STUDY ON THE APPLICATION OF DEEP
LEARNING TECHNOLOGY AND BIM MODEL
IN THE QUALITY MANAGEMENT OF BRIDGE
DESIGN AND CONSTRUCTION STAGE
Weiqi Zhu
Department of Economics and Management, Jiangyin Institute of Technology,
Jiangyin, Jiangsu, 214405, China
zhu20218882021@163.com
Reception: 28/02/2023 Acceptance: 19/04/2023 Publication: 06/05/2023
Suggested citation:
Zhu, W. (2023). Study on the application of deep learning technology and
BIM model in the quality management of bridge design and construction
stage. 3C TIC. Cuadernos de desarrollo aplicados a las TIC, 12(2), 137-154.
https://doi.org/10.17993/3ctic.2023.122.137-154
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
137
ABSTRACT
The development of the transportation industry can effectively accelerate the speed of
economic development, in which bridges occupy an important position in
transportation. The safety of the bridge design and construction process is a key part
of bridge construction, and relying on human resources to investigate safety hazards
greatly affects efficiency. In this paper, we combine deep learning technology and BIM
model to explore the synergistic effect of both on the quality management of bridge
construction phase and analyze the measured data. The results show that the
application of BIM model can improve the efficiency by 35% compared with the
traditional 2D CAD drawings, and the accuracy of data analysis can be improved by
12.51% and 14.26% for DNN and DBN models based on deep learning, respectively.
The addition of the GSO algorithm leads to a further 19.19% improvement in the
training accuracy of the coupled model. Finally, the optimization model was used to
analyze the load factors and force majeure factors that affect the safety of the bridge,
and to find the structural factors that affect the safety of the bridge design, which
provides guidance to ensure the quality of the bridge during the construction process.
KEYWORDS
BIM model; CATIA modeling; deep learning; bridge construction; quality and safety
INDEX
ABSTRACT
KEYWORDS
1. INTRODUCTION
2. DESIGN OF THE MODEL
2.1. Parametric modeling of the bridge structure
2.2. Deep learning model
3. ENGINEERING APPLICATION RESEARCH
3.1. Specific application of the BIM model in the design and construction of bridge
engineering
3.2. Specific application of deep learning technology in bridge engineering quality
monitoring
4. RESULTS AND DISCUSSION
4.1. Analysis of test results
4.2. Data validity analysis
4.3. Structural safety analysis
5. CONCLUSION
REFERENCES
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
138
ABSTRACT
The development of the transportation industry can effectively accelerate the speed of
economic development, in which bridges occupy an important position in
transportation. The safety of the bridge design and construction process is a key part
of bridge construction, and relying on human resources to investigate safety hazards
greatly affects efficiency. In this paper, we combine deep learning technology and BIM
model to explore the synergistic effect of both on the quality management of bridge
construction phase and analyze the measured data. The results show that the
application of BIM model can improve the efficiency by 35% compared with the
traditional 2D CAD drawings, and the accuracy of data analysis can be improved by
12.51% and 14.26% for DNN and DBN models based on deep learning, respectively.
The addition of the GSO algorithm leads to a further 19.19% improvement in the
training accuracy of the coupled model. Finally, the optimization model was used to
analyze the load factors and force majeure factors that affect the safety of the bridge,
and to find the structural factors that affect the safety of the bridge design, which
provides guidance to ensure the quality of the bridge during the construction process.
KEYWORDS
BIM model; CATIA modeling; deep learning; bridge construction; quality and safety
INDEX
ABSTRACT
KEYWORDS
1. INTRODUCTION
2. DESIGN OF THE MODEL
2.1. Parametric modeling of the bridge structure
2.2. Deep learning model
3. ENGINEERING APPLICATION RESEARCH
3.1. Specific application of the BIM model in the design and construction of bridge
engineering
3.2. Specific application of deep learning technology in bridge engineering quality
monitoring
4. RESULTS AND DISCUSSION
4.1. Analysis of test results
4.2. Data validity analysis
4.3. Structural safety analysis
5. CONCLUSION
REFERENCES
https://doi.org/10.17993/3ctic.2023.122.137-154
1. INTRODUCTION
With the continuous development of China's economy, transportation modes are
showing a trend of diversification. The establishment of a well-connected
transportation network is of great importance for the development of national
transportation, the promotion of inter-regional economic exchanges and the flow of
talents [1,2]. Among them, bridge construction is the core link in the transportation
network, which can effectively solve the problem of inconvenient traffic on both sides
of the river basin. The safety of bridges is a key consideration in the design and
construction process of bridges, which can cause very serious traffic accidents if the
quality is not up to par [3-5]. Safety problems not only cause national economic
losses, but also have a destructive effect on the ecological environment to some
extent [6-8]. The safety issues of bridges should have a preliminary prediction during
the construction phase and also throughout the construction phase of the project,
where data from all aspects of bridge construction should be collected, processed and
analyzed. Bridge quality and safety monitoring mainly include two aspects of data
acquisition as well as safety index evaluation [9,10]. BIM technology is a technology
that integrates the data in the construction process through a building information
model. By integrating a large amount of data and information, it can read the key
information and realize data interoperability. It enables information transfer and
resource sharing in the pre-project preparation, process implementation and quality
control stages at the end of the project. It plays an important role in quality inspection,
safety management, budgeting, and progress monitoring of the construction process
[11-14]. BIM technology can effectively improve efficiency, increase calculation
accuracy, shorten construction cycle time and scientifically maintain equipment.
