Evaluation of Deterioration in Brick Materials in Monumental Buildings Using CNN Models

Abstract

The materials of historical structures undergo deterioration and deformation over time due to physical, chemical, and biological factors. Regular inspections by experts are essential for the preservation of these structures. However, the detection of such deteriorations involves significant labor, time, and cost, and incorrect diagnoses may lead to irreversible damage and structural issues. This study aims to minimize human-induced errors in identifying types of deterioration in structures by utilizing deep learning-based convolutional neural network (CNN) models, a subfield of artificial intelligence. With this study, deteriorations in historical brick buildings, which have an important place among historical buildings, will be detected with nondestructive methods, thus contributing to the preservation of historical buildings and the literature. Within the scope of the study, a total of 1709 data consisting of physical deterioration (cracks and fractures, joint discharge, abrasion and piece loss), chemical deterioration and biological deterioration types that are frequently encountered in brick materials in historical buildings were discussed. The classification process was carried out with the inputs given to the ResNet-18, ResNet-50, ResNet-101, VGG16 and VGG19 networks. Model performances were evaluated with precision, recall and F1 score metrics. The best performance values were obtained with ResNet101 (88% Precision, 88% Recall, 87% F1 Score, 88% Accuracy). Then, using Grad-CAM, the points on which the model focused while making predictions were determined. This study, which will include planning the basic principles of interventions to be applied to cultural property, will prevent the problems encountered in deterioration and deformation, and objective solutions will be produced.