Prediction of Compressive Strength by Considering Practical Consideration Non-destructive Test by Artificial Neural Network

Accurate assessment of concrete compressive strength is critical for evaluating structural performance. While nondestructive testing (NDT) methods, such as Schmidt rebound hammer tests, offer rapid and NDT gives result with reasonable accurate based on environmental factors such as temperature, humidity etc of site and condition in which test is performed. Destructive testing (DT) methods, like core cutting, provide direct and accurate results. This study aimed to bridge the gap between these approaches by developing predictive models that correlate DT and NDT results. Experimental work involved 126 laboratory-prepared samples (grades M10–M40) with curing age of 14 day and 28 day and 231 field samples from a 20-year-old structure, tested using both methods. Total 357 no. of data samples were created with different mix proportion of design, curing ages and on-site environmental exposed concrete structure without unknown grade. Most of the researches were done while preparation of samples in the laboratory. For these purposes of taking mixing both variations such as control (Laboratory) and uncontrolled(on-site) samples were to prepare as a practical condition for prediction. For generation of predict model 70% data was used with methods such as regression analysis and Cascade forward back propagation neural network (CFBPNN) were used for investigation. To validate the prediction 30% data was used which was not used in model generation. The prediction results show that the coefficients of determination (R2) of the Regression analysis and the CFBPNN prediction models for the test set of concrete compressive strength are 95% and 99% respectively ANN model founded to be more accurate as compare to regression analysis. The validation by R2 of the Regression analysis and the CFBPNN prediction model for the compressive strength for above dataset was 89.0% and 98%. Statistical metrics (MSE, RMSE, MAPE) further confirmed the neural network’s superior accuracy.