Neural network deep learning is a modern tool for data interpretation and result
prediction, which can quickly read information and extract data feature values, and
input calculation results after training with embedded algorithms. This learning
approach is currently combined with various fields, and the use of a deep learning
approach can help us to quickly make predictions about the results and greatly
improve efficiency [15-19]. If the deep learning approach is applied to the quality
inspection in the bridge construction process, not only the rate of problem solving can
be improved but also the accuracy of calculation results will be enhanced. Currently,
some scholars have conducted studies on the use of BIM technology and deep
learning methods in bridge construction and obtained desirable results [20-24].
Pan [25] et al. investigated a clustering-based log mining method for building
information modeling (BIM), while combining a novel clustering algorithm with an
efficient fuzzy Kohonen clustering network (EFKCN) to classify information with
different characteristics. The data were analyzed using regression prediction method
and the results showed that the model can be better for model building. D Forgues et
al [26] used the BIM model to reduce the cost of project completion and shorten the
project cycle by classifying the data through linear regression. The results showed
that the use of the BIM model can effectively discover the causes affecting the quality
of bridge construction and improve efficiency. R Edirisinghe [27] developed a safety
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
139
life cycle BIM prediction model with high maturity and analyzed the factors affecting
the model by data from real cases. As a result, five main factors affecting the BIM
model were found, and the model was improved to propose a life cycle BIM maturity
model (LCBMM) supported by actual data. Liang et al [28] proposed a novel deep
learning model based on a three-stage image training strategy to analyze the bridge
design structure, and in order to train and analyze the data effectively focused on the
model's robustness of the model was focused on in order to train and analyze the
data effectively. The results show that the robustness of the model is very good and
the accuracy of the prediction is more than 90%. T Abbas [29] et al. used a neural
network (ANN) model to predict and analyze the aerodynamic phenomena around the
bridge design process. In the paper, the neural network and the structural model were
combined and trained on bridge data with different interfaces and different geometric
features. The results show that artificial neural networks can predict the bridge
construction process more accurately, and this method can provide ideas for the
design of bridges with larger spans. Qz A [30] proposed a vision-based method for the
detection of cracks in concrete bridge decks, which is a one-dimensional convolutional
neural network (1D-CNN) and long short-term memory (LSTM) method in the image
frequency domain. The method is trained using a large number of cracked or
uncracked bridge deck images with high efficiency and accuracy. The results show
that the developed model can reach 99.25% accuracy with respect to the test data.
Moreover, the 1D-CNN-LSTM model can effectively reduce the computation time
compared with other neural network deep learning approaches. Ma [31] et al.
proposed a data-driven method for strain data detection and differentiation of vehicles
for detecting vehicle operation in large-span bridges. A neural network deep learning
approach is used to track and identify vehicles to ensure traffic safety on bridge
pavements. The results show that this detection method is relatively robust and
accurate and is able to predict traffic conditions well despite the presence of noise.
Dinh K [32] et al. proposed a coupled algorithm combining traditional image
processing techniques and deep convolutional neural networks for the localization and
detection of steel reinforcement during the construction of bridge construction. The
images are first processed by offset and normalization methods for locating the pixels
containing potential rebar peaks. The results obtained in the first step were then
classified by a convolutional neural network (CNN) and a total of 26 bridge data were
analyzed. The results showed that the average accuracy of the model's calculation
results exceeded 97.75%, and the overall accuracy of the whole bridge test was about
99.60%. Kim [33] et al. proposed a vision sensor-based UAV bridge inspection
method to troubleshoot the deterioration of bridges over a long period of time to
ensure the quality and safety issues of bridges. The test first used a UAV to fly around
and obtain a point cloud-based background model. A regional convolutional neural
network (R-CNN) model was then used to detect the crack structure on the bridge
surface and calculate the thickness and length of the cracks. A new network is
generated from the pre-trained network and used to collect 384 crack images with 256
×
256 pixel resolution. The results show that the model is highly accurate in the
identification and detection of bridge quality. Yang [34] et al. proposed a deep learning
model to evaluate the stability and safety of bridge structures, and the model maps a
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
140
network of large data into a very small volume of eigenspheres. Where the data in the
spheres are normal values and the abnormal data are outside the eigenspheres. The
results show that this model learning approach is effective and the computed practical
results are superior compared to other advanced methods. In summary, the
application of BIM technology and deep learning neural networks for monitoring the
management of safety issues in bridge construction has ideal results in terms of both
data resources and access integration and project outcome prediction. However, the
two technologies are separate in the research work so far, and the combined use is
relatively rare.
