Development and Applications of Artificial Neural Network for Prediction of Ultimate Bearing Capacity of Soil and Compressive Strength of Concrete

• Main Author: Seyed Hakim, Seyed Jamalaldin
• Format: Thesis
• Language: English; English
• Published: 2006
• Online Access: http://psasir.upm.edu.my/id/eprint/631/1/600423_fk_2006_100_abstrak_je__dh_pdf_.pdf; http://psasir.upm.edu.my/id/eprint/631/

Artificial Neural Networks (ANNs) have recently been widely used to model some of the human activities in many areas of science and engineering. One of the distinct characteristics of the ANNs is its ability to learn from experience and examples and then to adapt with changing situations. ANNs does not need a specific equation form that differs from traditional prediction models. Instead of that, it needs enough input-output data. Also, it can continuously re-train the new data, so that it can conveniently adapt to new data. This research work focuses on development and application of artificial neural networks in some specific civil engineering problems such as prediction of ultimate bearing capacity of soil and compressive strength of concrete after 28 days. One of the main objectives of this study was the development and application of an ANN for predicting of the ultimate bearing capacity of soil. Hence, a large training set of actual ultimate bearing capacity of soil cases was used to train the network. A neural network model was developed using 1660 data set of nine inputs including the width of foundation, friction angle in three layer, cohesion of three layers and depth of first and second layer are selected as inputs for predicting of ultimate bearing capacity in soil. The model contained a training data set of 1180 cases, a verification data set of 240 cases and a testing data set of 240 cases. The training was terminated when the average training error reached 0.002. Many combinations of layers, number of neurons, activation functions, different values for learning rate and momentum were considered and the results were validated using an independent validation data set. Finally 9-15-1 is chosen as the architecture of neural network in this study. That means 9 inputs with a set of 15 neurons in hidden layer has the most reasonable agreement architecture. This architecture gave high accuracy and reasonable Mean Square Error (MSE). The network computes the mean squared error between the actual and predicted values for output over all patterns. Calculation of mean percentage relative error for training set data, show that artificial neural network predicted ultimate bearing capacity with error of 14.83%. The results prove that the artificial neural network can work sufficiently for predicting of ultimate bearing capacity as an expert system. It was observed that overall construction-related parameters played a role in affecting ultimate bearing capacity, but especially the parameter “friction angle” play a most important role. An important observation is that influencing of the parameter “cohesion” is too less than another parameters for calculating of ultimate bearing capacity of soil. Also in this thesis is aimed at demonstrating the possibilities of adapting artificial neural Also in this thesis is aimed at demonstrating the possibilities of adapting artificial neural networks (ANN) to predict the compressive strength of concrete. To predict the compressive strength of concrete the six input parameters, such as, cement, water, silica fume, superplasticizer, fine aggregate and coarse aggregate identified. Total of 639 different data sets of concrete were collected from the technical literature. Training data sets comprises 400 data entries, and the remaining data entries (239) are divided between the validation and testing sets. The training was stopped when the average training error reached 0.007. A detailed study was carried out, considering two hidden layers for the architecture of neural network. The performance of the 6-12-6-1 architecture was the best possible architecture. The MSE for the training set was 5.33% for the 400 training data points, 6.13% for the 100 verification data points and 6.02 % for the 139 testing data points. It can recognize the concrete in term of ‘strength’ with a confidence level of about 95%, which is considered as satisfactory from an engineering point of view. It was found from sensitivity analyses performed on a neural network model that the cement has the maximum impact on the compressive strength of concrete. Finally, the results of the present investigation were very encouraging and indicate that ANNs have strong potential as a feasible tool for predicting the ultimate bearing capacity of soil and compressive strength of concrete.