Classification of Porosity in Sulfur-Based Concrete Samples Using Deep Neural Networks

This study investigates the application of deep learning and machine learning techniques for the classification of sulfur-based concrete samples based on porosity, which is an important property that affects the strength, durability, and permeability of concrete. The first part of the research focused on creating a unique dataset of sulfur-based concrete samples and calculating features such as porosity. The percentage porosity was then calculated, and images were labeled as low porosity or high porosity based on the percentage of porosity. The images of physical samples were automatically annotated by image processing techniques to create a dataset. The second part of the study aimed to train and test a neural network to predict and classify samples based on porosity. We classified concrete images into two separate classes of low and high porosity using a basic Convolutional Neural Network (CNN) and transfer learning with a pre-trained model such as AlexNet. Porosity was calculated as the distribution of air voids and aggregates through the concrete sample. The comparison of two of the best models and finding the accuracy and other performance metrices of the networks were done using confusion matrices. The conclusion of this study shows that pre-trained models with transfer learning, such as AlexNet, can be used to accurately and automatically classify sulfur-based concrete samples based on porosity, which could lead to faster and more efficient quality control of concrete production. This study also sets the stage for further research into the application of artificial intelligence methods in the field of civil engineering, as it offers a new method for classifying and predicting the characteristics of construction materials such as concrete. In future studies, the dataset created in this study can also be used for regression analyses.