Therefore, this paper explores the synergistic effect of coupled models in the
application of bridge construction stage quality management based on the BIM model,
combined with a deep learning approach. Firstly, the bridge structure is parametrically
modeled by CATIA software, and the potential problems in the bridge construction
process are identified through the BIM model. Then the information is input into the
neural network through the input layer, while comparing the deep learning approach
with the traditional calculation method, and evaluating the accuracy of the deep
learning approach according to the calculation results. Further, the GSO algorithm is
used to optimize the deep learning method, and the validity of the computational
results of the optimized algorithm is analyzed. Finally, the structural factors affecting
safety and stability, including load factors and force majeure factors, are analyzed by
this optimization algorithm. It provides an idea for the bridge construction process
monitoring and the bridge quality and safety prediction.
2. DESIGN OF THE MODEL
2.1. PARAMETRIC MODELING OF THE BRIDGE
STRUCTURE
At present, BIM models are more widely used in all stages of engineering design,
construction and maintenance, among which the application in the transportation
industry is the most. Various urban transportation fields in China have made
application requirements for the application of BIM technology, and strict standards
have been set for the accuracy of the results. The standardized regulations are
important to ensure the construction quality and safety of the project, and all aspects
of bridge design should be carried out in strict accordance with the standards. the key
aspect of BIM application lies in the construction of the initial model, and the efficiency
and quality of the modeling directly affect the project. CATIA software is a good
modeling tool, which is a software developed by Dassault Systèmes, France, mainly
used for the construction of mechanical structure models. the most important feature
of CATIA is that it can model according to parametric spatial points, spatial curves and
surface features, and has a good effect on the parametric modeling of large bridges
(arch bridges, T-bridges, etc.).
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
141
CATIA is an integrated software with CAD/CAE and CAM, which can provide a set
of mature technical solutions from product design and final product landing. CATIA
has the functions of 3D parametric design, dimensional constraints, parametric
modification, etc. The specific connotations of parametric include the following.
1.
Custom parameters: Users can input specific parameters according to their
actual needs during the modeling process, and there are various types of
parameters to meet different types of customers.
2.
All-round dimensional constraints: This is the advantage of CATIA.
Dimensional constraints refer to the specific constraints on the dimensions of
each drawing that are mandatory for users in the modeling process. If the
markup is missed, the modeling process cannot proceed. Also, the user has to
give a global dimension so that the accuracy can be effectively controlled
during the modeling process.
3.
Dimensional parameter modification: In the process of parametric modeling,
there must be constraints on the overall dimensions. But the detail part may
exist with the subject management, the size can not be completely determined.
CATIA can give the drive size modification function to this part, to achieve size
modification and change.
4.
Structural logic: In the design process respect the top-to-bottom concept,
specifically expressed as a loop. In the setting or modification link soft solutions
are made to record, the user can check or modify again.
5. Standardized design: For the frequently used structure, the software provides a
parametric template. That is, the software can save this part of the design in
the inventory, and the same is defined in accordance with the standard size.
Users can transfer in and out from the library in the process of use, to facilitate
the work of users.
2.2. DEEP LEARNING MODEL
In recent years, deep learning (DL) and artificial intelligence have been very closely
integrated and have penetrated into many industrial fields. DL has unique advantages
in extracting feature values from big data and processing data, and therefore has a
wide range of applications in computer systems, speech idiosyncratic recognition and
language expression. The input is generally located at the lower level. The input
transmits the data further to a higher layer. The layers of transmission finally reach the
output layer, which is the highest layer. It is important to note that the data is mined
step by step during the transfer process, and finally, the data is obtained with some
distribution pattern. The bridge structure is composed of many different small parts,
and each part is related. Therefore, many details need to be considered when
evaluating construction safety and quality, and the amount of data is relatively large.
To address this problem this paper uses the DL model for analysis, and adds the deep
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
142
neural network (DNN) model and the deep belief network (DBN) model to the DL
model, aiming to further improve the accuracy of the model prediction results. Which
can be transformed into an output matrix when extracting the feature values, as
shown in Equation 1.
