Prediction of compressive strength of self-compacting concrete containing different fillers using artificial neural networks

Accuracy in pouring concrete, concrete density and also the appearance of concrete as an exposed material is always a concern of designers and executors of construction projects. Self-compacting concrete with weight compression properties can always be one of the options available to designers. The variety of materials used in self-compacting concrete, including recycled materials, with pozzolanic properties and fillers to achieve rheological and mechanical goals, is one of the challenges that designers face. Also, accurate determination of mixing ratios and their results is very time consuming and costly. Using soft computing and neural networks inspired by the biological structure of the human brain, computer science seeks to increase speed, accuracy, and cost reduction to prevent malicious experiments. In this study, with the help of ANN and LSTM networks, using 320 samples of self-compacting concrete with dispersion and comprehensiveness of common materials used in it by various researchers, tried to predict 28-day compressive strength of self-compacting concrete, evaluate performance and increase accuracy by 6 different training algorithms. In total, about 200 repetitions of training were performed on 320 samples of self-compacting concrete with 14 characteristics, which by comparing the best results obtained from training algorithms, best performance with root mean square error of 4.97 and correlation coefficient of 0.9484 in the test, for the network. ANN was reported with the Beyesian Regularization training algorithm, which indicates the high accuracy of that network.