The present study evaluates the performance of fiber-reinforced geopolymer, especially its flexural strength, using a Deep Learning (DL) approach, Deep Residual Network (ResNet), and the experimental work is presented. A total of 245 mixtures were employed to generate the data for the ResNet training and validating procedures. In the proposed model, the Fly Ash (FA) content, sodium silicate solution/solid binder ratio, curing temperature, curing time, fiber volume fraction, fiber length (l) and diameter (d), as well as fiber tensile strength, were considered as input factors. In contrast, flexural strength was the output parameter. The effectiveness of ResNet was evaluated by three statistical factors, correlation coefficient (R2), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). ResNet validation revealed the effectiveness of predictive methods with 94.5%, 0.292 MPa, and 4.068% for R2, RMSE, and MAPE, respectively. The suggested models may be used as standard mixtures for geopolymer concrete reinforced with steel fibers.
Read full abstract