(1)
Where denotes the fault information and denotes the fault parameters. By
constructing a mixture model about the features, the distribution expression of the
features of the faults can be obtained as follows.
(2)
Where denotes the frequency of the fault wave and the subscript expresses
the number of faults. According to the deep learning method, the dynamic
components of the faults can be obtained by the distribution probabilities of the visible
and implicit layers corresponding to the initial detection and model reconstruction of
the bridge as shown in the following equation.
(3)
Where denotes the fusion transfer parameter containing the fault features, and
based on the above-extracted eigenvalues, combined with the spectral analysis, the
density component of the monitoring fault phase coupling is obtained as shown in the
following equation.
(4)
Where represents the peak of the bridge fault state. The joint analysis method is
applied to it, and the expression of energy distribution can be obtained as follows:
(5)
Where denotes the dimensionless parameter for equipment failure in the bridge
monitoring process, and analysis of the data yields the output of automatic fault
tolerance for bridge equipment failure monitoring as:
R
=
r
1
r
2
r
1,n
r1,1 r2,2 r2,n
rn,1 rn,2 rk,n
rnk
R
i=
i=1
R+r
n,k
mi
mi
i
T
=
R
i
(n+1)+R
mi
n
R
x=
vx
T+Ri
×
i=1
R
vx
M
=
N
x
Ri+Rx
+
i=1
R
Nx
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
143
(6)
From the above analysis, it can be seen that when a bridge fails, the extraction of
fault feature values can help engineers quickly determine the cause of the failure and
find the location of the failure. The bridge fault tolerance model constructed can
improve the fault tolerance of the device.
DNN is an artificial neural network composed of the input layer, implicit layer and
output layer, DNN is mainly through the learning behavior to solve the mapping
relationship between the input layer and output layer. DNNs can achieve more
satisfactory results for both feature value extraction and computational result
prediction.
DBN is a generative model with multiple hidden layers because it is composed of
multiple restricted Boltzmann machines (RBMs) cumulatively, which is a model with
random probability distribution consisting of a set of visible units as well as hidden
units. DBN is able to retain data features and on this basis, the computational process
can be simplified by reducing the dimensionality as much as possible. Therefore, the
effective learning method of DBN can be thought of as reducing the complex model to
a combination of many simple models, and learning the input parameters by passing
them in layers. The DBN is more accommodating and can input different kinds of
data. And the data transfer is iterative, i.e., the previous data calculation result will be
used as the next input data, so DBN has an efficient learning method as well as
scientific classification performance.
3. ENGINEERING APPLICATION RESEARCH
Deep learning technology and BIM models have a role in the quality management
of bridge design and construction phases that should not be underestimated. Among
them, BIM technology and the CATIA software module available in recent years have
a guiding role in the design of the bridge in the early stage and the subsequent
construction problem ranking, which can reduce the manual input. Deep learning
technology, on the other hand, can be used to troubleshoot bridge quality problems
that occur during the construction phase and to monitor the bridge quality monitoring
system, which can provide statistical analysis efficiency by excluding invalid data. This
section will detail the specific application of the BIM model and deep learning
approach in bridge construction.
3.1. SPECIFIC APPLICATION OF THE BIM MODEL IN THE
DESIGN AND CONSTRUCTION OF BRIDGE
ENGINEERING
G
= 2T+
N
g
+N
x
mi
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
144
For the bridge project, the bridge design and construction and bridge engineering
feasibility links, through the application of BIM technology, can more effectively
accelerate the progress of the bridge project design. the advantages of BIM
technology applied to bridge design are mainly the following two points, one is that
through the use of the technology, you can check out in advance whether there are
defects and deficiencies in the project, timely improvement and processing, for the
later bridge engineering project The first is that by using this technology, the defects
and shortcomings of the project can be identified in advance, and timely
improvements can be made to bring more convenience to the later bridge projects.
The second is that the application of BIM technology helps designers and constructors
to have a deeper understanding of the project, and can accurately analyze and judge
the different cost situations involved in the project to facilitate further reduction of
bridge project costs. the main contents of the four main steps of BIM technology
application are:
1.
In the design and construction of bridge projects, the first step of BIM
technology application requires that the designer should continuously improve
the model according to the requirements of different periods and the actual
needs of the project so as to ensure the accuracy of the model. The
construction process mainly has the following three steps: the first is to
establish the relevant parameter library, the second is to build a scientific and
accurate model, and the third is to reasonably set the corresponding
reinforcement module. Among them, the establishment of the relevant
parameter library and the next step of building the model can use Dassault's
CATIA software to achieve parametric modeling. CATIA 3D model can pass
management object data, index data, etc. to the construction stage, which can
make the modeling process more convenient with higher accuracy.
2.
