Multistage semisupervised active learning framework for crack identification, segmentation, and measurement of bridges

Abstract

In bridge health monitoring (BHM), crack identification and width measurement are two of the most important indices for evaluating the functionality of bridges. In order to reduce the labor cost in field detection, researchers have proposed a variety of deep learning (DL)-based detection techniques for crack recognition. However, some problems still exist in extending these techniques to practical applications, such as data annotation difficulty, limited model generalization ability, and inaccuracy of the DL identification of the actual crack width measurement. In this paper, an application-oriented multistage crack recognition framework is proposed, namely, Convolutional Active Learning Identification-Segmentation-Measurement (CAL-ISM). It includes four steps:(1) pretraining of the benchmark classification model, (2) retraining of the semisupervised active learning model, (3) pixel-level crack segmentation, and (4) crack width measurement. Beyond numerical experiments, the performance of the CAL-ISM is validated for practical applications: (i) bridge column test specimen and (ii) field BHM projects. In conclusion, the obtained results from these applications shed light on the high potential of CAL-ISM for BHM applications, which is recommended in future deployments for BHM. BACKGROUND AND MOTIVATIONSBy the end of 2020, there are 912,800 highway bridges in China, including 6444 long-span bridges and 119,935 normal bridges. The total mileage of the highway is about 5.19 million kilometers (km), of which the maintenance mileage of the highway is 5.14 million km, accounting for 99.0% of the total mileage (Ministry of transport of the People's Republic of China, 2020). The remaining 1% is for newly constructed highways with no need for immediate © 2022 Computer-Aided Civil and Infrastructure Engineering. maintenance. Similarly, there are more than 618,000 bridges in the United States, where nearly 36% of the bridges need repair work, and 7.3% of them are considered structurally deficient (American Road and Transportation Builders Association, 2021). Therefore, surging demands in the maintenance work of highway bridges in China, the United States, and many other countries are worldwide realities and it is a necessity to develop effective methods to evaluate the service functionalities of bridge structures. Because of cost-efficiency and easy-forming, reinforced

Efficient data dimensionality reduction method for improving road crack classification algorithms

“…Crack detection is a step to be taken in a previous step in crack classification; however, they require different models. Several recent proposals based on DL were presented, however, mostly for concrete surfaces (Çelik & König 2022; Chun et al., 2021; Jang et al., 2021; Kong et al., 2021; Liu & Xu 2022; Xie et al., 2022; Zheng et al., 2022). Regarding road or pavement cracks, an image‐to‐image translation was presented for night images in Liu and Xu (2022), and finally J. Chen and He (2022) presented a neural network (NN) for the detection of four types of pavement cracks.…”

A vision monitoring system for multipoint deflection of large‐span bridge based on camera networking

“…Several structural control methods and systems have been proposed to reduce the effect of dynamic load (e.g., traffic, wind, and earthquake) on a bridge (Adeli & Saleh, 1997; Gutierrez Soto & Adeli, 2019; Kim & Adeli, 2005). Meanwhile, large efforts have been devoted to the structural health monitoring, including but not limited to the defect detection (Chun et al., 2022; Sajedi & Liang, 2021; Zhang & Lin, 2022; Zheng et al., 2022), stress or force estimation (Lee & Park, 2011; Tian et al., 2021), and strain or displacement measurement (Hampshire & Adeli, 2000; Park et al., 2007; Yu et al., 2022). As the most direct embodiment of structural health, the deflection of large‐span bridge is an important index for the evaluation of bridge safety and applicability, which can be further used for the bridge properties identification (Amezquita‐Sanchez & Adeli, 2019; Amezquita‐Sanchez, Park, & Adeli, 2017; Perez‐Ramirez et al., 2016; Pezeshki et al., 2023), structural response prediction (Perez‐Ramirez et al., 2019; Ren et al., 2021), and early warning (Bao et al., 2019; Li et al., 2015; Oh et al., 2017; Sousa Tome et al., 2020).…”

A decoupled algorithm for cable parameter identification by frequency‐ratio‐offset

“…Structural safety has always been the most concerned issue of civil engineers and scholars. In recent years, thanks to the rapid progress of various data acquisition technologies (Zhao et al, 2022), signal processing methods Qu et al, 2021), and artificial intelligence algorithms (Rafiei & Adeli, 2017;Zheng et al, 2022), the technology of structural health monitoring has or implicit frequency-tension relationship, which requires reasonable theoretical models. Scholars and engineers have struggled with these issues for decades.…”