sensors

Article
Digital Twin-Based Risk Control during Prefabricated Building
Hoisting Operations
Zhansheng Liu *, Anxiu Li, Zhe Sun, Guoliang Shi and Xintong Meng

Department of Urban Construction, Beijing University of Technology, Beijing 100124, China;

[email protected] (A.L.); [email protected] (Z.S.); [email protected] (G.S.);
[email protected] (X.M.)
* Correspondence: [email protected]

Abstract: Prefabricated buildings have advantages when it comes to environmental protection.

However, the dynamics and complexity of building hoisting operations bring significant safety risks.
Existing research on hoisting safety risk lacks a real-time information interaction mechanism and
lacks scientific control decision-making tools based on considering the correlation between safety
risks. Digital twin (DT) has the advantage of real-time interaction. This paper presents a safety risk
control framework for controlling prefabricated building hoisting operations based on DT. In the case
of considering the correlation of the safety risk index of hoisting, the safety risk hierarchy model of
hoisting is defined in the process of building the DT model. The authors have established a Bayesian
network model into the process of the integrated analysis of the digital twin mechanism model
and monitoring data to realize the visualization of the decision analysis process of hoisting safety
risk control. The key degree of the indirect inducement variable to direct inducement variable was
calculated according to probability. The key factor leading to the occurrence of risk was found. The
effectiveness of the hoisting safety risk control method is verified by a large, prefabricated building



project. This method provides decision tools for hoisting safety risk control, assists in formulating

 effective control schemes, and improves the efficiency of information integration and sharing.
Citation: Liu, Z.; Li, A.; Sun, Z.; Shi,
G.; Meng, X. Digital Twin-Based Risk Keywords: digital twin; prefabricated building hoisting; safety risk control; decision visualization
Control during Prefabricated
Building Hoisting Operations.
Sensors 2022, 22, 2522. https://
doi.org/10.3390/s22072522 1. Introduction
Academic Editors: Filippo Ubertini The traditional construction industry has the disadvantages of extensive construc-
and Chunhua Yang tion mode, waste of resources, frequent safety accidents, and low intelligence level [1].
Received: 11 February 2022
Integrating modern information technology into the current practice of the construction in-
Accepted: 23 March 2022
dustry has promoted the rapid development of new building industrialization represented
Published: 25 March 2022
by prefabricated buildings. Prefabricated building is a new construction method, which
involves many new technologies in the construction stage [2]. Such a new construction
Publisher’s Note: MDPI stays neutral
method has the advantages of resource-saving, environmental protection, and a short
with regard to jurisdictional claims in
construction period, which is an important way to transform the traditional construction
published maps and institutional affil-
mode to the modern industrialized construction mode [3]. Unfortunately, due to the low
iations.
level of professional skills of construction workers, the construction stage of prefabricated
buildings has also become a stage of frequent safety accidents.
Prefabricated construction processes contain the transportation stage, hoisting stage,
Copyright: © 2022 by the authors.
and installation stage. The hoisting stage is the most critical part of various prefabricated
Licensee MDPI, Basel, Switzerland. construction processes. Hoisting operations have the following characteristics [4]: (1)
This article is an open access article prefabricated components with large weight and high-altitude operation; (2) the hoisting
distributed under the terms and operation is a combined movement, with many and scattered dangerous points; (3) large
conditions of the Creative Commons range of movement; (4) cooperation of multiple working types; (5) the influence relation-
Attribution (CC BY) license (https:// ship between various hoisting safety risk factors is complex. In the hoisting stage, safety
creativecommons.org/licenses/by/ accidents such as the overturning of tower cranes, rope breaking, collision, decoupling,
4.0/). and the falling of components occur easily [5]. The reasons for safety accidents are mostly

Sensors 2022, 22, 2522. https://doi.org/10.3390/s22072522 https://www.mdpi.com/journal/sensors

Sensors 2022, 22, 2522 2 of 24

inadequate on-site safety management, workers’ illegal operation, and inadequate equip-
ment maintenance. Therefore, the safety control problem in the hoisting stage is worthy of
in-depth study.
Safety has always been the focus of the construction industry. According to official
statistics, there were 773 accidents and 904 deaths in the production of housing and munici-
pal engineering in 2019. The object strike accidents accounted for 15.91% of the total, and
the hoisting machinery injury accidents accounted for 5.43% of the total [6]. The occurrence
of the two accidents is closely related to the hoisting operation. Many scholars have done
a lot of research in the field of safety risk management, but there are still few research
results in the field of prefabricated building safety risk management. A considerable num-
ber of scholars focus on the optimal arrangement of tower cranes [7,8], transportation
scheduling [9], and motion control [10,11]. With the development of safety risk research
theory and technology, the research concept of safety risk issues has also been transformed
from post-analysis to pre-prevention. The research on safety issues is roughly divided
into three stages [12]: the first stage focuses on the statistical analysis of safety accidents
and safety measures; the second stage enters the active safety management stage, using
more advanced statistical techniques to identify and evaluate the safety risk; the third stage
is the real-time monitoring stage of security risks. The use of virtual reality technology
for security training and the real-time collection of security risk information can prevent
and control security risks in a timely fashion. However, the dynamics of construction site
elements are still difficult to track, and lack effective information integration and interaction
mechanisms.
At present, relying on the supervision of on-site safety management personnel to
control the occurrence of safety accidents is very time-consuming. Visualization technol-
ogy has been gradually applied to the safety management of construction sites. With
the help of location information, imaging information, and alarm technology, workers’
behaviors and site environments (including equipment) can be monitored, analyzed, and
warned [13]. Although visualization technology has developed, it is often only managed
for individual safety risk factors. The hoisting operation process of prefabricated buildings
is a dynamic system that contains multiple factors of the human-machine-component-
environment. Therefore, the elements contained in the system cannot be separated and
controlled separately. Digital twin provides a solution to complex system problems and has
been widely used in workshop scheduling [14], fault detection, and diagnosis [15] in the
field of intelligent manufacturing. Digital twin can integrate various technologies. Based
on establishing a virtual model, the data collected from the site are fused with the model,
and the analysis results are fed back to the construction site in real-time to realize the visual
control of safety risks. However, the application of digital twin in the construction industry
is not mature enough. The introduction of a digital twin to the safety management of
prefabricated building hoisting mainly faces the following problems: based on considering
the correlation between the safety risk factors of prefabricated building hoisting, it is very
important to integrate the safety risk mechanism model with the data collected in real-time
and achieve the purpose of the scientific decision-making of safety control.
This paper presents a framework for the safety risk control of prefabricated building
hoisting based on digital twin. To ensure the better integration of the mechanism model
and data to express the relationship between safety risk factors, the hierarchical model of
hoisting safety risk is defined in the process of establishing the digital twin model. A multi-
level hierarchical model is established by using the Decision-making Trial and Evaluation
Laboratory (DEMATEL) and Interpretative Structural Modeling Method (ISM) to show
the influential relationship between the factors of hoisting safety risk. Bayesian network is
introduced in the process of driving the digital twin mechanism model and monitoring
data integration analysis. After determining the direct and indirect inducement variables
of security risk, the prior probability of each node is obtained by training Bayesian network
parameters with the expectation–maximization (EM) algorithm. If the direct inducement
variable is in a dangerous state, the occurrence probability is set to 100% and input into the
Sensors 2022, 22, 2522 3 of 24

Bayesian network structure as evidence to obtain the posterior probability. By calculating

the criticality of each indirect inducement variable to the direct inducement variable in a
dangerous state and comparing the criticality, the key factors leading to the occurrence
of danger are screened to achieve the purpose of the visual analysis of safety risk control
decision. Finally, the feasibility of the proposed hoisting safety risk control framework is
further verified by taking the risk of the cable inclination at a certain time in a specific case
as an example.
The organization of the paper is as follows. Section 2 presents an in-depth literature
review of safety risk management and digital twin. Section 3 introduces the DT-based
prefabricated building hoisting safety risk control framework. Section 4 describes the
establishment process of hoisting a safety risk control model based on digital twinning,
including the physical model, virtual model, and safety risk hierarchy model. Section 5
introduces the data fusion steps of the DT model driven by the Bayesian network. Section 6
uses a case study to validate the proposed method. Section 7 summarizes the major findings
of this study and concludes.

2. Literature Review
The literature review of this paper includes two parts: construction safety risk manage-
ment and digital twin. This section focuses on the development of safety risk management
methods and related research on the safety risk management of prefabricated building
construction, and analyzes the shortcomings of prefabricated building safety management
compared with other fields. In addition, the deep application of digital twin in the field
of intelligent manufacturing and related research in the field of intelligent construction is
introduced to highlight the advantages of digital twin technology.

2.1. Construction Safety Risk Management

Many scholars have studied the issue of safety risk management and have formed
many mature methods and technologies [16]. Safety risk management includes safety
risk identification and assessment, security risk prediction, and security risk control. At
present, work breakdown structure-risk breakdown structure (WBS-RBS) [17], analytic hi-
erarchy process (AHP) [18], fault tree analysis (FTA) [19], Fuzzy comprehensive evaluation
(FCE) [20], Cloud model [21], D-S evidence theory [16], and other methods have been used
in construction safety risk management. The safety risk management scheme integrated by
various methods has also been applied. For example, Yang et al. introduced data-driven
Bayesian networks into the TOPSIS method. A new ship detention risk assessment method
and reasonable risk control strategy are proposed [22]. With the deepening of theoretical
research on safety risk management, the research focus has gradually turned to the essence
of safety risk. Based on the extraction of the accident chain, the construction of the risk
network, and the formulation of safety measures, Chen et al. proposed a data-driven
research framework of construction engineering safety promotion strategy based on the
risk network [23].
In recent years, BIM, artificial intelligence, and Internet of Things are also applied to
construction safety risk management. Choe et al. proposed a four-dimensional building
safety planning process, which solved the problem of spatio-temporal safety information
integration at specific locations, and provided an active safety planning tool for safety
managers [24]. Lin et al. used the Fuzzy set theory and machine learning method in the
risk assessment of mining process and integrated real-time collected data into the BIM
management platform to monitor, control, and manage dynamic safety risks [25]. Li et al.
proposed the safety risk identification system (SRIS) and early warning system (SREWS)
for subway construction in China. The contents of the risk identification knowledge base,
retrieval matching algorithm, data fusion, and BIM technology monitoring safety state are
mainly studied [26]. Some scholars also apply ontology knowledge [27], computer vision,
and semantic reasoning [28] to construction safety management, to more accurately assess
the level of safety risks.
Sensors 2022, 22, 2522 4 of 24

