Prediction of concrete compressive strength using deep learning

Earthquake disasters in Turkey in recent years have brought the issue of building safety back to the agenda. The most effective way to construct reliable buildings is to use high-strength building materials. Among these materials, concrete is paramount. Concrete is a composite building material consisting of water, aggregate, cement, and various chemicals. Predicting the compressive strength of concrete is crucial for both building safety and cost efficiency. With advancing technology, traditional prediction methods have been replaced by artificial intelligence (AI) techniques, which can provide fast and reliable results. AI uses machine learning to analyze past data and generate useful predictions about the future. Machine learning models excel at finding linear patterns but often struggle with non-linear ones. A subfield of machine learning, deep learning is highly effective at identifying complex non- linear patterns. In this study, Convolutional Neural Network (CNN) algorithm, one of the deep learning techniques, is used to predict the 28-day compressive strength of conventional concrete made with CEM I type Portland cement. In the application section of the study, we performed sustainability analyses on a dataset from a concrete company. Statistical analyses were performed on a concrete company's dataset during the study's implementation phase. Randomized Search technique was used to find the best hyperparameter combination for the prediction model. The CNN model, developed using Python, takes the following inputs: Raw Material (Df) Avg. LSF, SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, Na2O Eq, 200µ, 90µ, 200µ.1, 90µ.1, Blaine, amount of water, water and cement ratio, specific gravity, and the 2-day (B2) and 7-day (B7) compressive strengths of the concrete. The model's output is the 28day concrete compressive strength (B28). The CNN model achieved strong results, with an R2 value of 0.84, a Root Mean Square Error (RMSE) of 1.72, and a Mean Absolute Error (MAE) of 1.39. The statistical analysis showed that the variables with the greatest influence on the 28-day concrete compressive strength were B7, B2, aggregate size (90µ.1), and MgO. The 28- day concrete compressive strength was also predicted using the same dataset, with machine learning algorithms such as Linear Regression, Decision Tree, Random Forest, and Support Vector Machines. Moreover, the 28-day concrete compressive strength was predicted using the same dataset, with machine learning algorithms such as Linear Regression, Decision Tree, Random Forest, and Support Vector Machines. The results showed that the CNN algorithm outperformed the other algorithms. In future studies, a hybrid approach that combines the CNN algorithm with other deep learning algorithms like Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU) will be utilized.