The degradation of infrastructures such as bridges, highways, buildings, and dams has accelerated due to environmental and loading consequences. The most popular method for inspecting existing concrete structures has been visual inspection. Inspectors assess defects visually based on their engineering expertise, competence, and experience. This method, however, is subjective, tiresome, time-consuming, and constrained by the requirement for access to multiple components of complex structures. The angle, width, and length of the crack allow investigators to figure out the cause of the propagation and extent of the damage, and rehabilitation can be suggested based on that. This research proposes an algorithm based on a pre-trained convolutional neural network (CNN) and image processing to find the crack's angle, width, endpoint length, and actual path length in a concrete structure. The results show low relative errors of 2.19%, 14.88%, and 1.11% for the crack's angle, width, and endpoint length.
Top1. Introduction
According to a survey taken in 2010, National Highway Authority own about 5000 bridges; about 30% percent of these bridges have already reached their design or require rehabilitation (Qureshi, 2014). Similarly, in the USA, according to Federal Highway Administration (FWHA), 40% of the 5,7000 bridges are defiant and need maintenance (Nowak, 2012). Civil engineering structures, including bridges, buildings, tunnels, etc., are prone to further damage due to catastrophic events such as earthquakes, wind, explosions, and temperature variations. These damages mainly include cracks, efflorescence, scaling, and spalling. As a result, various structural health monitoring (SHM) systems for detecting, identifying, and monitoring such degradation have been proposed. Different types of concrete can appear in the concrete. Mainly it includes construction cracks, shrinkage cracks, despair cracks, plastic settlement cracks, crazing cracks, and chemical cracks. Many studies have been made on classifying concrete surfaces between cracked and uncracked. Few studies have been conducted to find the characteristic of cracks, i.e., endpoint length, actual path length, angle, and width of the crack. Based on these factors, the extent of the damage can be measured, and the reasons for the crack propagation can be figured out. The angle, width, and length of a crack in a concrete structure are all essential factors that can impact the strength and stability of the structure. The crack angle can affect the stress distribution in the concrete structure, which can impact the structure's overall stability and load-bearing capacity. For example, a crack with a vertical angle is more likely to cause vertical separation of the concrete and is therefore considered more severe than a crack with a horizontal angle. The width of the crack can provide information about the severity of the damage and the stress that the crack is under. The width of the cracks indicates a greater amount of damage and can impact the structural integrity of the concrete. In general, wider cracks are considered more severe than narrower cracks. The crack's length can impact the concrete structure's stability and load-bearing capacity. Longer cracks can weaken the concrete and may result in the structure collapsing. In general, longer cracks are considered more severe than shorter cracks. It is vital to accurately assess the angle, width, and length of cracks in concrete structures to determine the appropriate repair method and ensure the structure's long-term stability and safety.
CNN is a type of image processing tool that uses deep learning (DL) algorithms to achieve mainly three types of tasks, firstly classification of the object or image, secondly the detection of the object in such a manner that the boundary box, and thirdly the segmentation of the image in which groups of pixels are separated of the image. To extract the object from the image, different techniques are widely adopted. One of them is using morphology. The term “morphology” refers to various image-processing techniques that analyze and manipulate pictures based on their geometrical forms. Morphological operations use an input picture template to generate an identically sized new image. Each pixel in the final picture is assigned a value determined by how the input image's relevant pixel is compared to its neighbours.