After constructing an accurate parametric model of the bridge project in the
early stage, to develop a perfect construction strategy, advanced BIM
technology needs to be further applied. By means of simulation, the link is
clearly presented, and the construction differences are compared, so that the
simulated construction process is visualized and dynamic. This step will enable
the designer to make a more scientific and accurate judgment on the safety
and economy of the bridge structure.
3.
Volume statistics, refers to the application of BIM technology application
statistics to obtain accurate volume calculation values in the bridge engineering
design period. Compared with the traditional method of calculating the points,
lines and surfaces of the two-dimensional plane, the calculation task is
completed with Excel tables. Using BIM technology, in building an accurate
three-dimensional model of the bridge, it is possible to scientifically select the
design components and suitable materials, and use the automated
measurement function to achieve the purpose of determining the bridge
volume. This method, in addition to reducing the calculation time, improves the
accuracy of calculation by about 35% on average and saves the energy input
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
145
of design staff. In addition, for the modeling of complex curves under linear
engineering, it is also possible to accurately determine the twisted part and
accurately measure the length, area and volume of the bridge by using the
CATIA software in BIM technology.
4.
In the design and construction of bridge projects, after completing the above
work, Navisworks and other related software can be further used. For the
construction of the BIM model for collision analysis, to complete the preliminary
check, proofreading and audit and other different work. At the same time, it
helps to find out whether there are defects and deficiencies in the construction
drawings, and further correct them. In addition, the use of BIM technology can
also achieve the task of comparing procurement lists and determining what
needs to be procured.
3.2. SPECIFIC APPLICATION OF DEEP LEARNING
TECHNOLOGY IN BRIDGE ENGINEERING QUALITY
MONITORING
It is necessary to monitor the health system data in the design and construction of
bridges. Compared to most traditional test and analysis methods that rely on statistical
theory and require extensive domain knowledge, monitoring systems based on deep
learning techniques are more suitable for large-scale data sets. The main work of the
deep learning technique monitoring system is to perform health monitoring and data
characterization of bridge structures. Its main purpose is to study the different
distribution characteristics of the data for subsequent processing of the data. One of
the bridge quality monitoring devices for monitoring bridge structure data is an
important device for monitoring faults and status, and the reliability of bridge quality
monitoring devices is to be ensured in performing bridge quality monitoring. However,
the reliability of the bridge quality monitoring device cannot be guaranteed due to the
bridge's own structural factors that easily affect the bridge quality monitoring device. In
order to guarantee the reliability of monitoring data, this paper proposes an intelligent
bridge quality monitoring device fault tolerance system automatically based on deep
learning technology, and carries out hardware design and application testing.
Establish the communication module of bridge quality monitoring equipment in the
fault-tolerant system of the upper computer, and carry out the interface interaction
design of bridge quality monitoring equipment fault judgment through reset control and
Internet networking control technology. Set the dual ports as RAM, apply the Internet
of Things networking technology to obtain the PCI protocol of the dual ports, and
obtain the bus control parameter analysis model for fault-tolerant judgment of bridge
quality monitoring equipment according to the PCI protocol control signal. Set the
operating main frequency of the system to 180MHz/MIPS, and carry out the joint
multi-channel control of bridge quality monitoring equipment fault judgment by PCI
protocol. The bridge quality monitoring sensor module with data bus set to LD[16:0]
signal is used for VXI transmission of bridge quality monitoring equipment fault
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
146
tolerance system automatically to build a network monitoring model for bridge quality
monitoring equipment fault tolerance judgment. To reduce the error of fault
information, the register initialization process is carried out, and finally, the monitoring
information of bridge quality monitoring equipment fault automatic fault tolerance
system sends data to the main control computer through the bus.
4. RESULTS AND DISCUSSION
4.1. ANALYSIS OF TEST RESULTS
A part of the data from the China Construction Project Database was used as the
experimental data set for the experimental testing of the bridge quality monitoring
device fault automatic fault tolerance system designed based on deep learning. The
output curve of fault feature monitoring of the bridge quality monitoring device is
shown in Figure 1.
Figure 1. Monitoring output amplitude change curve
As can be seen from Figure 1, the deep learning technique-based test method for
bridge quality monitoring equipment fault monitoring has a high level of sample fusion.
In addition, compared with the traditional test method without troubleshooting, the
highest output value of the deep learning technique test is 28.64% higher in the first
125S and 69.97% higher in the second 250S, which greatly improves the fault
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
147
tolerance level of the bridge quality monitoring equipment and the reliability of the
measured data. The fault tolerance level of the bridge quality monitoring equipment is
greatly improved and the reliability of the measured data is also greatly enhanced. On
this basis, the bridge quality monitoring equipment fault judgment was implemented,
and the fault tolerance convergence level evaluation results were obtained as shown
in Figure 2.
