Prediction of ultimate Strength of prestressed concrete beams using artificial neural network

Other Title(s)

تقييم المقاومة القصوى للعتبات الخرسانية المسبقة الجهد باستخدام الشبكات العصبية الصناعية

Dissertant

Abd, Ruaa Talib

Thesis advisor

Jasim, Nabil Abd al-Razzaq

University

University of Basrah

Faculty

Engineering College

Department

Department of Civil Engineering

University Country

Iraq

Degree

Master

Degree Date

2010

English Abstract

This thesis investigated the use of the artificial neural networks in predicting the ultimate strength of prestressed concrete beams.

Because neural networks are massively parallel processors and have the ability to learn patterns through training experience, they are often well suited for modeling complex and non-linear processes.

Multilayered feed forward back propagation neural networks are used in this research, which are implemented using neural network toolbox that is available in MATLAB version 7.0.0 (2004).

This program implements several different neural network algorithms, including back propagation algorithm.

The descent gradient back propagation algorithm was employed for predicting the ultimate bending moment capacity of simply supported rectangular prestressed concrete beams.

The optimum topology (which gives least mean square error for both training and testing with fewer numbers of epochs) is presented.

Thus, the effects of the parameters, such as the number of hidden layer(s), number of nodes in the input layer, output layer and hidden layer (s), the pre-process of the training patterns, initialization weight factors and the selection of the learning rate and momentum coefficient, on the behaviour of the neural network have been investigated. Because of slow convergence of results when using descent gradient back propagation, another algorithm which is faster called " resilient back propagation algorithm " has been used to improve the performance of the neural network and the results have been compared with those obtained using the descent gradient back propagation algorithm.

Further, by using this model the optimum topology of network for predicting the ultimate shear capacity of simply supported I-section prestressed concrete beams and the ultimate capacity of simply supported T-section prestressed concrete beams under moment, shear and torsion were presented.

It was found that normalizing the input and target values of the training data by using Maximum and Minimum normalization method reduces the training time.

Also the initial value of weight factors and biases has greatly influenced the performance (mean square error) of the network model.

The Widrow-Hoff method was found to give a minimum mean square error. Finally, once the neural network has been trained, a parametric study is made to explore the effect of the various parameters on the behaviour of prestressed concrete beams. The increase in compressive strength from (35 to 80) MPa, increases the ultimate moment capacity by (35.26) %, while an increase in compressive strength from (20 to 60) MPa, increases the ultimate shear strength by (63.36) %.

The increase in prestressing steel stresses (fps) from (1500 to 1900) MPa, increases the ultimate moment capacity by (16.99) %.

However an increase in (fps) from (1500 to 1750) MPa, increases the ultimate shear strength by (37.82) %.

Also the increase in area of prestressing steel from (51.613 to 371.61) mm2 increases the ultimate moment capacity by (62.8 %).

Main Subjects

Civil Engineering

Topics

• Neural networks(Computer science)

No. of Pages

141

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-317176