Neuro-fuzzy application for concrete strength prediction using combined non-destructive tests

The application of the neuro-fuzzy inference system to predict the compressive strength of concrete is presented in this study. The adaptive neuro-fuzzy inference system (ANFIS) is introduced for training and testing the data sets consisting of various parameters. To investigate the influence of various parameters which affect the compressive strength, 1551 data pairs are collected from the technical literature. These data sets cover early and late compressive strengths from 3 to 365 days and low and high strength in the range 6.3–107.7 MPa. To reflect the effects of other uncertain parameters and in situ conditions, the results of non-destructive tests (NDTs) such as ultrasonic pulse velocity (UPV) and rebound hammer test are also included as input parameters, in addition to mix proportion and curing histories. For the testing of trained ANFIS models, 20 cube specimens and 210 cylinders are prepared, and compressive test and NDTs are conducted. For the comparative study of the applicability of ANFIS models combined with NDT results, four ANFIS models are developed. Depending on whether the input parameters of ANFIS models include NDT results or not, these are distinguished from each other. Among the four models, the ‘ANFIS-UR’ model having the parameters for both UPV and rebound hammer test results shows the best accuracy in the prediction of compressive strength.