Figure 2. Monitoring convergence level curve
From the convergence curve comparison, it is known that the convergence curve of
the method designed in this paper for bridge quality monitoring equipment fault
monitoring is smooth and of high quality, with small overall changes and reliable data
measured by side feedback. In contrast, with the data obtained from the traditional
method test, the data reliability cannot be guaranteed due to the failure of the
monitoring equipment and automatic error reporting is not excluded. Although the
highest convergence level and the lowest convergence level of the test method based
on deep learning technology are smaller than the traditional test method by about
0.01-0.02, the overall level change rate is only 2.1% per 25 iterative steps, the fault
reliability is large, the judgment ability is high, and the overall fault tolerance
convergence is high.
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
148
4.2. DATA VALIDITY ANALYSIS
After excluding the bridge quality monitoring equipment fault monitoring data, the
validity analysis of the measured monitoring data is performed. The prediction results
of decision-tree, random-forest, SVM, DNN, DBN and other machine learning
classification methods were compared. For the same dataset collected, the DBN
model can improve the accuracy by 14.27% compared with the decision tree. The
experiments prove that the deep learning model has stronger data analysis capability
and is more suitable for validity analysis of bridge quality monitoring data obtained
from monitoring.
Although deep learning models such as DNN and DBN have a great improvement
for data analysis accuracy, but because there are more parameters in the two models
and the initial values of the parameters are set randomly, it is easy to lead to local
optimum during the training process of deep learning, which affects the training results
of network models and reduces the testing accuracy. For this reason, a firefly swarm
optimization algorithm, combined with artificial GSO, is used to further optimize the
DBN model. Since GSO has a strong ability to solve local optimization problems, its
objective function can be a loss function, and the initial parameters of the model are
optimized through GSO to improve the model's applicability and test accuracy. the
DBN-GSO model represents a random initialization of parameters to the original DBN-
R model, and the results show that the data accuracy obtained from testing through
the DBN-GSO optimized model is higher than that of the initial model by The accuracy
is improved by 19.19% compared with the initial model, indicating that the GSO
optimization algorithm can further optimize the DBN model and improve the training
test accuracy.
4.3. STRUCTURAL SAFETY ANALYSIS
Based on the above analysis, it is clear that the optimized DBN algorithm has
improved the model prediction accuracy, so this subsection further analyzes the
structural parameters affecting bridge safety by the optimized DBN algorithm. The
bridge construction process is fixed by the connection between the piers, the bridge
body and the bridge deck, and its structural integrity is an important factor affecting
the stability of the bridge. This paper focuses on the structural analysis of bridges,
including the bearing of bridges under overload conditions of use and force majeure
factors. These factors include mainly pedestrian and vehicle loads (lateral and
longitudinal), ambient temperature, typhoons and seismic natural disasters, and the
input layer data of the influencing factors under certain criteria are imported into DL,
as shown in Table 1.
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
149
Table 1. Input layer parameter characteristics
For the input parameters, the DL model is calculated iteratively to obtain the output
results as shown in Table 4. The output layer results represent the structural
influences on the safety and stability of the bridge, highlighting the most jointed areas.
This is because of the large and complex structure of the bridge and the high degree
of correlation between the internal structures. Relying solely on the engineer's
experience to make judgments about safety issues at design time would slow down
the project schedule on the one hand, and there is no guarantee that the experience
is accurate and that any safety-threatening issues in the bridge design must be
eliminated. The structure obtained from the DL model can help engineers find the key
aspects of the construction problem and provide guidance for the bridge design. Table
2 lists the force problems during the use of the bridge. The main sources of force are
load and wind, and the parts that are subject to the greatest axial force, bending
moment, and shear can be obtained from the input results. The offset is then the
stability problem of the bridge in the presence of stresses. The same analysis was
carried out for different parts to get the sensitive areas affecting the main girders, piers
and towers.
No. Load Type
Load design limit value
Base value setting
basis Minimal value Maximum value
1Vehicle load /
kN·m-2
Consider the
overload situation 0 1.3
2Crowd Load /
kN·m-2
Number of people in
large events 0 2.8
3Temperature /
Meteorological
statistics temperature
maximum value
-15 40
4Windward load
m·s-1Category 10 typhoon 0 27.6
5Upwind load
m·s-1Category 10 typhoon 0 27.9
6Transverse vehicle
load /.kN
Consider the
overload situation 0 45000
7Longitudinal vehicle
load /kN
Consider the
overload situation 0 18000
8 Seismic load /g 8 magnitude
earthquake 0 0.24
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
150
Table 2. Output layer parameters and characteristics
5. CONCLUSION
This paper is based on the definition and role of both deep learning technology and
the BIM model, and discusses the significance of both in the quality management of
bridge design and construction stage, systematically analyzes the guidance
significance of the BIM model in bridge design and construction, describes the
application process of deep learning technology and its role in monitoring
experimental test in bridge quality monitoring equipment failure, and obtains the
following conclusions.
