Concrete Crack Detection Using Thermal Imaging and Neural Network

In the field of building integrity testing, the structural integrity of concrete structures can be adversely affected by various impact actions, such as conflict and warfare. These actions can result in subsurface defects that compromise the safety of the buildings, even if the impacts are indirect. However, detecting and assessing these hidden defects typically requires significant time and expert knowledge. Currently, there is a lack of techniques that allow for rapid evaluation of usability and safety without the need for expert intervention. This study proposes a non-contact method for testing the integrity of structures, utilising the unique characteristics of thermography and Deep Learning. By leveraging these technologies, hidden defects in concrete structures can be detected. The Deep Learning model used in this study is based on the pretrained ResNet50 model, which was fine-tuned using simulated data. It achieved an impressive overall accuracy of 99.93% in classifying defected concrete blocks. The training process involved two types of thermograms. The first type consisted of simulated concrete blocks that were heated and subjected to pressure. The second type involved real concrete blocks from the lab, which were subjected to pressure using a pressure machine.
Laboratory experiments involving the compression of concrete blocks and thermal imaging yielded a dataset used to train a new model with the same architecture and hyperparameters, resulting in a validation accuracy of 100%. This investigation demonstrates that AI can effectively classify thermal images of concrete surfaces, enabling the detection of subsurface cracks and hidden defects.