In the methods mentioned above, the objects of construction safety risk assessment and
control are mostly subway, tunnel, pipe gallery, and other underground projects. Although
BIM, Internet of Things, artificial intelligence, and other technologies are integrated, the
technologies are not very well integrated, the information interaction mechanism is still
imperfect, and most studies ignore the correlation between safety risks. At present, the
research on the safety risk management of prefabricated building hoisting operation is
relatively minor. Safety risk assessment methods are mostly static and subjective. Xun
et al. proposed the DEMATEL method to calculate the mutual influence between the
index weights and the method of applying game theory to calculate the combined weights.
The safety risk of the prefabricated building hoisting operation is evaluated by the cloud
model [29]. Zhao et al., based on the method of determining the weight of the conventional
safety risk index, introduced the system dynamics theory and established the SD model
of the safety risk measurement of prefabricated building hoisting operation combined
with qualitative and quantitative, which solved the dynamic and complex problems in the
construction safety risk system [30]. Shen et al. constructed a safety risk assessment model
based on Cloud-BN to dynamically evaluate the safety risk of prefabricated hoisting [31].
Liu et al. established a safety risk early warning model of hoisting operation based on RVM
(relevance vector machine) to accurately judge the safety status of hoisting operation [32].
In addition, some scholars analyzed the safety accidents of tower cranes and determined
the safety risk factors of tower cranes. From the perspective of complex social and technical
systems, the safety of tower cranes was evaluated by qualitative and quantitative meth-
ods [4], the safety factor of tower cranes was developed [33], and the safety standards were
improved [34].

2.2. Digital Twin

Digital twinning refers to mapping physical entities in the real world into the digital
world, and forming a digital model corresponding to physical entities. Digital models
and physical entities can be bidirectional information exchange fusion and iterative opti-
mization, which can improve the performance of physical systems in the real world. The
concept of digital twin was first proposed by Professor Grieves in the product life cycle
management course of the University of Michigan in 2003 [35]. It was mainly used in the
military and aerospace fields in the early period [36,37]. Digital twin is now widely used in
intelligent manufacturing. Tao et al. proposed the concept of a digital twin five-dimensional
model and applied it to 10 fields, such as satellite/space communication network, ship,
vehicle, manufacturing workshop, and smart city [38]. He also proposed the construction
theory of the digital twin model [39] and the evaluation index system of the digital twin
model [40]. Digital twin technology is suitable for solving complex system problems and
has entered a more in-depth research stage of dynamic modeling [41], data fusion [42], and
virtual-real interaction [43] in the manufacturing industry.
Scholars in the construction industry have gradually realized that digital twin tech-
nology is the key technology to realize intelligent construction [44], which can meet the
needs of data fusion and interaction. The application and components of digital twin in
the construction industry have not been clearly defined, but some scholars have made
preliminary explorations in design, construction, operation, and maintenance in the con-
struction field [45]. Akula et al. realized the real-time monitoring of drilling process
hazards by processing and integrating point clouds from 3D imaging technology to the
drilling process [46]. Jiang et al. proposed an intelligent safety management framework for
construction sites based on the concepts of digital twinning and the CPS system, which
can alarm and control dangerous states such as personnel and machinery [47]. Liu et al.
proposed an intelligent fire evacuation method based on digital twin, which was verified
in the Winter Olympics venues, and laid a foundation for the development of intelligent
operation and maintenance [48]. However, these research results have not been confirmed
under more complex conditions. In the field of hoisting safety risk management, Liu et al.
also proposed a set of modeling theories for the space-time evolution of the prefabricated
Sensors 2022, 22, 2522 5 of 24

building construction process, and focused on the modeling method and process [49].
A framework of hoisting safety risk coupling rule mining based on digital twin is also
proposed [50]. In summary, the realization of digital twin in the construction field depends
on the integrated application of BIM technology and Internet of Things technology [51].
The application of digital twin in the construction stage is in the real-time simulation
stage [1]. In the existing research on hoisting safety risk management, the modeling method
of digital twin has not risen to the research on the essence of safety risk, and the fusion and
interaction mechanism between real-time data and the model has become a key issue of
research.

2.3. Research Gap

Although many scholars have studied the construction safety management methods
in the above literature, there are still some limitations. The safety risk management method
integrating BIM, Internet of Things, and other technologies has been applied to the tunnel,
subway, and other underground engineering construction fields, and the concept of digital
twin has also been introduced. However, the interactive feedback mechanism of safety risk
information is still not perfect. The method of real-time warning of safety risk state based
on threshold lacks the research on the interaction of safety risk factors and the effective
control and decision-making basis. Compared with the safety management methods such
as tunnels and subways, the safety management method of prefabricated building hoisting
has the limitations of a static state, relying on expert experience, and low intelligence level.
The existing modeling methods of digital twin hoisting safety risk management have not
risen to the research level of the essence of hoisting safety risk, and there is still a large
gap in the research when it comes to how to integrate hoisting safety risk correlation into
the establishment of a digital twin model. The introduction of digital twin can integrate
BIM, Internet of Things, and other technologies, and establish a virtual–real interaction
mechanism through real-time data collection, transmission, and analysis. In the study
of the safety management and control of prefabricated building hoisting, how to better
integrate the correlation between safety risk indicators and the virtual model is also a
problem to be solved. This paper presents a framework for the safety risk control of
prefabricated building hoisting based on digital twin. To ensure the better integration of
the mechanism model expressing the relationship between the safety risk index factors
and the monitoring data, the safety risk hierarchical model is defined in the process of
establishing the digital twin model. The Bayesian network is introduced to effectively
combine the real-time collected data with the visual network structure, which establishes
an effective virtual–real interaction mechanism for the scientific decision-making of the
safety risk control of prefabricated building hoisting based on digital twin.

3. Digital Twin Framework for Safety Risk Control of Prefabricated Building Hoisting
The establishment of the safety risk index of prefabricated building hoisting is the
foundation of the building safety risk control digital twin framework. The safety risk factors
of a prefabricated building hoisting operation site include personnel factors, equipment
factors, component factors, environmental factors, and management factors. By consulting
a large number of research studies [4,29–32,50] and discussing them with experts, the safety
risk index system of prefabricated building hoisting is established, as shown in Figure 1.
The prefabricated building hoisting site is a dynamic and complex system that includes
multi-dimensional human-machine-component-environment. When building a digital
twin framework for hoisting safety risk control, the idea of dynamic modeling should be
integrated based on the geometry-physics-behavior-rule model. Not only the establishment
method of each component of the digital twin model, but also the real-time interaction
between the model and the data should be considered. In this way, the analysis results
of the virtual hoisting site can be fed back to the physical hoisting site in real-time to
achieve scientific decision-making. The digital twin framework for the safety risk control
of prefabricated building hoisting is shown in Figure 2.
SensorsSensors
Sensors 2022, 22,
2022, 22, xx FOR
2022, FOR PEER REVIEW
22, 2522
PEER REVIEW 67 of
6 of 24 of 25
25

Table 1. Monitorable variables of hoisting space [53].

Safety Risk Indicator (ri) Monitorable Variables

Technical level of workers (r1) basic information of workers
blink frequency/facial features of tower
Physical and mental state of workers (r2)
crane drivers
workers’ position/wearing of protective
Workers’ safety awareness (r3)
equipment
stress or strain of load/key parts of tower
Equipment bearing capacity test (r5)
crane/verticality of tower crane
the inclination angle of the hanger/the stress
Connection strength of sling (r6)
and strain of the hanger
Temporary support strength (r8) stress and strain of supporting key parts
Component manufacturing quality (r9) component basic information
stress and strain at lifting point of hoisting
Strength of hoisting components (r10)
components
component location information
Temporary stacking of components (r11)
(whether secondary handling is required)
Climate environment of hoisting operation
wind grade/wind speed
(r16)of prefabricated building hoisting.
Figure 1. Safety risk index
Figure 1. Safety risk index of prefabricated building hoisting.

The prefabricated building hoisting site is a dynamic and complex system that in-
cludes multi-dimensionalinteractive feedback
human-machine-component-environment. When building a
digital twin framework for hoisting safety risk control, the idea of dynamic modeling
should be integrated based on the geometry-physics-behavior-rule model. Not only the
establishment method of each component of the digital twin model, but also the real-time
Virtual hoisting site

interaction between the model and the data should be considered. In this way, the analysis
Dispersion of variables
results of the virtual hoisting
Personnel data site can be fed back to the physical hoisting site in real-time
equipment to achieve scientific decision-making.
hierarchical Bayesian The digital twin framework for the safety risk con-
model model Network Determining Variable
trol of prefabricated building hoisting Parameter is shown
State Levels in Figure 2. Risk control decision
Update visualization
component environmentBased on the idea of establishing the digital twin evolution model [52], the hoisting
equipment
model model data
safety risk control framework is built according to the realization logic of the digital twin
model. Firstly, Revit software is used to establish the geometric model of the hoisting op-
Twin Database

eration site (equipment, Component environment, prefabricated position of danger components, etc.) and the position
Risk control decision analysis

data NO
relationship between each element on the site is clarified. The assembly and integration of
Yes
each element model are carried out to complete the high-precision simulation of each el-
Dynamic Simulation of
embedded LoRa
terminal ement
module and its attributes,
environmental
from the physical Bayesianhoisting
network site to the virtual hoisting site. Sec-
Hoisting Process
reasoning
ondly, the influential data
relationship between safety risk factors in the hoisting process is
analyzed,
force
and the hierarchical relationship between risk factors is given to establish a hi-
Physical hoisting site

RFID tag
sensor key degree analysis
erarchical model of safety risk. By introducing a Bayesian network, the variable infor-
management
mation on safety riskdataelements collected on-site Making control
is discretized and its state-level is judged.
The variable information is shown in Table 1. If it is in a dangerous state, the posterior
decision

