in efficient allocation of the regional water needs. Modeling such systems based on satellite remotely sensed data may be a viable option to predict water table fluctuations. Two types of water table prediction models based on Artificial Neural Network (ANN) technology were developed to simulate the water table fluctuations at two well sites in Maryland. One was based on the relationship between the variations of brightness temperature and water table depth. The other one was based on the relationship between the changes of soil moisture and water table depth. Water table depths recorded at these two wells, brightness temperature retrieved from the Advanced Microwave Scanning Radiometer, and soil moisture data produced by the Land Data Assimilation System were used to train and validate the models. Three models were constructed and they all performed well in predicting water table fluctuations. The root mean square errors of the water table depth forecasts for 12 months were between 0.043m and 0.047m for these three models. The results of sensitivity test showed that the models were more sensitive to the uncertainty in water table depth than to that in brightness temperature or in soil moisture content. This suggests that for situations where high resolution remotely sensed data is not available, an ANN water table prediction model still can be built if the trend of the time series of the data, such as brightness temperature or soil moisture, over the study site correlates well with the trend of the time series of the ground measurement at the study site. An extension of the study to a regional scale was also performed at 12 available well sites in Piedmont Plateau, Maryland. Hydrologic soil types, LDAS soil moistures, and water table depths at these locations were used in the ANN modeling. The root mean square error of one month long water table depth forecast was 0.142m. However, the accuracy of the monthly forecast decreases with the increase of time. A further study to improve the accuracy of long-term water table fluctuation forecast is recommended.