PREDICTION OF CONCRETE MODE I CRITICAL STRESS INTENSITY FACTOR VIA ARTIFICIAL NEURAL NETWORKS

In this paper, the potential of using Artificial Neural Networks (ANN) to predict concrete mode I critical stress intensity factor KeIc ; for the Two Parameter Model (TPM) ; and KeIc ; for the effective crack model (ECM); is investigated. The experimental results from three hundred and sixteen tests of three point bend specimens were utilized in this study. To train the network to discover the relationship between the concrete mode I critical stress intensity factor KeIc and the main parameters affecting it two hundred and fifty three tests were utilized. To check the effectiveness of the trained network, the calculated mode I critical stress intensity factors of the training data sets and another sixty-three data sets not known before to the network were compared. Excellent agreement was found. Several neural-networks were implemented. The General Regression Neural Network (GRNN) proves to be the best to the present study. The influence of material and specimen dimension parameters on the prediction of mode I critical stress intensity factors of concrete were checked. Concrete compressive strength was found to be the most influencing parameter. The GRNN was successful to predict KeIc for high and normal strength concrete The relationship between KeIc and KeIc was investigated. Excellent correlation was found. Finally, the Group Method of Data Handling (GMDH) network was utilized to produce a formula for KeIc as a function of the parameters affecting it.