Predicting compressive strength of AAC blocks through machine learning advancements

Abstract

Determining the strength properties of Autoclaved Aerated Concrete (AAC) through conventional compression experiments is both time-consuming and costly. Using sophisticated Machine Learning (ML) algorithms to forecast concrete compressive strength can expedite time-consuming experimental procedures and reduce expenses. In this study, four ML models were proposed, including Random Forest (RF), Support Vector Regression (SVR), Linear Regression (LR), and Stochastic Gradient Descent (SGD). These models were developed to forecast the compressive strength of AAC blocks based on a dataset of 525 cubic samples. By comparing the results using different evaluation indices, the study analyzed each input variable’s relative importance and impact on the output. The findings revealed that the SVR model had the least error and is thus the most suitable for concrete compressive strength estimation. This approach results in cost savings on both specimens and laboratory tests. Out of the seven input factors, which encompass the proportions of water, cement, sand, lime, fly ash, aluminum powder, and gypsum, the proportions of cement and water content were pinpointed as the most crucial characteristics. In contrast, aluminum powder and gypsum displayed less prominent significance.