Page 96 - Read Online

Page 402 Wang et al. Intell Robot 2022;2(4):391-406 https://dx.doi.org/10.20517/ir.2022.25

Table 3. Intelligent detection methods
Research Method/approach Advantage Drawback
Wang et al. [61]
[62]
Yan et al. Machine learning Excellent predictive performance Unsuitable for large sample data
[63]
Liu et al.
[66]
Guo et al.
Broad learning system High accuracy and short time Unsuitable for sequential data
[67,68]
Xu et al.
Li et al. [69]
[70]
Islam et al.
Liang et al. [71] Deep learning Sufficient for deep mining big data information Long computing time
[72]
Li et al.

Figure 3. Structure of the deep neural network mapping model.

(length and width) of cracks, and research on three-dimensional cracks is limited. Although three-
dimensional crack reconstruction technology has gradually attracted researchers’ attention [73,74] , its evolution
mechanism remains ambiguous. There is an urgent need in the field of bridge disease detection to
investigate the law of crack evolution and to predict the trend of crack degradation.

In bridge crack research based on data, bridge health monitoring systems has an irreplaceable role. The
main purpose of research on the long-span bridge health monitoring system is to accumulate the design and
scientific research data of bridge health monitoring, realize the real-time damage diagnosis and safety
assessment of the structure, and support management and maintenance decision-making. Presently, the
challenges faced in bridge health monitoring mainly include decision-making and the design of health
monitoring systems, sensor signal preprocessing, and signal data noise reduction.

4. CONCLUSION
This paper presents a comprehensive review of recent advances in the field of data-driven bridge crack
health detection. The latest achievements in bridge crack feature extraction, data fusion and intelligent