Detection of structural and non-structural damages in reinforced concrete structures with deep learning

Earthquakes, an unpredictable and unpreventable natural event that can be considered the most dangerous of natural disasters, have caused serious losses in our country, which is located on fault lines, as in the whole world. After each earthquake that occurs, the importance of earthquake-resistant building design increases, and it is observed that serious damages occur in structures built without complying with the regulations in damage assessment studies, even in small earthquakes. While it is of great importance that the damage assessment studies carried out in all buildings damaged in the earthquake are carried out correctly, it is extremely important that the works are also fast and practical due to the danger of destruction of aftershocks in seriously damaged buildings. The limited number of experts and the large number of buildings damaged during the earthquake cause the damage assessment studies to be insufficient. In this case, people continue to live in damaged buildings unconsciously, which may lead to increased loss of life and property. One of the important decisions made in the damage assessment studies made after the earthquake is to differentiate the cracks that occur in reinforced concrete structures according to whether they are in the structural element or the non-structural element. Earthquake damages on structural elements, especially on columns and shear walls, show that the structure has been damaged significantly, while earthquake damages on non-structural elements do not significantly affect the use of the building.In this thesis, an artificial intelligence algorithm has been developed that can help experts during damage assessment studies and ensure that decisions are made accurately and quickly. In the study, the Deep Learning model, which is one of the sub-branches of artificial intelligence, which can make predictions on different data by learning by itself from the data, was preferred. Since the aim of this thesis study is to classify the damage images as damage to the structural element or damage to the non-structural element, the Convolutional Neural Networks model, which has been used in very successful studies in this field, has been utilized. The damage images obtained from the damage assessment studies carried out after the Istanbul (Silivri), Elazığ (Sivrice) and İzmir (Seferihisar) earthquakes with a magnitude of Mw ≥ 5 that occurred in our country in recent years were collected to develop an algorithm. The model, which was developed by selecting the optimum values, detected and classified the structural and non-structural element damages in reinforced concrete structures with a success rate of 93.13%. In addition to the fact that Deep Learning systems, which are still very new in the field of civil engineering, can be developed and used, it has been proven that the model developed in this study can help experts in damage assessment studies.