Semester

Summer

Date of Graduation

2003

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Civil and Environmental Engineering

Committee Chair

Robert N. Eli.

Abstract

As a result of an earlier WV DOH study, the idea came to the forefront of using a completely new approach to analyzing the complex subject of culvert hydrodynamics. The literature indicates that there have been no reports of artificial intelligence, to include neural networks, fuzzy logic, or combined neural-fuzzy logic, used to investigate and predict culvert hydrodynamics.;The scope of this dissertation is to investigate the applicability of using neural-fuzzy logic to predict culvert diameters. To analyze these flows, commercial culvert software was employed to account for all types of flow conditions. This included different slopes, lengths, flow-rates, pipe sizes, and headwater and tail water conditions. For all of the variables included in the analysis of culvert flow, some are complex in nature and require selection of different parameters. A large data set was created, from which to draw out different flow types for analysis. The use of fuzzy logic enables the user to enter variables and the developed code then interprets the data and solves for diameter. These trained data sets have a compliment checking data which is derived from similar calculations, with one variable slightly larger. These data sets were trained in a neural-fuzzy model and the result was a predicted culvert diameter data set. The predicted diameters were then compared to the actual diameters to determine the accuracy of the model. For all data sets evaluated, the root mean square error was less than 12 inches. The overall weighted root mean squared error for the training data sets was 1.989 inches and 2.658 inches for the checking data sets.

Download

Share

COinS