Pull‐out capacity prediction of sustainable cementitious composites with artificial intelligence and statistical methods

Abstract

Concrete is used with reinforcement in structures so adherence gains importance especially in fire scenarios. To contribute production of sustainable concrete, by-products like granulated blast furnace slag, fly ash, and silica fume were used, generally. In this study, to determine the elevated temperature resistance of by-product-added composites, samples were exposed to 250, 500, and 750 C and bonding behavior of composites was determined with pull-out test. Moreover, prediction models were developed to estimate concrete-reinforcement adherence values without experimentation and importantly before a fire. For the prediction models, considering the cross correlation, the mixture type, temperature, compressive strength, and flexural strength of concretes were selected as independent variables different from literature and pull-out capacity (POC) as dependent variable. Artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) were used for prediction models. As a statistical method, multiple linear regression was used. The performances of pull out prediction models were compared according to coefficient of correlation (R), mean square error (MSE), and mean percentage error (MPE). Among the statistical models, Quadratic and pure quadratic performance surpassed the ANFIS model, which is a combination of ANN and fuzzy techniques. The best performance was obtained in ANN model with 99.85%, 3.65 kg, 3.1% for R, MSE and MPE, respectively. Therefore, ANN can provide a new applicable model to effectively predict POC of by-product-added cementituous composites under fire effect.