Evaluation of compressive strength and rapid chloride permeability test of concretes containing metakaolin using Bayesian inference and GEP methods

Compressive strength (CS) and rapid chloride permeability test (RCPT) are the most important tests in the concrete industry. CS is the most significant characteristic of concrete mechanical properties that can show other mechanical properties like the module of elasticity. Chloride penetration could degradation of concrete durability. In the Persian Gulf, chloride penetration is the most dangerous effect on steel rebar corrosion. Therefore, CS and RCPT are related to mechanical and durability properties and should be studied more carefully. In this research, CS and RCPT are predicted using soft computing. For this purpose, Bayesian inference is used for prediction of them. Bayesian inference is a subset of linear regression but unlike conventional regressions that are deterministic, this type of regression is probabilistic. So, in this research is used of probabilistic analysis replaced deterministic analysis. Gene expression programming (GEP) is used for comparison of their results versus Bayesian inference. For research performing, 100 concrete samples containing metakaolin are considered that 75 samples are selected as training, and 25 samples are selected as testing data. seven input data are considered for prediction of CS and RCPT that contains the age of concrete (day), cement (kg/m3), water (kg/m3), metakaolin (kg/m3), fine aggregate (kg/m3), coarse aggregate (kg/m3) and surface resistance (KΩS). Output parameters are CS (MPa) and RCPT (Coulomb) that for predicting them, independent analysis should be performed. Results show that Bayesian inference in CS prediction has an excellent ability that the R2 coefficient for training and testing is 0.96. These values for GEP were 0.93 and 0.96 respectively. Values of root mean square error (RMSE) and mean absolute error (MAE) in Bayesian inference for training are 2.55 and 1.84 MPa respectively. These values for testing are 2.75 and 2.25 MPa. The values of RMSE and MAE for GEP training are 3.46 and 2.60 MPa and for testing these values are 3.43 and 2.65 MPa respectively. A comparison between evaluation parameters (i.e. R2, RMSE, and MAE) showed that Bayesian inference and GEP have excellent accuracy. In Bayesian inference, R2 coefficients for RCPT training and testing are 0.98 and 0.97 respectively. These values for GEP are 0.96 and 0.97 respectively. RMSE and MAE values in Bayesian inference for training are 223.14 and 161.58 Coulomb and these values for testing are 269.56 and 233.25 Coulomb respectively. RMSE and MAE values for GEP in training are 311.73 and 239.34 Coulomb respectively and these values for testing are 306.92 and 252.67 respectively. Results of CS and RCPT are showed that Bayesian inference is a good method for the prediction of concrete properties. On the other side, Bayesian is linear and has a little time consuming compared to nonlinear methods like GEP. In the next part of this study, first-order reliability method (FORM) is used for reliability analysis of CS and RCPT. Reliability index or beta and probability of failure (Pf) are the most important component in FORM analysis that are calculated in each analysis. For this purpose, mean values of input data are selected as inputs in reliability analysis. Results of reliability analysis indicated that when the CS is considered less than 45 MPa, the probability of failure is not considerable. Reliability analysis of RCPT in concrete samples is indicated that the value of 2000 Coulomb is a threshold value for the probability of failure. Therefore, if the RCPT of concrete samples is less than 2000 Coulomb, the probability of permeability is increased.