1.
The application of the BIM model in bridge engineering design and
construction mainly through the construction of models, the development of
construction strategies, statistical engineering volume and fault monitoring and
material statistics during construction, compared to the traditional two-
dimensional CAD drawings BIM model has more aspects of bridge engineering
design and construction, improving the efficiency of about 35%.
2.
Deep learning techniques applied in bridge quality management monitoring
can improve the reliability of measurement data and analysis efficiency. The
bridge quality monitoring equipment fault monitoring and troubleshooting
system designed in this paper can improve up to 69.97% in 250S iteration time
compared with the data obtained from the traditional method testing, while the
No.
Type
Major Categories Minor Categories position
1
Inner Strength
Axial force
Tower main beam
2 Tower Root Section
3
Bending moment
Bridge pier main girders
4Tower Headquarters
(Yokohama direction)
5
Shear force
Bridge pier main girders
6Tower root (longitudinal
bridge direction)
7
Offset degree
Main beam Middle of main beam
8 Bridge pier Left side of the bridge pier
9Bridge Tower Tower top (cross-bridge
direction)
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
151
level value of the convergence curve is high, with a change rate of only 2.1%
per 25 iterative steps.
3.
The validity analysis of the measured data after the bridge quality monitoring
equipment fault monitoring and elimination, the DNN model and DBN model in
the deep learning technique can improve the accuracy of data analysis by
12.51% and 14.26%, respectively, and the DBN-GSO model combined with the
GSO optimization algorithm can also avoid the local optimization results and
improve the training accuracy by 19.19% compared with the original model.
The optimized model is further analyzed for load-stress analysis and safety
issues of force majeure factors for bridge structures. Based on the imported
parameters, the specific structural parameters affecting the bridge quality and
safety were obtained by iterative learning.
REFERENCES
(1)
Hu, Y., & Chen, Y. (2022). Coupling of Urban Economic Development and
Transportation System: An Urban Agglomeration Case. Sustainability, 14.
(2)
Li, B., Gao, S., Liang, Y., et al. (2020). Estimation of Regional Economic
Development Indicator from Transportation Network Analytics. Scientific Reports,
10(1).
(3) Wang, X., Wang, X., Dong, Y., et al. (2020). A Novel Construction Technology for
Self-Anchored Suspension Bridge Considering Safety and Sustainability
Performance. Sustainability, 12.
(4)
Hu, J., Carvel, R., & Usmani, A. (2021). Bridge fires in the 21st century: A
literature review. Fire Safety Journal, 126, 103487.
(5)
Guo, X., Zhang, C., & Chen, Z. Q. (2020). Dynamic performance and damage
evaluation of a scoured double-pylon cable-stayed bridge under ship impact.
Engineering Structures, 216(3), 110772.
(6) Wang, X., Wang, H., Sun, Y., Liu, Y., Liu, Y., & Liang, P., et al. (2022). Fault-
tolerant interval inversion for accelerated bridge construction based on
geometric nonlinear redundancy of cable system. Automation in Construction.
(7)
Wu, Y., & Lu, P. (2022). Comparative analysis and evaluation of bridge
construction risk with multiple intelligent algorithms. Mathematical Problems in
Engineering, 2022.
(8)
Farre-Checa, J., Komarizadehasl, S., Ma, H., Lozano-Galant, J. A., & Turmo, J.
(2022). Direct simulation of the tensioning process of cable-stayed bridge
cantilever construction. Automation in Construction.
(9)
Yu, S., Xu, Z., Su, Z., et al. (2021). Two Flexible Vision-based Methods for
Remote Deflection Monitoring of a Long-span Bridge. Measurement, 2021(1),
109658.
(10)
Zhou, J., Li, T., Ye, X., et al. (2020). Safety assessment of widened bridges
considering uneven multi-lane traffic load modeling: a case study in China.
Journal of Bridge Engineering, 25(9), 1-12.
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
152
(11) Yang, Y., Ng, S. T., Dao, J., et al. (2021). BIM-GIS-DCEs Enabled Vulnerability
Assessment of Interdependent Infrastructures – A Case of Stormwater Drainage-
Building-Road Transport Nexus in Urban Flooding. Automation in Construction,
125(4).
(12) Montiel-Santiago, F. J., Hermoso-Orzáez, M. J., Terrados-Cepeda, J., et al.
