Page 87 - Read Online

Wang et al. Intell Robot 2022;2(4):391-406 https://dx.doi.org/10.20517/ir.2022.25 Page 393

(2) Multisource heterogeneous data fusion representation of bridge cracks
Different kinds of data, such as the length of a bridge crack, the load, and the environmental humidity,
belong to multisource heterogeneous data.

From the level of description for the bridge crack damage state, it is difficult to describe the damage
evolution state of bridge cracks due to the large amount of multisource heterogeneous data. When bridge
cracks are initially generated, the impact on the bridge is small, and the state change of each indicator is not
obvious under the same conditions.

Due to the large variety, quantity, variation in sampling methods and many random interference factors of
the sensors, the quality of the acquired data is reduced, and the monitoring data are uncertain and low-
density, which makes it difficult for the collected data to accurately and pertinently reflect the evolution of
the bridge crack.

(3) Intelligent detection methods of bridge cracks
From the level of diagnosis for the bridge crack damage state, the characterization information of bridge
cracks is mixed and weak, so it is difficult to scientifically diagnose them. Due to the diversity of monitoring
equipment, the diversity of monitoring methods and the variability of monitoring locations, the monitoring
data show the characteristics of multimodal, strong correlation and high dimension, which makes the crack
information extremely complex, highly mixed and weak separability. It is difficult to analyze the monitoring
data and to scientifically describe and diagnose the cracks of concrete bridges. Additionally, the
deterioration trend of cracks cannot be scientifically predicted.

By analyzing the feature extraction method of crack multimodal data, the problem of difficult multimodal
parameter feature extraction due to the multiscale and diversity of characterization information can be
solved. By investigating the multisource heterogeneous data fusion representation of cracks, the difficulty in
describing the damage evolution state due to the large amount of multisource heterogeneous data can be
addressed. By examining the intelligent prediction model of crack deterioration trends, the problem of
difficult scientific diagnosis due to the overlapping and weak separability of bridge crack characterization
information can be solved.

The remainder of the article is organized as follows: Section 2 describes the common difficulties in research
on feature extraction, data fusion and the detection of concrete bridge cracks. The global research status
from three aspects is elaborated and analyzed in Section 3. Section 4 summarizes the principal concluding
remarks of the current research with a statement about its strengths and limitations. Figure 1 shows the
research framework.

2. RESEARCH STATUS
Research on concrete bridges mainly includes three key aspects: feature extraction of multimodal data,
fusion representation of multisource heterogeneous data, and intelligent detection models of bridge cracks.
The review of the global research status and development trends mainly focuses on these three aspects.

2.1. Research status of multimodal data feature extraction of bridge cracks
Although some nondestructive methods, including the ultrasonic pulse [24,25] and elastic wave [26-28] , have
achieved good results for bridge crack detection, feature extraction from the multimodal data of bridge
cracks remains challenging.