Date of Award

Fall 1-1-2020

Degree Type


Degree Name

Master of Civil Engineering Degree


Civil And Environmental Engineering

First Advisor

Owusu-danquah, Josiah

Second Advisor

Dr. Srinivas Allena

Third Advisor

Dr. Stephen F Duffy


Artificial Neural Network (ANN), which is inspired by biological neural networks in the human brain, is one important tool of machine learning that creates artificial intelligence through computational systems. The creation of this intelligence is contingent on learning from available data regarding a specific subject. Although machine learning, in general, has profuse applications in most scientific disciplines, yet few have been developed in civil engineering due to the required time consuming and demanding programming. In order to minimize this, intelligible ANN software has been developed in this research capable of training networks with any number of hidden layers and nodes for each layer. Furthermore, two models have been created to demonstrate the robust applications of ANN. The first application involves a simulation of the straintemperature behavior of a shape memory alloy (SMA) under thermal cycling. In the second case, the bond strength between the concrete and the steel-reinforced bars is predicted considering the effects of steel corrosion level, concrete compressive strength, and concrete cover. Java programming language was used in developing the ANN software and a simple graphical user interface (GUI) has been designed, allowing the user to control the inputs and the training progress, make predictions and save the outputs. In this study, the ANN models were developed with different structures and activation functions to prove the ANN eminent idiosyncrasy of modeling data from different fields. Comparison is made between these models as well as models created by statistical iv regression and other models available in the literature. The developed software can efficiently train ANNs with any structure, as less time is needed to develop one ANN using the software than using programming methods. Moreover, the user will have the option to save the weights and the biases at any iteration and predict responses for the currently trained or previously trained ANN. The model predicted results can be saved or exported as an excel file. In terms of the created models, ANN can capture highly complicated relationships accurately and effectively compared to traditional modeling methods. Based on that, more accurate predictions are expected using ANN.