(2020). Sustainability and Energy Efficiency: BIM 6D. Study of the BIM
Methodology Applied to Hospital Buildings. Value of Interior Lighting and
Daylight in Energy Simulation. Sustainability.
(13) Zhang, J., Zhu, W., Wu, X., et al. (2021). Traffic State Detection Based on
Multidimensional Data Fusion System of Internet of Things. Wireless
Communications and Mobile Computing, 2021(1), 1-12.
(14) Zhang, Hongzhi. (2021). Regression function model in risk management of bank
resource allocation. Applied Mathematics and Nonlinear Sciences, doi:10.2478/
AMNS.2021.1.00093.
(15) Gao, Y., Li, H., Xiong, G., & Song, H. (2023). AIoT-informed digital twin
communication for bridge maintenance. Automation in Construction.
(16) Phares, B., Lee, Y. S., Hosteng, T. K., & Nelson, J. (2021). Investigation of
grouted coupler connection details for accelerated bridge construction.
Transportation Research Record, 2675(8), 562-569.
(17) Ugural, M. N., & Burgan, H. I. (2021). Project performance evaluation using EVA
technique: Kotay Bridge construction project on Kayto River in Afghanistan.
Tehnicki Vjesnik, (1).
(18) Clarke-Sather, A. R., McConnell, J. R., & Masoud, E. (2021). Application of lean
engineering to bridge inspection. Journal of Bridge Engineering, 26(2),
04020120.
(19) Wang, H., Ranade, R., & Okumus, P. (2023). Calibrating a physics-based
corrosion model with field-based bridge condition data. Journal of Bridge
Engineering.
(20) Chen, B., Liu, Q., Chen, H., et al. (2021). Multiobjective optimization of building
energy consumption based on BIM-DB and LSSVM-NSGA-II. Journal of Cleaner
Production, 294(5-6), 126153.
(21) Utkucu, D., & Szer, H. (2020). Interoperability and data exchange within BIM
platform to evaluate building energy performance and indoor comfort.
Automation in Construction, 116, 103225.
(22) Caterino, N., Nuzzo, I., Ianniello, A., et al. (2021). A BIM-based decision-making
framework for optimal seismic retrofit of existing buildings. Engineering
Structures, 242(3), 112544.
(23) Xu, B., & Yuan, X. (2022). A Novel Method of BP Neural Network Based Green
Building Design—The Case of Hotel Buildings in Hot Summer and Cold Winter
Region of China. Sustainability, 14.
(24) Wang, S., Zeng, X., Zhou, T., et al. (2020). Image super resolution via nonlocal
and second-order feature fusion network. Journal of Electronic Imaging, 29(2), 1.
(25) Pan, Y., & Zhang, L. (2020). BIM log mining: Exploring design productivity
characteristics. Automation in Construction, 109(Jan.), 102997.1-102997.14.
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
153
(26)
Forgues, D. (2021). Bim Machine Learning and Design Rules to Improve the
Assembly Time in Steel Construction Projects. Sustainability, 14.
(27) Edirisinghe, R., Pablo, Z., Anumba, C., et al. (2021). An Actor–Network Approach
to Developing a Life Cycle BIM Maturity Model (LCBMM). Sustainability, 13.
(28) Liang, X. (2019). Image-based post
isaster inspection of reinforced concrete
bridge systems using deep learning with Bayesian optimization. Computer-Aided
Civil and Infrastructure Engineering, 34(5), 415-430.
(29)
Abbas, T., Kavrakov, I., Morgenthal, G., et al. (2020). Prediction of aeroelastic
response of bridge decks using artificial neural networks. Computers &
Structures, 231, 106198.
(30) Oz, A., Kb, A., Skb, B., et al. (2020). Real-Time Detection of Cracks on Concrete
Bridge Decks Using Deep Learning in the Frequency Domain. Engineering.
(31)
Ma, R., Zhang, Z., Dong, Y., et al. (2020). Deep Learning Based Vehicle
Detection and Classification Methodology Using Strain Sensors under Bridge
Deck. Sensors, 20(18).
(32)
Dinh, K., Gucunski, N., Duong, T. H. (2018). An algorithm for automatic
localization and detection of rebars from GPR data of concrete bridge decks.
Automation in Construction, 89(MAY), 292–298.
(33)
In-Ho, K., Haemin, J., Seung-Chan, B., et al. (2018). Application of Crack
Identification Techniques for an Aging Concrete Bridge Inspection Using an
Unmanned Aerial Vehicle. Sensors, 18(6), 1881.
(34)
Yang, J. X., Yang, F., Zhang, L., et al. (2021). Bridge health anomaly detection
using deep support vector data description. Neurocomputing, 444.
https://doi.org/10.17993/3ctic.2023.122.137-154
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed.43 | Iss.12 | N.2 April - June 2023
154