Towards the “Third Wave”: An SCO-enabled Occupational Health and Safety
Management System for Construction
Yuhan Niu, Weisheng Lu, Fan Xue, Diandian Liu, Ke Chen, Dongping Fang, and Chimay
Anumba
This is the peer-reviewed post-print version of the paper:
Niu, Y., Lu, W., Xue, F., Liu, D., Chen, K., Fang, D., & Anumba, C. (2019). Towards
the “third wave”: An SCO-enabled occupational health and safety management system
for construction. Safety science, 111, 213-223. Doi: 10.1016/j.ssci.2018.07.013
The final version of this paper is available at: https://doi.org/10.1016/j.ssci.2018.07.013. The
use of this file must follow the Creative Commons Attribution Non-Commercial No
Derivatives License, as required by Elsevier’s policy.

Abstract
Occupational health and safety (OHS) is of the utmost concern in the construction sector. For
decades, researchers and practitioners have endeavoured to enhance construction OHS
performance through various measures ranging from “hard” technologies (in this paper, the “first
wave” of construction OHS management) such as provision of personal protective equipment, to
the more recent “soft”, managerial approaches (the “second wave”) such as fostering a safety
culture. Although considerable improvements have been made in construction OHS, the general
sentiment is that construction remains one of the most dangerous sectors, warranting more
innovative or even revolutionary approaches. This research seeks to develop a smart construction
object (SCO)-enabled OHS management system. The central tenet of the system is that artificial
intelligence (AI), as the art of creating machines that perform functions that require intelligence
when performed by people, represents a direction of the “third wave” in construction OHS
management. The system embraces emergent SCOs and harnesses the power of their smart
properties of awareness, communicativeness, and autonomy. The system is demonstrated and
validated in real-life construction practice and a controlled lab test with a tower crane, the cause
of many construction-related injuries and fatalities, as the subject. It is found that the SCO-enabled
OHS management system can identify dangerous situations and respond to them autonomously.
This research suggests that smarter construction, through incorporation of AI in particular, is a
direction of much promise in terms of improving construction OHS.
1

Keywords: Occupational health and safety (OHS); construction safety; smart construction
object (SCO); tower crane; artificial intelligence (AI)
1. Introduction
According to the International Labour Office (2001), OHS management refers to a coordinated
and systematic approach undertaken by an organization to protect the safety and health of all
members through prevention of work-related injury, illness and disease. Despite strenuous efforts
to manage OHS in the construction industry, its safety performance is still alarmingly poor. In the
United States, for example, construction accounted for no more than 5 per cent of the workforce
but 20 per cent of occupational deaths in the years 2003 to 2013 (National Safety Council 2015).
This disproportionate pattern is similar or worse in developing economies (Raheem and Hinze
2014). It is estimated that a total of 60,000 construction fatalities occur every year around the world;
on average, one every nine minutes (Somavia 2005). The construction industry, while
“instrumental in influencing human health, economic activities and social behaviour as well as
cultural identity and civic pride” (Pearce 2003), is also one of the most dangerous.
Efforts of researchers and practitioners to improve construction OHS management have been
ongoing across several historical stages of development. The early days of OHS management can
be characterised by a reliance on “hard” protection, using personal protective equipment (PPE) as
a physical buffer between users and hazards. This was termed the “first wave” of OHS
management in this paper. With the growing attentions on the root causes of accidents, a “second
wave” of construction OHS management arises with the emphasis on safety training and safety
education to reduce unsafe behaviours and dangerous situations. While considerable progress has
been achieved during the first and second waves, often insufficient for making rational decisions
and taking appropriate action when dangers suddenly emerge, making alerts ineffective. As a
consequence of such limitations, construction around the world is witnessing stagnant OHS
management. Inspired by smart technologies (e.g., artificial intelligence [AI], robotics) in other
sectors, the construction industry is also vigorously exploring how these technologies as having
the capacity can provide a revolutionary approach to improving OHS management in construction.
This AI-based OHS management is to be argued as the “third wave” development in construction.
2

