Developing Artificial Neural Network and Multiple Linear Regression Models to Predict the Ultimate Load Carrying Capacity of Reactive Powder Concrete Columns

Abstract

The study focuses on development a model to predict the ultimate load carrying capacity of Reactive Powder Concrete (RPC) columns. Two different statistical methods regression techniques (RT) and the artificial neural network (ANN) methods were used for determining the RPC columns ultimate load carrying capacity. The data is collected from three experimental studies the first used to develop the model and the other two used as a case study. Experimental results used as input data to develop prediction models. Two different techniques adopted to develop the models the first was Artificial Neural Network (ANN) and the second was multi linear regression techniques (RT). The models use to predict the ultimate load carrying capacity of RPC columns. To predict the ultimate load carrying capacity of RPC columns four input parameters were identified cross-section, micro steel fiber volume fraction content, compressive strength and main steel reinforcement area. Both models build with assistance of MATLAB software. The results exhibit that the cross section area has most significant effect on ultimate load carrying capacity. The performance of ANNs with different architecture was considered to adopt the pest ANN. An ANN with one layer consist of 7 neurons provide the best prediction. The results of this investigation indicate that ANNs have strong potential as statistical method for prediction the ultimate load carrying capacity of RPC columns.

Keywords: Reactive powder concrete, artificial neural network, multiple linear regressions, ultimate load carrying capacity, Statistical analysis.