Prediction of compressive strengths of different concrete classes by artificial learning algorithms

With the production of ready-mixed concrete, the quality of concrete used in the construction sector is increasing day by day. The quality of concrete is directly related to its compressive strength and related tests are labour intensive and time consuming. Therefore, numerical simulations, regression analyses, artificial intelligence-based learning algorithms are used to predict the compressive strength of concrete. However, the non-linear and complex correlation between different types of materials in concrete makes it difficult to predict the compressive strength of concrete at the desired level. For this reason, artificial learning algorithms such as artificial neural networks, adaptive network-based fuzzy inference system, and Gradient Boosting (Extreme Gradient Boosting, Light Gradient Boosting, Categorical Boosting) are used extensively in research in construction technology. In the models, for each concrete class, samples were taken from 20 separate concrete pours prepared at a ready-mixed concrete plant and delivered to the construction site. Two samples were taken from these specimens and compressive strength tests were performed on the 7th and 28th hydration days. Of the total 480 data obtained, 70% for training and 30% for testing were used. For the reliability of the predicted results, R2, MAPE and RMSE statistical methods most frequently used in the literature were utilised. In addition, in order to compare the actual results with the predicted results, the average of all samples taken during casting was taken for each concrete class and hydration day separately. As a result, it was determined that the actual results obtained and the test data of all models were very close to each other. Accordingly, the compressive strengths of concrete classes C16/20, C20/25, C25/30, C30/37, C35/45 and C40/50 were determined for 7 days with an error between 0.930% and 0.116%, 1.886% and 0.562%, 1.194% and 0.582%, 2.071% and 1.251%, 1.170% and 0.278% and 1. 275% to 0.360%, 1.896% to 0.927%, 1.450% to 0.364%, 1.030% to 0.699%, 0.749% to 0.050%, 0.727% to 0.199% and 0.808% to 0.774% for 28 days, respectively. Therefore, it is concluded that there is a good agreement between the results obtained from the experiments and the prediction results and that the compressive strengths can be predicted with "high accuracy" or "very good" with all the models created.