Crack Width Detection for Concrete Structures Using Thermal Images and Artificial Neural Network

Detection of concrete element cracks and their properties is essential to maintain the health of structures, as cracks affect the durability and serviceability of concrete structures. Several inspection techniques were developed to assess the performance of concrete structures. These techniques include visual inspection and non-destructive testing techniques such as ground penetrating radar, ultrasonic testing, and infrared thermography (IRT). In addition, several techniques were developed to measure crack properties, including image-based techniques remotely. Therefore, this research proposed a crack width detection method that can be used on different concrete structures. Moreover, this method uses thermal images and develops artificial neural network (ANN) models to predict the crack width. The method presented a simple technique to detect crack properties remotely and without any disturbance. Three models with different numbers of inputs (one input, two inputs, and three inputs) were tested under eight different neural network architectures. The one input model that linked the crack temperature (as an input) to the crack width failed to detect the crack widths accurately for the different points in the testing sets. In contrast, in the two input models that were used in addition to the crack temperature, the average intact region temperature in predicting the crack width outperformed the three inputs model that was used as a third input in the form of binary variable to indicate if the concrete structure contains fiber or not. The results indicated that the ANN provided a good crack properties detection technique using the crack temperature and intact region temperature.