Real-time safety risk

probability is obtained by Bayesian network reasoning, and the key degree of indirect
assessment
inducing factors to dangerous state variables is calculated to lay the foundation for rea-
sonable decision-making. Finally, to ensure the real-time interaction between the above
interactive feedback
data and the mechanism model, data acquisition and transmission devices such as RFID
tags and LoRa modules are set at the hoisting site to monitor the status of each factor at
Figure
the Figure2. 2.
Digital
hoisting sitetwin
Digital andarchitecture
twin architecture
upload the fordata
for hoisting
hoisting safety
to safety
the risk
twin risk control. for analysis. The established
control.
database
BIM model is lightweight, and processed and loaded on the web page by WebGL technol-
4. Establishment of Digital Twin Model for Safety Risk Control of Prefabricated
ogy to display the real-time sensing data and visualization results of decision analysis,
Building Hoisting
and realize the visualization of risk control decision, the dynamic simulation of the hoist-
The digital
ing process, andtwin risk model
the real-time is built from
assessment three risk
of safety aspects: thephysical
on the physicalhoisting space resource
site. al-
Sensors 2022, 22, 2522 7 of 24

Based on the idea of establishing the digital twin evolution model [52], the hoisting
safety risk control framework is built according to the realization logic of the digital twin
model. Firstly, Revit software is used to establish the geometric model of the hoisting
operation site (equipment, environment, prefabricated components, etc.) and the position
relationship between each element on the site is clarified. The assembly and integration of
each element model are carried out to complete the high-precision simulation of each ele-
ment and its attributes, from the physical hoisting site to the virtual hoisting site. Secondly,
the influential relationship between safety risk factors in the hoisting process is analyzed,
and the hierarchical relationship between risk factors is given to establish a hierarchical
model of safety risk. By introducing a Bayesian network, the variable information on safety
risk elements collected on-site is discretized and its state-level is judged. The variable
information is shown in Table 1. If it is in a dangerous state, the posterior probability is
obtained by Bayesian network reasoning, and the key degree of indirect inducing factors to
dangerous state variables is calculated to lay the foundation for reasonable decision-making.
Finally, to ensure the real-time interaction between the above data and the mechanism
model, data acquisition and transmission devices such as RFID tags and LoRa modules are
set at the hoisting site to monitor the status of each factor at the hoisting site and upload
the data to the twin database for analysis. The established BIM model is lightweight, and
processed and loaded on the web page by WebGL technology to display the real-time
sensing data and visualization results of decision analysis, and realize the visualization
of risk control decision, the dynamic simulation of the hoisting process, and the real-time
assessment of safety risk on the physical hoisting site.

Table 1. Monitorable variables of hoisting space [53].

Safety Risk Indicator (ri ) Monitorable Variables

Technical level of workers (r1 ) basic information of workers
Physical and mental state of workers (r2 ) blink frequency/facial features of tower crane drivers
Workers’ safety awareness (r3 ) workers’ position/wearing of protective equipment
Equipment bearing capacity test (r5 ) stress or strain of load/key parts of tower crane/verticality of tower crane
Connection strength of sling (r6 ) the inclination angle of the hanger/the stress and strain of the hanger
Temporary support strength (r8 ) stress and strain of supporting key parts
Component manufacturing quality (r9 ) component basic information
Strength of hoisting components (r10 ) stress and strain at lifting point of hoisting components
component location information
Temporary stacking of components (r11 )
(whether secondary handling is required)
Climate environment of hoisting operation (r16 ) wind grade/wind speed

4. Establishment of Digital Twin Model for Safety Risk Control of Prefabricated

Building Hoisting
The digital twin risk model is built from three aspects: the physical space resource
allocation of hoisting operation, the virtual space modeling and simulation of hoisting
operation, and the hierarchical model of hoisting safety risk, aiming at effective virtual-real
interaction and the full integration of the data and model of the digital twin model.

4.1. Physical Space Resource Allocation in Hoisting Operation

According to the digital twin framework of hoisting safety risk control, the physical
entities in hoisting operation space include workers, tower crane equipment, prefabri-
cated components, and hoisting operation site layout. To better drive the digital twin
framework of hoisting safety risk control, it is necessary to connect various elements of
human-machine-component-environment in real-time, so that the physical space of the
hoisting operation has the virtual and real interaction ability of intelligent perception
Sensors 2022, 22, 2522 8 of 24

and transmission. At present, the Internet of Things technology system composed of a

multi-sensor, radiofrequency technology and a wireless network has been widely used,
which provides support for the virtual-real interaction of the digital twin model. The data
variables that can be monitored in the physical space of hoisting operation according to the
safety risk index system are shown in Table 1.
According to the physical space modeling method [50] proposed by the research
group, for example, active RFID tags can be used to locate the position of personnel and
components, and the relevant basic information of personnel and components is input
into the tags for subsequent data retrieval. The wind speed, the stress and strain of
components and tower cranes, and the inclination change of hoisting components can
be collected through embedded sensors. Different information perception schemes are
selected according to the variable information required for monitoring in actual projects.
The collected monitoring data can be connected to the hoisting operation space re-
sources through a communication protocol. Combined with the configured intelligent
sensing device and wireless transmission network, the full interconnection and integra-
tion of multi-source heterogeneous elements such as human, machine, component, and
environment in the physical workshop can be realized.

4.2. Virtual Space Modeling and Simulation of Hoisting Operation

The hoisting process mainly includes the hoisting stage, hoisting transportation stage,
and installation stage of the component. The specific process is shown in Figure 3. The
hoisting process is a complex, continuous, and dynamic process. To ensure the high fidelity
of the digital twin model, this paper incorporates the modeling idea of “geometry-logic”
into the existing virtual model establishment method of the research group [54]. According
to the characteristics of various entities in the physical space, modeling is carried out
from two dimensions of geometry and logic. The geometric model mainly includes the
personnel attributes of space, tower crane equipment, prefabricated components, and
field working environment layout in the physical space. Specifically, the geometric model
established by BIM technology should reflect the appearance, size information, attribute
information of components, equipment, hoisting site layout, and the assembly relationship
between various entity elements. The logical model mainly drives the consistency between
the virtual space of hoisting operation and the physical space in behavior, state, and
operation in the form of data, such as the working state of hoisting equipment and the
state of prefabricated components. To realize the faithful mapping of virtual space to
physical space, BIM4D technology can be used to add time factors to define the logical
model containing factors such as behavior, rules, and constraints; for example, the position
Sensors 2022, 22, x FOR PEER REVIEWchange of the component. The geometric model and logical model are combined to 9form of 25a
completely virtual space model of hoisting operation.

Figure 3. Division of prefabricated building hoisting process.

Figure 3. Division of prefabricated building hoisting process.

The BIM model can provide a visualization platform for the virtual space of the hoist-
ing operation. The BIM model is lightweight, and processed and mounted on the web
page, and embedded in the data analysis and processing module with intelligent algo-
rithm as the core. On this basis, the virtual and real information interaction relies on the
active RFID tags, sensors, and wireless transmission networks configured in the hoisting
site to collect real-time data for the dynamic change process of human-machine-compo-
Sensors 2022, 22, 2522 9 of 24

The BIM model can provide a visualization platform for the virtual space of the
hoisting operation. The BIM model is lightweight, and processed and mounted on the web
page, and embedded in the data analysis and processing module with intelligent algorithm
as the core. On this basis, the virtual and real information interaction relies on the active
RFID tags, sensors, and wireless transmission networks configured in the hoisting site
to collect real-time data for the dynamic change process of human-machine-component-
environment and integrate them. After the conversion of the gateway communication
protocol, the data are sent to the virtual space by a wireless network. The virtual space
based on the lightweight BIM model visually displays, analyzes, and feeds back the decision
results to realize the real-time and effective control of the safety risk of the physical space
of the hoisting operation.

4.3. Construction of Hoisting Safety Risk Hierarchical Model

There are many factors influencing hoisting safety risks. There is a certain correlation
between the various factors, and the strength of the correlation effect is also different. In
this paper, a hierarchical model of hoisting safety risk is added to the original digital twin
model, to clearly express the direct and indirect causes of safety accidents, ensure the better
integration of the data and virtual model, and support better scientific control decisions.
The integration of the DEMATEL method and the ISM method to determine the hierarchical
structure of safety risk can effectively determine the causal relationship between factors and
obtain the deep-seated factors leading to accidents. This lays the foundation for subsequent
expert knowledge and real-time data fusion. The idea for the DEMATEL-ISM method is to
obtain the comprehensive influence matrix based on the DEMATEL method. In view of the
influence of factors on the ISM method, it is used to obtain the overall influence matrix, and
the reachable matrix is obtained by optimizing the given threshold. Finally, the hierarchical
structure between accidents is divided, and the relationship between accident systems is
obtained. The specific steps are as follows:
1. According to the safety risk index system of prefabricated building hoisting proposed
above, the influence levels of safety risk factors are divided. The index system set
is denoted as A. Questionnaires were distributed to three experts, namely techni-
cal research experts of prefabricated construction, enterprise managers, and project
managers in the field. The influence relationship between the two indicators is de-
noted as β. The questionnaire results from the three experts are further discussed to
comprehensively determine the direct impact matrix B between risk factors.
2. The initial direct influence matrix B is normalized according to Equation (1) to obtain
the normalized matrix C.

1
C= n B (1)
max ∑ β ij
1≤ i ≤ n j =1

n
In the formula, max ∑ β ij is the row sum maximum of the direct influence matrix.
1 i  n j =1

3. The normalized matrix is processed according to Equation (2) and the comprehensive
influence matrix T is obtained.

T = C ( I, −, C )−1 (2)
Based on the comprehensive influence matrix, the overall influence matrix H is ob-
tained according to Equation (3). I is the unit matrix.

H = I+T (3)

4. Referring to the method of λ in Reference [55], the threshold λ is set to 0.1 and the
reachable matrix K = (kij )nxn is obtained. A reachable matrix indicates whether there
Sensors 2022, 22, 2522 10 of 24

is a connection path from one element to another. kij is the element in the reachability
matrix and hij is the element in the comprehensive influence matrix.
(
1 hij ≥ λ
kij = i, j = 1, 2, · · ·, n (4)
0 hij < λ
kij = 1 indicates that factor i has a direct impact on factor j, and kij = 0 indicates that
there is no direct impact between the two factors.
5. Based on the reachable matrix, the reachable set and the antecedent set of factors are
constructed, as shown in Equation (5).

