Determination of Moment Capacities of Spiral Columns with Artificial Neural Networks

Abstract

There have been a number of experimental studies in the literature concerning spiral columns, which are preferred in reinforced concrete buildings in some cases. An analytical study is conducted by using an artificial intelligence technique, which is capable of giving almost perfect results provided that a closely associated information is entered, and also considerably shortens the time spent to solve the problem, through the data set obtained from these experimental studies. Moment bearing capacities of spiral columns, which have different engineering characteristics, are determined with ANN technique, which is commonly used today in construction engineering as well as in many disciplines. Furthermore, the results obtained from the analyzes are compared with the results of laboratory studies and, thus, success of ANN in this field is tested. The generated data set is composed of 79 experimental elements. 50 of this data set are used for training of algorithms and the remainder 29 evaluated for testing purposes. In the study, parameters, such as column diameter, concrete cover, column effective length, mechanical properties of concrete and reinforcement, longitudinal and transverse reinforcement ratio, axial load level, are specified as input layer while creating ANN network architecture. Moment bearing capacities of spiral columns are specified individually for 12 different algorithms and performance criteria values are referred for the accuracy of the prediction values found. As a result of training of the created network architecture with the SCG algorithm, a sensitivity of 97.62% is achieved, which indicates a highly appropriate prediction in this field.