Detection of building physics problems with convolutional neural networks

Buildings are produced to fulfill the functions needed for many years to be used healthily. It is of great importance to solve the structural physics damages that occur in their structures due to different reasons over time, for the structures to function properly and for sustainability. The timely and accurate detection of damages is the most important step in solving the problems. This situation directly impacts the smooth progress of the repair works to be carried out in the later stages and the costs of the repair. There are different approaches to detecting building physics problems, trying to detect them with the naked eye involves subjective judgments, and destructive methods damage the structure. This thesis aims to objectively and non-destructively detect deteriorations, cracks and moisture damages from structural physics damages by using the digital image processing method, which is one of the artificial intelligence applications. A total of building 7200 image data, 5400 of which were taken from the surfaces of damaged samples, and 1800 of them were taken from undamaged, were subjected to machine learning using a convolutional artificial neural network, and the test data were classified. The application studies were expanded by changing the parameters. It has been observed that structural physics damages can be detected with an accuracy rate of 95,21% with the use of Convolutional Neural Networks with the most appropriate values determined. As a result, it has been seen that image processing applications can be used in the detection of structural physics damages and provide high accuracy for different problems. This thesis study reveals that image processing, one of the artificial intelligence applications used in various fields today, can be used to detect structural physics problems.