Ri =  r j r j ∈ R, k ij 6= 0 i = 1, 2, · · ·n
(5)
Si = r j r j ∈ R, k ji 6= 0 i = 1, 2, · · ·n
In the formula, Ri is the reachable set of each factor. Si is the antecedent set of each
factor. R is the correlation matrix of the prefabricated building hoisting safety risk index. rj
is the hoisting safety risk index.
6. Equation (6) is used to verify each factor. If it is established, it is determined as the first
layer factor, and the corresponding rows and columns are divided in the reachable
matrix. The above steps are repeated to determine the level of other factors in turn,
until all the factors in the reachable matrix are divided. Thus, the safety risk hierarchy
Sensors 2022, 22, x FOR PEER REVIEW
model is determined, as shown in Figure 4. 11 of 25

Ri = Ri ∩ Si , i = 1, 2, · · ·, n (6)

Figure4.4.Hierarchical
Figure Hierarchicalmodel
modelofofhoisting
hoistingsafety
safetyrisk
riskfactors.
factors.
5. Safety Risk Control Method of Digital Twin Hoisting Driven by Bayesian Network
5. Safety Risk Control Method of Digital Twin Hoisting Driven by Bayesian Network
Based on the above digital twin model of hoisting safety risk control, the configuration
Based on the above digital twin model of hoisting safety risk control, the configura-
of intelligent resources in the physical space of hoisting operation can ensure the real-
tion of intelligent resources in the physical space of hoisting operation can ensure the real-
time acquisition and transmission of data. The establishment of a virtual space model
time acquisition and transmission of data. The establishment of a virtual space model lays
lays a foundation for the visualization of risk control decision-making and the dynamic
a foundation
monitoring for the
of the visualization
hoisting of risk control
process. However, decision-making
integrating and the
data and virtual dynamic
models mon-
requires
itoring of the hoisting process. However, integrating data and virtual models requires ar-
tificial intelligence algorithms to drive it. Considering the correlation between safety risks
and incomplete safety monitoring mechanisms, this paper introduces a Bayesian network
to drive a better combination of digital twin model and data.
Sensors 2022, 22, 2522 11 of 24

artificial intelligence algorithms to drive it. Considering the correlation between safety risks
and incomplete safety monitoring mechanisms, this paper introduces a Bayesian network
to drive a better combination of digital twin model and data.

5.1. Overall Operation Mechanism of Hoisting Safety Risk Control Method

Based on the establishment of the digital twin model, this paper fuses the digital
twin model and data through the Bayesian network, as shown in Figure 5. The safety risk
management of prefabricated building hoisting includes multiple dimensions of human-
machine-component-environment. It has great advantages to solve the complex system of
the safety risk management of prefabricated building hoisting by using the real-time and
dynamic characteristics of digital twin technology. Considering the relationship between
safety risks and the absence of a large number of historical data on safety risk management,
the graph model is undoubtedly an effective way to solve the problem. The graph model
is one of the most popular methods in the field of machine learning. It is composed of
directed or undirected graphs through a series of nodes and wires. In the field of solving
construction safety problems, the Bayesian network, as a kind of graph model algorithm,
has been widely used. The establishment of its network structure can show a correlation
between safety risk factors. Bayesian network reasoning can be used to assist decision-
making, providing an internal driving force for hoisting safety risk control method12based
Sensors 2022, 22, x FOR PEER REVIEW of 25
on digital twin.

Learning Data Storage for Bayesian Network Parameters

Update Bayesian New variable Parameter Learning of New Analysis Model

Network Structure discretization Bayesian Networks of Safety Risk Control

Offline learning stage Update model / data

Unrecognized factors Factors of safety state fusion module

Sling angle / strength, etc. Model / data fusion module

Data sources Data exchange
Classification Bayesian Posterior
Online use stage of safety risk network probability
factors status reasoning of each node

Physical hoisting site Virtual hoisting site

Safety risk control decision visualization

Calculation and Analysis of key degree Updates and feedback

Figure 5. Overall operation mechanism of safety risk control method for hoisting.
Figure 5. Overall operation mechanism of safety risk control method for hoisting.

The
The overall
overall operation
operation mechanism
mechanism of of the
thehoisting
hoisting safety
safetyrisk
riskcontrol
controlmethod
methodisismainly
mainly
divided
divided into the offline learning stage and the online use stage. In the online use stage,
into the offline learning stage and the online use stage. In the online use stage,
digital
digital twin
twin technology
technology is is used
usedtotoestablish
establishthe thephysical
physicalhoisting
hoistingsite
siteand
andthe
the virtual
virtual hoist-
hoisting
ing
site,site,
andandthethe model/data
model/data fusionmodule
fusion moduleisisconfigured.
configured.The
Thebridge
bridge between
between the
the physical
physical
hoisting
hoisting site and the virtual hoisting site is established, and the data of the virtualhoisting
site and the virtual hoisting site is established, and the data of the virtual hoisting
site
site are interconnected.The
are interconnected. Thevariable
variableinformation,
information, such
such as sling
as sling angle
angle andand strength
strength col-
collected
lected
from the from the physical
physical site, is uploaded
site, is uploaded to thehoisting
to the virtual virtualsite
hoisting
through siteintelligent
through intelligent
perception
perception
and transmissionand transmission
equipment.equipment.
At the same Attime,
the same time, thenetwork
the Bayesian Bayesianisnetwork is used
used to process
to process the perception data in the model/data fusion module, including
the perception data in the model/data fusion module, including the state classification the state clas-
of
sification of safety risk factors and the Bayesian network reasoning to calculate
safety risk factors and the Bayesian network reasoning to calculate the posterior probability the poste-
rior probability of each node. When the Bayesian network model cannot determine and
analyze the perceived data, it is judged as a new hoisting safety risk factor index, and the
data need to be reprocessed in the offline learning stage. The main work of the offline
learning stage is to establish the hoisting safety risk control model based on the Bayesian
network. The specific process includes the determination of the Bayesian network struc-
Sensors 2022, 22, 2522 12 of 24

of each node. When the Bayesian network model cannot determine and analyze the
perceived data, it is judged as a new hoisting safety risk factor index, and the data need to
be reprocessed in the offline learning stage. The main work of the offline learning stage is to
establish the hoisting safety risk control model based on the Bayesian network. The specific
process includes the determination of the Bayesian network structure, the determination of
nodes, the discretization of variables, and the learning of Bayesian network parameters.
The safety risk factors that cannot be identified and processed in the online use stage are
combined with the original Bayesian network structure model to form a new Bayesian
network hoisting safety risk control model and update the model and data fusion module.
In this process, it is necessary to store the data of the Bayesian network parameter learning
for subsequent calls. Through data analysis and calculation, the key degree is obtained. The
safety risk control strategy is formed and fed back to the physical hoisting site. The fusion
of model and data is realized, and the safety of the hoisting site is effectively guaranteed.

5.2. Specific Process of Hoisting Safety Risk Control Analysis

The hoisting operation is a complex, discrete, and irregular system, so it is difficult to
establish
Sensors 2022, 22, x FOR PEER REVIEW the corresponding mathematical model. The amount of data on the construction 13 of 25
site is larger, and the construction monitoring data are difficult to retain and form a
historical database with. Based on the above characteristics of the hoisting operation site,
the Bayesian network (BN) is selected to drive the digital twin model. BN has been used in
in fault
fault diagnosis,
diagnosis, abnormal
abnormal condition
condition identification,
identification, and
and other
other researchfields.
research fields.ItIthas
hasgood
good
advantagesin
advantages inthe
theexpression
expressionofofnonlinear
nonlinearand anduncertain
uncertainrelations.
relations.
BN is an uncertain causal correlation model with strongreasoning
BN is an uncertain causal correlation model with strong reasoningability.
ability.TheThedefini-
defi-
tion ofofa aBayesian
nition Bayesian network
network can be be
can expressed
expressed in the form
in the of aof
form tuple: G  (V,E)
a tuple: G = (V, Emeans ) meansdi-
directed acyclic
rected acyclic graphs.
graphs. V represents
V represents theofset
the set nodesof nodes in a directed
in a directed acyclicE graph.
acyclic graph. representsE
represents
the directed theedge
directed edge connecting
connecting two related two related
nodes nodes
in the in theacyclic
directed directed acyclic
graph, graph,
indicating
indicating
qualitativequalitative
information.information.
 refers θtorefers to the probability
the probability distribution
distribution between between variables,
variables, which
which is essentially
is essentially a conditional
a conditional probability
probability table,table, representing
representing quantitative
quantitative information.
information. The
The Bayesian
Bayesian network
network topology
topology is shown
is shown in Figure
in Figure 6. Taking
6. Taking node node
X4 as Xan4 as an example,
example, X1 andXX1 2
and X2 are
are the parentthe parent
nodes of nodes of XX44, is
X4, and and X4 is affected
affected by X1 and byXX and X2the
2.1Under . Under
premise theofpremise of
given evi-
given evidence, the Bayesian network can update the probability of variables
dence, the Bayesian network can update the probability of variables by probability propa- by probability
propagation or reasoning
gation or reasoning underunder limited,
limited, incomplete,
incomplete, and uncertain
and uncertain conditions.
conditions.

X1
X2 X3

X4 X5

X6 X7

Figure6.6.Bayesian
Figure Bayesiannetwork
networktopology.
topology.

BN
BNisisquantitatively
quantitativelyanalyzed
analyzedbybythe
theD-separation
D-separationcriterion
criterionand
andchain
chainrule.
rule.All
Allroot
root
nodes
nodesareareconditional
conditionalindependent according
independent to the
according toD-separation criterion,
the D-separation and other
criterion, andnodes
other
depend only ononly
nodes depend theirondirect
their parent nodes.nodes.
direct parent According to conditional
According independence
to conditional and
independence
and chain rule, BN denotes the joint probability distribution P(U) of variables U = {A1,
A2..., An} contained in the network (as shown in Equation (7)).
n
P(U )   i 1 P( Ai | Pa ( Ai )) (7)

Pa (Ai) is the parent node of Ai in BN. When new information is given, called evidence
Sensors 2022, 22, 2522 13 of 24

chain rule, BN denotes the joint probability distribution P(U) of variables U = {A1 , A2 . . . ,
An } contained in the network (as shown in Equation (7)).
n
P (U ) = ∏i=1 P( Ai | Pa( Ai )) (7)

Pa (Ai ) is the parent node of Ai in BN. When new information is given, called evidence
E, BN can update the prior probability of the event by reasoning engine (as shown in
Equation (8)).
P(U E) = P(U, E)/P( E) = P(U, E)/∑U P(U, E) (8)

Aimed at the specific scene of prefabricated building hoisting safety risk control, the
operation process of the digital twin safety risk control framework driven by the Bayesian
network is as follows:
(1) Step 1: Transform the hierarchical model determined by the DEMATEL-ISM method
in Section 2.3 into a Bayesian network structure, which requires transforming the
quantifiable index ri in the original hierarchical model into the corresponding variables.
The bottom layer of the hierarchical model is divided into direct incentive variables,
and the other layers are divided into indirect incentive variables.
(2) Step 2: According to the Bayesian network structure, the network nodes are graded
by expert scoring and the continuous data are discretized, as shown in Tables 2 and 3.
(3) Step 3: After the Bayesian network structure is determined, it is necessary to deter-
mine the Bayesian network parameters. The goal of Bayesian network parameter
learning is to determine the conditional probability of nodes in the Bayesian network
model by using prior knowledge combined with the given network topology and
training sample set. The expectation–maximization algorithm (EM algorithm) is ap-
plied to Bayesian network parameter learning based on the above data processing
characteristics.

Table 2. 2 Risk state division of state discrete variables.

Index Corresponding Variables S0 S1

r8 Stability of temporary support stable unstable
r14 Completeness of safety standards complete incomplete
(S0 and S1 are used to represent non-risk free and risky).

Table 3. 3 Risk state division of state discrete variables.

Risk State Division

Index Corresponding Variables
S0 (Good) S1 (General) S2 (Poor)
r1 Technical level of workers [90,100) [60,90) [0,60)
r3 Workers’ safety awareness [90,100) [60,90) [0,60)
r5 Actual load ratio [0,80) [80,100) [100,150)
r6 Sling angle/◦ [30,40) [40,50) [50,60)
r7 Wear rate of hoisting equipment (%) [0,10) [10,40) [40,50)
r9 Quality of Component Production [90,100) [60,90) [0,60)
r10 Strength of hoisting components (MPa) [0,4) [4,9) [9,13.5)
r11 Temporary stacking of components [90,100) [60,90) [0,60)
r12 On-site supervision staffing level [90,100) [60,90) [0,60)
r13 The ratio of participants in safety clearance (%) [90,100) [60,90) [0,60)
r15 Safety measures cost investment ratio (%) [3,5) [1.5,3) [0,1.5)
r16 wind velocity (m·s−1 ) [0,7.9) [7.9,10.8) [10.8,16)
Sensors 2022, 22, 2522 14 of 24

Existing security information monitoring mechanisms are incomplete, resulting in no

large database or complete historical data to refer to. Relying on expert experience is too
subjective. The EM algorithm is a general algorithm for calculating maximum likelihood
function (ML), which is mainly used for Bayesian network parameter learning under
incomplete data. The obtained data samples of hoisting safety risk indicators are recorded
as D = { D1 , D2 , · · ·, Dm }. Each sample Di is composed of unknown variables Xi and
observed variables Yi . The probability distribution function Q( X = xi |Y ) of the residual
variable X is defined, which represents the probability of X = xi when the observation
value is Yi . It means ∑ Q( X = xi |Y ) = 1.
Any sample Di is composed of unknown variables Xi and observation variables Yi . At
this time, it is assumed that the likelihood function L is shown in Equation (9).
m
L(θ ) = ∑ log ∑ P(X = xi , Y = D) (9)
i =1

Assuming that Q( X |Y ) is a convex function with extreme values, according to Jensen

inequality:
m
P( X = xi ,Y = D )
L(θ ) = ∑ log ∑ Q( X |Y ) × Q ( X |Y )
i =1 n
m (10)
≥ ∑ ∑ Q( X |Y ) × log P(XQ=(xXi ,Y =D)
|Y )
i =1 n

The EM algorithm mainly calculates the expected value of the log-likelihood function
according to Q( X |Y ) and continuously optimizes the Q( X |Y ) value to find the L value
that satisfies the maximum value of the log-likelihood function. The EM algorithm is a
continuous iteration process described above [56].
(4) Step 4: According to the data sample, the EM algorithm is used to solve the prior
probability and conditional probability of each node.
(5) Step 5: If the direct inducement variable is in a dangerous state, the probability of the
occurrence of the state at this time is changed to 100%, and as evidence, information
is input into the Bayesian network model for reasoning, and continued in Step 6. If
there is no danger, the data are stored in the offline learning stage to update the prior
probability.
(6) Step 6: Based on the results of Bayesian network reasoning, the key degree of indirect
incentive variables to direct incentive variables is calculated, which makes the factors
that need to be controlled clearer, as shown in Equation (11).
v
u pk
1u
pk ∑
Xt (Ck ) = t ( P0 (C = c | R = r ) − P(C = c )) 2 (11)
k k k k
ck

pk represents the number of discrete states of risk indicators. P0 (Ck = ck | R = r )

represents the posterior probability of Ck in ck state when node R is in r state; P(Ck = ck )
denotes the prior probability of Ck in ck state.
(7) Step 7: In the process of data input into the Bayesian network structure, it is necessary
to determine whether new anomalies occur. If they do occur, Step 1 is entered to
re-determine the Bayesian network structure and update the Bayesian network model
offline.

6. Case Study
Taking a prefabricated building project in Tianjin as an example, this paper verifies the
methods mentioned in the article. The prefabricated building project is located in the leisure
area of Huanggang, Tianjin Binhai New Area. The total building area is 71,172.43 square
meters. The structural type is a reinforced concrete shear wall structure and some pre-
fabricated structural members are used. The project has an assembly rate of 37.5%, a
 Sensors 2022, 22, 2522 15 of 24

maximum height of 54 m, and a maximum number of 18 floors. The prefabricated parts of

the prefabricated project are prefabricated composite floor slabs, prefabricated stair slabs,
and partially prefabricated shear walls.
This project has problems such as prefabricated component management and safety
risk aversion. The construction site involves many tower cranes and many types of com-
Sensors 2022, 22, x FOR PEER REVIEW ponents. The relationship among equipment, components, and environment is 16 of 25
complex.
The construction site is shown in Figure 7. Therefore, it is necessary to study the safety risk
management in the hoisting process of this project to prevent and control the safety risk.

(a) (b)
Figure
Figure7. 7.
Construction site
Construction of of
site thethe
project. (a)(a)
project. BIM model
BIM ofofconstruction
model constructionproject;
project;(b)
(b)On-site
On-sitecon-
construc-
struction.
tion.

6.1.
6.1. Establishment
Establishment of of Digital
Digital TwinTwin Model
Model
On the level of physical entity
On the level of physical entity modeling, modeling, intelligent
intelligent sensing
sensing and
and data
data transmission
transmission
equipment
equipment areare mainly
mainly configured
configured on the
on the fourfour dimensions
dimensions of theof human,
the human, machine,
machine, compo- com-
ponent,
nent, and environment
and environment of the of the hoisting
hoisting site. RFID
site. Active Active RFID
tags tags are embedded
are embedded in the
in the prefab-
prefabricated
ricated componentscomponents
to store to thestore the ID
ID and andattributes
basic basic attributes of the components.
of the components. RFIDare
RFID tags tags
are also configured on the worker’s safety hat to store the identity
also configured on the worker’s safety hat to store the identity and location information and location information
of of
thethe personnel.
personnel. ToTo read
read thethe inclination,
inclination, strength,
strength, position,
position, andand other
other information
information ofof
thethe
component, IMU (inertial measurement unit) and mechanical sensors are installed on thethe
component, IMU (inertial measurement unit) and mechanical sensors are installed on
component.
component. However,
However, thethe information
information needed
needed forfor tower
tower craneequipment
crane equipmentmonitoring
monitoring
(load-weight
(load-weight ratio,
ratio, wind
wind speed,
speed, height,
height, amplitude,
amplitude, etc.)
etc.) is more
is more andand more
more complex.
complex. There-
There-
fore, mechanical sensors, wind speed sensors, height sensors, and other types of sensorsare
fore, mechanical sensors, wind speed sensors, height sensors, and other types of sensors
aremainly
mainlyusedusedfor
fordata
dataacquisition.
acquisition.RFID RFIDreaders
readersare arealso
alsoconfigured
configuredtotoobtain
obtainthe
theposition
posi-
and attribute information of personnel and components in
tion and attribute information of personnel and components in real-time. The sensorreal-time. The sensor accuracy
ac-
or error
curacy usedused
or error above is shown
above is shown in Table 4. In4.the
in Table process
In the process of data acquisition,
of data considering
acquisition, consid-
the process of hoisting activities and the real-time nature of digital twin, the time interval
ering the process of hoisting activities and the real-time nature of digital twin, the time
of sensor data collection is uniformly set to 90 sec, and the redundant data are deleted,
interval of sensor data collection is uniformly set to 90 sec, and the redundant data are
which is also convenient for data analysis at the same time, and more accurate results are
deleted, which is also convenient for data analysis at the same time, and more accurate
obtained.
results are obtained.
Table 4. Accuracy or error of various sensors.
Table 4. Accuracy or error of various sensors.
Sensor Type Accuracy/Error
Sensor Type Accuracy/Error
RFID tag RFID tag detection accuracy: 3–5m
detection accuracy: 3–5m
IMU (inertial measurement unit) accuracy: ±0.1
IMU (inertial
stress and strain sensor
measurement unit) accuracy: ±0.1
system uncertainty: ≤0.2% ± 1 µε
Tower crane black box (height sensor, windstress
speedand strain
sensor, sensor
rotary sensor, weight system uncertainty: ≤0.2% ± 1 µε
System comprehensive error: ≤0.5%
Tower
sensor, inclination crane
sensor, black box
amplitude (height sensor, wind
sensor)
speed sensor, rotary sensor, weight sensor, incli- System comprehensive error: ≤0.5%
Onnation sensor,
the level of amplitude sensor)site modeling, considering the dynamic hoisting
virtual hoisting
process, the virtual model with high fidelity is built by BIM technology. The overall model
On hoisting
of the the level site
of virtual hoisting
is divided into site modeling, considering
the equipment the dynamic
model, prefabricated hoisting model,
component pro-
cess, the virtual model with high fidelity is built by BIM technology. The overall model of
the hoisting site is divided into the equipment model, prefabricated component model,
personnel model, and hoisting environment model, and each model is further divided in
detail. Taking the establishment of the hoisting environment model as an example, it is
 Sensors 2022, 22, 2522 16 of 24

personnel model, and hoisting environment model, and each model is further divided
in detail. Taking the establishment of the hoisting environment model as an example,
Sensors 2022, 22, x FOR PEER REVIEW 17 of 25
it is necessary to consider the display of site layout and the stacking of prefabricated
components. Based on establishing their geometric models, Revit software is used to couple
with the environmental model to form a digital twin model, with high fidelity according
between equipment,
to the position prefabricated
relationship, components,
assembly relationship,and theconstruction
and environmental model.
process At the
relationship
same time, the models such as equipment and prefabricated components
between equipment, prefabricated components, and the environmental model. At the sameare parameter-
ized,
time,and
the amodels
familysuch
library is established
as equipment andto realize the call
prefabricated of subsequent
components modeling of the
are parameterized, and
same scene, and realize rapid modeling, as shown in Figure 8. The modeling
a family library is established to realize the call of subsequent modeling of the of same
the equip-
scene,
ment model and
and realize rapidcomponent
modeling,model is the
as shown in same
Figureas8.above.
The modeling of the equipment model
and component model is the same as above.

(a) (b) (c)

Figure 8. Prefabricated components and tower crane equipment model. (a) BIM model of prefabri-
Figure 8. Prefabricated components and tower crane equipment model. (a) BIM model of prefabri-
cated stairs, (b) BIM model of composite board, (c) BIM model of tower crane.
cated stairs, (b) BIM model of composite board, (c) BIM model of tower crane.
Considering
Consideringthe thespecific
specificscene
sceneofofsafety
safetyrisk
riskcontrol
controlininthe
thehoisting
hoistingprocess,
process,totobetter
better
integrate data with the digital twin model, a hierarchical model
integrate data with the digital twin model, a hierarchical model of safety of safety riskrisk
is added
is addedto
characterize the correlation between various hoisting safety risks. The
to characterize the correlation between various hoisting safety risks. The security risk security risk hier-
archy model
hierarchy established
model in Section
established 2.3 is2.3
in Section generally applicable
is generally applicableand and
can can
be directly
be directlyusedused
in
this case.
in this case.
On
Onthethelevel
levelofofvirtual-real
virtual-realinteraction
interactioncorrelation
correlationmodeling,
modeling,the theimportant
importanttask taskisistoto
establish
establish the bridge of information interaction between the physical hoisting siteand
the bridge of information interaction between the physical hoisting site andthethe
virtual
virtual hoisting site. Virtual-real interaction modeling mainly includes two aspects:one
hoisting site. Virtual-real interaction modeling mainly includes two aspects: oneisis
data
datacollection
collectionand andtransmission;
transmission;another
anotherisisthe
thepackaging
packagingservice
servicefor fordigital
digitaltwin
twinmodels.
models.
InInthis
thiscase,
case,LoRa
LoRatechnology
technologywith withlow-power
low-powerconsumption,
consumption,long-distance
long-distanceand andlowlowcostcostisis
used
used to solve the problem of connecting a large number of sensors to the gateway, andaa
to solve the problem of connecting a large number of sensors to the gateway, and
star
startopology
topologyisisbuilt.
built.The
Thedata
datacollected
collectedfrom
fromthethephysical
physicalhoisting
hoistingsite siteare
areuploaded
uploadedtoto
the LoRa gateway.
the LoRa gateway. The The LoRa gateway uploads the data to the cloud server through thethe
gateway uploads the data to the cloud server through 4G
4G network,
network, andandthethe local
local server
server andand customer
customer endend
cancan access
access thethe site
site information
information through
through the
the cloud
cloud server,
server, as shown
as shown in Figure
in Figure 9. 9.
The operation mechanism of the overall digital twin model of this project and the
partial virtual hoisting site model are shown in Figure 10. Taking the whole BIM model as
a data carrier and visualization platform, the collected data are uploaded through the data
acquisition and transmission system of physical space. After judging the safety risk state,
the node probability of each variable in the Bayesian network is updated. The key cause
analysis of the dangerous state nodes and the feedback of the results are displayed on the
BIM platform to realize the visual analysis and decision-making of safety risks, avoid the
blind judgment of the causes of safety risks, and thus effectively control the safety risks.
SensorsSensors
2022, 22, x FOR
2022, PEER REVIEW
22, 2522 17 of 18
24 of 25

Figure 9. LoRa network architecture.

The operation mechanism of the overall digital twin model of this project and the
partial virtual hoisting site model are shown in Figure 10. Taking the whole BIM model
as a data carrier and visualization platform, the collected data are uploaded through the
data acquisition and transmission system of physical space. After judging the safety risk
state, the node probability of each variable in the Bayesian network is updated. The key
cause analysis of the dangerous state nodes and the feedback of the results are displayed
on the BIM platform to realize the visual analysis and decision-making of safety risks,
avoid the blind judgment of the causes of safety risks, and thus effectively control the
safety risks. Figure 9. LoRa network architecture.
Figure 9. LoRa network architecture.

The operation mechanism of the overall digital twin model of this project and the
partial virtual hoisting site model are shown in Figure 10. Taking the whole BIM model
as a data carrier and visualization platform, the collected data are uploaded through the
data acquisition and transmission system of physical space. After judging the safety risk
state, the node probability of each variable in the Bayesian network is updated. The key
cause analysis of the dangerous state nodes and the feedback of the results are displayed
on the BIM platform to realize the visual analysis and decision-making of safety risks,
avoid the blind judgment of the causes of safety risks, and thus effectively control the
safety risks.

Figure 10. Operating Figure

mechanism of digital
10. Operating twin of
mechanism holistic model.
digital twin holistic model.

6.2. Safety Risk Control Method of Digital Twin Hoisting Driven by Bayesian Network
6.2. Safety Risk Control Method of Digital Twin Hoisting Driven by Bayesian Network
This paper proposes a safety risk control method for prefabricated building hoisting
based on digital twin, focusing on the fusion mechanism of the digital twin model and
field monitoring data. Therefore, this part focuses on the introduction of the Bayesian

Figure 10. Operating mechanism of digital twin holistic model.

6.2. Safety Risk Control Method of Digital Twin Hoisting Driven by Bayesian Network
Sensors 2022, 22, 2522 18 of 24

network to realize the visualization of the risk decision analysis method process and verify
the feasibility of the method.

6.2.1. Data Preparation

Based on the digital twin model of prefabricated building hoisting safety risk control,
according to the method of data perception and transmission on the physical hoisting field
level, the data collection of each safety risk index variable is carried out at the construction
site. Due to the high repeatability of the data collected in one day, we further integrate
the data collected for many consecutive days to obtain the data at three different times,
as shown in Table 4. To ensure the accuracy of Bayesian network parameter learning,
according to the opinions of experts, the integrated data samples are further copied and
expanded. For the three-state variables in the Table 5, 0,1,2 respectively represent the safety
state of the index in good, general, and poor. For 2 state variables, 0 and 1 represent better
and worse, respectively.

Table 5. Data samples.

Time 1 Time 2 Time 3

r1 0 2 1
r3 0 2 1
r5 1 2 2
r6 0 1 2
r7 1 2 1
r8 0 1 1
r9 0 1 1
r10 0 2 1
r11 0 0 0
r12 0 1 1
r13 0 1 0
r14 0 1 1
r15 1 1 0
r16 0 2 1

6.2.2. Visualization Analysis of Security Risk Control Decision-Making

On the visualization analysis of safety risk control decisions, this paper uses GeNIe
software to visualize the Bayesian network. Combined with the safety risk level model
described in Section 2.3, the Bayesian network structure is built. The ID of each node in
the network structure and the number of their states are defined. The EM algorithm is
embedded in GeNIe software. The data collected in Section 6.2.1 are input to learn the
Bayesian network node parameters to obtain the prior probability of each node, as shown
in Figure 11.
At time 3 in Table 5, it can be seen that factor r6 is in a dangerous state. The probability
of r6 in a dangerous state is set to 100% and input into the Bayesian network structure as
evidence. The posterior probability of other nodes is obtained when the sling angle is in
a dangerous state during hoisting, as shown in Figure 12. At this time, the principle of
maximum membership is used to determine the state level of each node. When r6 is in a
dangerous state, the state of the associated indirect inducing factors is consistent with the
state of the data that we collected. Equation (11) is used to solve the key degree Xt (Ck ) of
the indirect inducing factors associated with r6 to the risk factor r6 , as shown in Table 6. By
comparing the values of criticality, the path of safety risk causes is analyzed, and the safety
risk control decision is given with emphasis and rationality.
,, 22,
22, xx FOR
FOR PEER
PEER REVIEW
REVIEW 20 of
20 of 25
25

Sensors 2022, 22, 2522 19 of 24

r16
16 0 2 1

Figure 11.
Figure 11. Prior
Prior distribution
distribution of nodes
nodes *. *.
Figure 11. Priorofdistribution of nodes *.

Figure 12.
Figure 12. The
The posterior
posterior probability
Figure 12.probability
distribution
The posteriordistribution
of each
of each node
node
probability distribution
when
when
of each
thewhen
the
node
slingthe
sling angle
angle is in
in aa is
slingisangle
dangerous
dangerous
in a dangerous
state *.
state *. state *.

* Note: The uppercase letters representing the node number in Figures 11 and 12 cor-
respond to the lowercase letters in Tables 5 and 6, and the uppercase letters in the figure
are used to distinguish them from numbers.
Sensors 2022, 22, 2522 20 of 24

Table 6. The key degree calculation.

Prior Probability Posterior Probability

Code Risk Factor Xt /10−3
S0 S1 S2 S0 S1 S2
r1 Technical level of workers 0.34 0.36 0.29 0.31 0.58 0.11 95.277
r3 Workers’ safety awareness 0.53 0.28 0.19 0.24 0.52 0.24 126.579
r9 Quality of Component Production 0.33 0.57 0.10 0.14 0.69 0.18 79.512
r12 On-site supervision staffing level 0.33 0.52 0.15 0.31 0.64 0.06 50.442
r13 Ratio of participants in safety clearance (%) 0.53 0.28 0.19 0.86 0.11 0.04 133.458
r14 Safety policies and regulations 0.38 0.62 - 0.43 0.57 - 35.355
r15 Safety measures cost investment ratio (%) 0.38 0.42 0.20 0.84 0.06 0.10 187.380
r16 wind velocity (m·s− 1 ) 0.33 0.33 0.33 0.08 0.72 0.20 160.381

* Note: The uppercase letters representing the node number in Figures 11 and 12
correspond to the lowercase letters in Tables 5 and 6, and the uppercase letters in the figure
are used to distinguish them from numbers.
According to the key degree calculated in the above table, the order of the key degree
of indirect inducing factors associated with r6 is as follows: Safety measures cost investment
ratio r15 > wind velocity r16 > ratio of participants in safety clearance r13 > workers’ safety
awareness r3 > technical level of workers r1 > quality of component production r9 > on-site
supervision staffing level r12 > safety policies and regulations r14 . The input of safety risk
measures and the disturbance of the environment have a great influence on the strength
of the connection parts of the hoisting components. In the hoisting site, attention should
be paid to whether the wind speed is suitable for the hoisting operation, and whether
the input of safety measures meets the requirements. When formulating safety strategies,
attention should be paid to emphasizing safety awareness to workers and strengthening
the control of the number of participants in the disclosure of safety technology to improve
the technical level of workers. At this time, the obtained data will be stored in the offline
learning stage of the Bayesian network and used for the next risk state identification and
decision analysis.
After the above analysis and processing of the data related to the safety risk factors of
hoisting operation, the analysis and processing results will be displayed on the BIM-based
visualization platform. Taking the abovementioned factor r6 as an example, the platform
interface will show that r6 (cable inclination) is in a dangerous state, and the position is
marked in the BIM model. The risk-causing paths are listed, and the key causes leading
to the occurrence of factor r6 are analyzed by calculating the key degree. After receiving
the risk factors information, the hoisting site management personnel formulate the safety
risk control scheme and feed it back to the site operation personnel to rectify the site. The
analysis data related to the risk factors will be stored in the cloud database, and the safety
risk control measures formulated by the management personnel will also be uploaded to
the platform for recording, to serve as a follow-up call reference and improve the efficiency
of safety risk management.

7. Conclusions and Future Works

With the development of information technology such as BIM, Internet of Things, and
artificial intelligence, the construction safety management mode has changed from passive
management to active management and real-time monitoring. In the safety management
of underground engineering such as subway and tunnel, the integrated application of BIM
and Internet of Things technology has been formed, but there is still a lack of effective
information interaction mechanism and it ignores the interaction between safety risk factors.
In the field of prefabricated building hoisting safety risk management, most research focuses
on the importance evaluation of safety risk indicators, mainly relying on expert experience.
Sensors 2022, 22, 2522 21 of 24

In the existing research on hoisting safety risk management based on digital twin, the
modeling method of digital twin has not risen to the research on the essence of safety
risk. On the basis of considering the correlation between safety risk factors, the interaction
mechanism of the real-time data and model needs to be improved.
This paper presents a framework for the safety risk control of prefabricated building
hoisting based on digital twin. To ensure the better integration of the digital twin model
and data, the DEMATEL-ISM method is used to establish the hierarchical model of hoisting
safety risk in the process of establishing the digital twin model, so as to express the correla-
tion and risk transfer relationship between safety risks. The Bayesian network is introduced
to visualize the risk control decision model based on digital twin. After determining the
direct and indirect incentive variables of safety risks, the prior probability of each node is
obtained by training the Bayesian network parameters through the EM algorithm. If the
direct inducement variable is in a dangerous state, the occurrence probability is set to 100%
and input into the Bayesian network structure as evidence to obtain the posterior probabil-
ity. The key degree of each indirect incentive variable to the direct incentive variable in a
dangerous state is calculated, and the safety risk control decision is made by comparing
the key degree. The method has been verified in a large prefabricated residential project,
which proves the value and effectiveness of the proposed method. The main contributions
of this study are as follows:
(1) A complete framework for the safety risk control of prefabricated building hoisting
based on digital twin is established. The hoisting activities of prefabricated buildings
are analyzed in detail, and the hoisting safety risk index system is sorted out. Aiming
at the safety risk factors of hoisting, a digital twin model is established from the aspects
of the physical hoisting process, virtual hoisting process, and virtual-real interaction,
and the interaction mechanism of each component of digital twin is expounded.
(2) The research on digital twin in the field of prefabricated building hoisting is raised
to the level of the research on the essence of safety risk. This paper fully considers
the correlation of hoisting safety risk indicators. The hierarchical model of safety risk
is added to the digital twin modeling, and the cause path of hoisting safety risk is
intuitively expressed. The fusion mechanism of the safety risk mechanism model and
real-time monitoring data is established by introducing the Bayesian network, which
lays the foundation for the real-time visualization analysis of hoisting safety risk.
(3) The proposed hoisting safety risk control method based on digital twin is verified in
practical engineering projects. Through the BIM visualization platform, the location of
risk factors, the attribute information of factors themselves, and the cause information
are visualized. It is convenient for hoisting site managers to clearly analyze the safety
risk state and effectively formulate safety risk control measures, avoid the occurrence
of safety risks, and fill in the blanks of the safety risk control decision-making module
in the hoisting safety risk management platform.
This study provides a good solution for a visual analysis of safety risk control de-
cisions of prefabricated building hoisting, and solves the problem that the digital twin
model cannot effectively integrate with real-time monitoring data under the condition of
considering the correlation between safety risks. There are still some limitations to this
method. The existing training sample set is limited, and the intelligent decision system
supporting hoisting safety risk control is lacking. In addition, how to integrate this method
with prefabricated building hoisting safety risk assessment and prediction methods needs
further exploration. In future research, we will continue to expand the scale of data sam-
ples to improve the accuracy of the hoisting safety risk control method based on digital
twinning, and continuously improve the visual display effect and functional modules of
the control platform based on BIM technology.
Sensors 2022, 22, 2522 22 of 24

Author Contributions: Conceptualization, Z.L.; Methodology, Z.L.; Software, Z.L., A.L., Z.S., G.S.
and X.M.; Writing—original draft preparation, A.L.; Writing—review and editing, Z.L. and Z.S.;
Project administration, Z.L.; Funding acquisition, Z.L. All authors have read and agreed to the
published version of the manuscript.
Funding: The research was funded by Beijing Municipal Science & Technology Commission, the
grant number 8202001.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Data sharing not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Opoku, D.-G.J.; Perera, S.; Osei-Kyei, R.; Rashidi, M. Digital twin application in the construction industry: A literature review. J.
Build. Eng. 2021, 40, 102726. [CrossRef]
2. Hsu, P.-Y.; Aurisicchio, M.; Angeloudis, P. Risk-averse supply chain for modular construction projects. Autom. Constr. 2019, 106,
102898. [CrossRef]
3. Tran, H.; Nguyen, T.N.; Christopher, P.; Bui, D.-K.; Khoshelham, K.; Ngo, T.D. A digital twin approach for geometric quality
assessment of as-built prefabricated façades. J. Build. Eng. 2021, 41, 102377. [CrossRef]
4. Zhou, W.; Zhao, T.; Liu, W.; Tang, J. Tower crane safety on construction sites: A complex sociotechnical system perspective. Saf.
Sci. 2018, 109, 95–108. [CrossRef]
5. Shin, I.J. Factors that affect safety of tower crane installation/dismantling in construction industry. Saf. Sci. 2015, 72, 379–390.
[CrossRef]
6. Ministry of Housing and Urban-Rural Development of the People’s Republic of China. The Ministry of Housing and Urban-Rural
Development informed the production safety accidents of housing and municipal engineering in 2019. Eng. Constr. Stand. 2020, 7,
51–53.
7. Riga, K.; Jahr, K.; Thielen, C.; Borrmann, A. Mixed integer programming for dynamic tower crane and storage area optimization
on construction sites. Autom. Constr. 2020, 120, 103259. [CrossRef]
8. Younes, A.; Marzouk, M. Tower cranes layout planning using agent-based simulation considering activity conflicts. Autom.
Constr. 2018, 93, 348–360. [CrossRef]
9. Huang, C.; Li, W.; Lu, W.; Xue, F.; Liu, M.; Liu, Z. Optimization of multiple-crane service schedules in overlapping areas through
consideration of transportation efficiency and operational safety. Autom. Constr. 2021, 127, 103716. [CrossRef]
10. Dutta, S.; Cai, Y.; Huang, L.; Zheng, J. Automatic re-planning of lifting paths for robotized tower cranes in dynamic BIM
environments. Autom. Constr. 2020, 110, 102998. [CrossRef]
11. Tak, A.N.; Taghaddos, H.; Mousaei, A.; Bolourani, A.; Hermann, U. BIM-based 4D mobile crane simulation and onsite operation
management. Autom. Constr. 2021, 128, 103766. [CrossRef]
12. Zhou, Z.; Goh, Y.M.; Li, Q. Overview and analysis of safety management studies in the construction industry. Saf. Sci. 2015, 72,
337–350. [CrossRef]
13. Guo, H.; Yu, Y.; Skitmore, M. Visualization technology-based construction safety management: A review. Autom. Constr. 2017, 73,
135–144. [CrossRef]
14. Liu, Z.; Chen, W.; Zhang, C.; Yang, C.; Cheng, Q. Intelligent scheduling of a feature-process-machine tool supernetwork based on
digital twin workshop. J. Manuf. Syst. 2021, 58, 157–167. [CrossRef]
15. Nguyen, T.N.; Ponciroli, R.; Bruck, P.; Esselman, T.C.; Rigatti, J.A.; Vilim, R.B. A digital twin approach to system-level fault
detection and diagnosis for improved equipment health monitoring. Ann. Nucl. Energy 2022, 170, 109002. [CrossRef]
16. Ou, G.-Z.; Jiao, Y.-Y.; Zhang, G.-H.; Zou, J.-P.; Tan, F.; Zhang, W.-S. Collapse risk assessment of deep-buried tunnel during
construction and its application. Tunn. Undergr. Space Technol. 2021, 115, 104019. [CrossRef]
17. Jeong, J.; Jeong, J. Novel approach of the integrated work & risk breakdown structure for identifying the hierarchy of fatal
incident in construction industry. J. Build. Eng. 2021, 41, 102406. [CrossRef]
18. Raviv, G.; Shapira, A.; Fishbain, B. AHP-based analysis of the risk potential of safety incidents: Case study of cranes in the
construction industry. Saf. Sci. 2017, 91, 298–309. [CrossRef]
19. Ding, X.; Huang, Q.; Zhu, H.; Hu, H.; Liu, Z. Subway Tunnel Disease Associations Mining Based on Fault Tree Analysis Algorithm.
Math. Probl. Eng. 2017, 2017, 2621493. [CrossRef]
20. Wang, Z.Z.; Chen, C. Fuzzy comprehensive Bayesian network-based safety risk assessment for metro construction projects. Tunn.
Undergr. Space Technol. 2017, 70, 330–342. [CrossRef]
21. Guo, Q.; Amin, S.; Hao, Q.; Haas, O. Resilience assessment of safety system at subway construction sites applying analytic
network process and extension cloud models. Reliab. Eng. Syst. Saf. 2020, 201, 106956. [CrossRef]
Sensors 2022, 22, 2522 23 of 24

22. Yang, Z.; Wan, C.; Yang, Z.; Yu, Q. Using Bayesian network-based TOPSIS to aid dynamic port state control detention risk control
decision. Reliab. Eng. Syst. Saf. 2021, 213, 107784. [CrossRef]
23. Chen, F.; Wang, H.; Xu, G.; Ji, H.; Ding, S.; Wei, Y. Data-driven safety enhancing strategies for risk networks in construction
engineering. Reliab. Eng. Syst. Saf. 2020, 197, 106806. [CrossRef]
24. Choe, S.; Leite, F. Construction safety planning: Site-specific temporal and spatial information integration. Autom. Constr. 2017,
84, 335–344. [CrossRef]
25. Lin, S.-S.; Shen, S.-L.; Zhou, A.; Xu, Y.-S. Risk assessment and management of excavation system based on fuzzy set theory and
machine learning methods. Autom. Constr. 2021, 122, 103490. [CrossRef]
26. Li, M.; Yu, H.; Jin, H.; Liu, P. Methodologies of safety risk control for China’s metro construction based on BIM. Saf. Sci. 2018, 110,
418–426. [CrossRef]
27. Hai, N.; Gong, D.; Liu, S. Ontology knowledge base combined with Bayesian networks for integrated corridor risk warning.
Comput. Commun. 2021, 174, 190–204. [CrossRef]
28. Wu, H.; Zhong, B.; Li, H.; Love, P.; Pan, X.; Zhao, N. Combining computer vision with semantic reasoning for on-site safety
management in construction. J. Build. Eng. 2021, 42, 103036. [CrossRef]
29. Xun, Z.-y.; Zhang, L.-M.; Xu, Y.-l.; Zhao, Z.-y. Safety Risk Assessment of Prefabricated Buildings Based on Combination Weighting
Cloud Model. Math. Pract. Underst. 2020, 50, 302–310.
30. Zhao, P.; Xiong, Q.; Tang, Y.-N. Study on safety risk measurement of prefabricated building construction based on SD. Acad. Exch.
Pap. Ind. Archit. 2020, 2, 814–817. [CrossRef]
31. Shen, L.; Tang, Y.-W.; Mou, Y. Safety evaluation of assembly housing component hoisting based on Cloud-BN. Chin. J. Saf. Sci.
2019, 29, 147–153. [CrossRef]
32. Liu, M.-Q.; Li, Y.-P.; Wang, F.; Chen, X.; Li, R.-G.; Li, X.-Z. Safety warning model of assembly building hoisting operation based on
RVM. J. Saf. Sci. China 2018, 28, 109–114. [CrossRef]
33. Mohandes, S.R.; Abdelmageed, S.; Hem, S.; Yoo, J.S.; Abhayajeewa, T.; Zayed, T. Occupational Health and Safety in Modular
Integrated Construction projects: The case of crane operations. J. Clean. Prod. 2022, 342, 130950. [CrossRef]
34. Im, S.; Park, D. Crane safety standards: Problem analysis and safety assurance planning. Saf. Sci. 2020, 127, 104686. [CrossRef]
35. Grieves, M.; Vickers, J. Digital Twin: Mitigating Unpredictable, Undesirable Emergent Behavior in Complex Systems. In
Transdisciplinary Perspectives on Complex Systems; Kahlen, F.J., Flumerfelt, S., Alves, A., Eds.; Springer: Cham, Switzerland, 2017;
pp. 85–113.
36. Glaessgen, E.; Stargel, D. The Digital Twin Paradigm for Future NASA and U.S. Air Force Vehicles. In Proceedings of the 53rd
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference, Honolulu, HI, USA, 23–26 April
2012.
37. Tuegel, E.J.; Ingraffea, A.R.; Eason, T.G.; Spottswood, S.M. Reengineering Aircraft Structural Life Prediction Using a Digital Twin.
Int. J. Aerosp. Eng. 2011, 2011, 154798. [CrossRef]
38. Tao, F.; Liu, W.; Zhang, M.; Hu, T.; Qi, Q.; Zhang, H.; Sui, F.; Wang, T.; Xu, H.; Huang, Z.; et al. Five-dimension digital twin model
and its ten applications. Jisuanji Jicheng Zhizao Xitong/Computer Integrated Manufacturing Systems. Comput. Integr. Manuf.
Syst. 2019, 25, 1–18. [CrossRef]
39. Tao, F.; Zhang, H.; Qi, Q.; Xu, J.; Sun, Z.; Hu, T.; Liu, X.; Liu, T.; Guan, J.; Chen, C.; et al. Theory of digital twin modeling and its
application. Jisuanji Jicheng Zhizao Xitong/Computer Integrated Manufacturing Systems. Comput. Integr. Manuf. Syst. 2021, 27,
1–15. [CrossRef]
40. Zhang, C.; Tao, F. Evaluation index system for digital twin model. Jisuanji Jicheng Zhizao Xitong/Computer Integrated
Manufacturing Systems. Comput. Integr. Manuf. Syst. 2021, 27, 2171–2186. [CrossRef]
41. Zheng, M.; Tian, L. Digital product twin modeling of massive dynamic data based on a time-series database. Qinghua Daxue
Xuebao. J. Tsinghua Univ. 2021, 61, 1281–1288. [CrossRef]
42. Wang, G.; Cao, Y.; Zhang, Y. Digital twin-driven clamping force control for thin-walled parts. Adv. Eng. Inform. 2022, 51, 101468.
[CrossRef]
43. Zhang, X.; Zhang, C.; Wang, M.; Wang, Y.; Du, Y.; Mao, Q.; Lyu, X. Digital twin-driven virtual control technology of cantilever
roadheader. Jisuanji Jicheng Zhizao Xitong/Computer Integrated Manufacturing Systems. Comput. Integr. Manuf. Syst. 2021, 27,
1617–1628. [CrossRef]
44. Liu, Z.-S.; Shi, G.-L.; Sun, J.-J. Digital twinning technology and its application in intelligent construction. Ind. Buildings. 2021, 51,
184–192. [CrossRef]
45. Jiang, F.; Ma, L.; Broyd, T.; Chen, K. Digital twin and its implementations in the civil engineering sector. Autom. Constr. 2021, 130,
103838. [CrossRef]
46. Akula, M.; Lipman, R.R.; Franaszek, M.; Saidi, K.S.; Cheok, G.S.; Kamat, V.R. Real-time drill monitoring and control using
building information models augmented with 3D imaging data. Autom. Constr. 2013, 36, 1–15. [CrossRef]
47. Jiang, W.; Ding, L.; Zhou, C. Cyber physical system for safety management in smart construction site. Eng. Constr. Archit. Manag.
2020, 28, 788–808. [CrossRef]
48. Liu, Z.; Zhang, A.; Wang, W.; Wang, J. Dynamic Fire Evacuation Guidance Method for Winter Olympic Venues Based on Digital
Twin-Driven Model. Tongji Daxue Xuebao. J. Tongji Univ. 2020, 48, 962–971. [CrossRef]
Sensors 2022, 22, 2522 24 of 24

49. Liu, Z.; Xing, Z.; Huang, C.; Du, X. Digital twin modeling method for construction process of assembled building. Jianzhu Jiegou
Xuebao. J. Build. Struct. 2021, 42, 213–222. [CrossRef]
50. Liu, Z.; Meng, X.; Xing, Z.; Jiang, A. Digital Twin-Based Safety Risk Coupling of Prefabricated Building Hoisting. Sensors 2021, 21,
3583. [CrossRef]
51. Khajavi, S.H.; Motlagh, N.H.; Jaribion, A.; Werner, L.C.; Holmstrom, J. Digital Twin: Vision, Benefits, Boundaries, and Creation
for Buildings. IEEE Access 2019, 7, 147406–147419. [CrossRef]
52. Jiang, H.; Ding, G.; Zhang, J. Evolution and Operation Mechanism of Digital Twin Shopfloors. Zhongguo Jixie Gongcheng. China
Mech. Eng. 2020, 31, 824–832. [CrossRef]
53. Zhang, C.; Zhao, T.-s.; Jiang, L.; Wang, X.; Tang, J.-J. Selection of structural safety monitoring parameters and layout of measuring
points of tower crane. J. Saf. Sci. China 2021, 31, 112–118. [CrossRef]
54. Chang, X.; Jia, X. Knowledge-driven proactive management and control method for digital twin enabled aircraft overhaul
shop-floor. Jisuanji Jicheng Zhizao Xitong. Comput. Integr. Manuf. Syst. 2021, 27, 390–402. [CrossRef]
55. Wang, J.-W.; Lu, C. Analysis of the cause mechanism of assembly building hoisting accident from the perspective of relevance. J.
Saf. Environ. 2021, 21, 1158–1164. [CrossRef]
56. Zhang, W.; Jiang, Y.; Yin, G.; Yu, L. Handling missing values in software effort data based on naive Bayes and EM algorithm.
Xitong Gongcheng Lilun yu Shijian. Syst. Eng. Theory Pract. 2017, 37, 2965–2974. [CrossRef]