However, the understanding of this “third wave” is in its infant stage. For example, there are
exhortations to develop full “AI”, or totally disruptive solutions to construction OHS management,
while the take-up of these advocacies is rather low in reality.
Building on previous studies of smart construction objects (SCOs), the primary aim of this research
is to (a) develop a SCO-enabled construction OHS management system, and (b) argue that SCOs
augment construction resources with a “narrow AI” should be the “third wave” development of
construction OHS management. Central to the “third wave” of construction OHS management is
not completely departing from existing OHS management methods. While acknowledging the
adoption of PPE and importance of preventive strategies, an active AI-based solution is proposed
with the deployment of smart construction objects (SCOs). While SCOs provide OHS-related
decision-making information to human decision-makers, they can also talk to each other directly.
Thus, actions that can eliminate a hazard at source can be taken by SCOs promptly and
autonomously; that is, without necessarily involving human decision-makers in the loop.
The remainder of this paper comprises seven sections. Subsequent to this introductory section is a
review of the literature on the revolution of construction OHS management. By introducing the
definition and properties of SCOs, the potentials and advantages of using SCOs for OHS
management are presented in Section 3. In Section 4, the architecture and workflow of an SCOenabled OHS management system is presented. With a tower crane selected as the target, the
system is prototyped and validated in the context of a real-life on-site project in Section 5. A lab
experiment is also presented demonstrating the system and how the SCO-enabled OHS
management framework could be used in management strategy development. Section 6 discusses
the prospects and challenges of the SCO-enabled OHS management framework, and conclusions
are drawn in Section 7.
2. The three “waves” development of construction OHS management
The early days of OHS management can be characterised by a reliance on “hard” technologies,
which is termed as the “first wave” OHS management in this paper. The protection is mainly relied
on physical buffers provided by personal protective equipment (PPE) such as safety helmets, boots,
gloves, and goggles (Hinze et al. 2013). Fundamentally, PPE work in the way of imposing a barrier
3

between the user and the working environment, thus reducing the user’s exposure to hazards
including physical, electrical, heat, chemicals, biohazards, and airborne particulate matter. The
“first wave” OHS management is not uniquely used in construction. A cross-sectoral analogue is
the automotive industry, where car manufacturers have adopted physical protection (e.g. safety
belts, air bags, and anti-lock braking systems) to protect drivers and passengers.
Despite the widespread applications of PPE in construction and continuing advances in
technological approaches to its provision, a general limitation of PPE is that it does not eliminate
hazards at their source (Holt 2008). Thus, significant efforts have been directed in recent years to
investigating the root causes of accidents. Heinrich (1941), a pioneer in accident causation
investigation, developed the domino theory, which states that injuries occur as a result of linear,
sequential factors. Building on this theory, enriched causation models incorporate factors such as
unsafe conditions, unsafe behaviour and worker response (Abdelhamid and Everett 2000).
Managerial approaches to tackling these causes have also been explored, such as developing a
behaviour-based safety system (Choudhry and Fang 2008), conducting safety training
(Hadikusumo and Rowlinson 2002), and fostering a safety culture (Mohamed 2003) and climate
(Hahn and Murphy 2008). These efforts echo developments attributing accidents largely to
overload of human capabilities, both physical and psychological, such as the human-error
causation model (Petersen 1984) and the DeJoy (1990) model. While unavoidably intertwined with
traditional technological approaches, such efforts focus on “soft” aspects and can be collectively
referred to as the “second wave” in OHS management. As in the case of “hard” technologies, an
emphasis on “soft” aspects can also be found in the automotive industry, for example through safedriver education and the enforcement of strict traffic rules and regulations.
While human error-related accidents can be reduced with safety training and safety culture
development, they cannot be completely eliminated due to unexpected conditions such as fatigue
or sudden site distractions (Fang et al. 2015). Studies have been made for safety management
systems using emerging technologies, most of which focus on detecting hazardous conditions and
issuing alerts. For example, sensing technologies such as Radio Frequency Identification (RFID)
(Lu et al. 2011; Flanagan et al. 2014) and wireless networks such as ZigBee have been used to
capture real-time construction site conditions (Wu et al. 2010), while cyber-physical systems have
4

been developed to model the complexities of construction safety (Yuan et al. 2016). Alerts can be
issued when people enter pre-defined danger zones (Yang et al. 2012) or are too close to moving
objects (Teizer et al. 2010).
However, the safety protection provided by these technologies is imperfect. Although timely alerts
can be provided, in-time mitigations and actions in response to dangerous situation still largely
rely on humans. Researchers have theorized OHS management as decision making, recognizing
that the rationality of human decision-makers (e.g. safety managers and construction workers) is
generally bounded by a “triangle of limits” (Simon 1976): available information, cognitive ability,
and finite amount of time. The latter is often insufficient for making rational decisions and taking
appropriate action when dangers suddenly emerge, making alerts ineffective. Thus, a more
intelligent, in-time solution is desired to manage OHS events proactively and promptly.
The development of construction OHS management, toward the next wave, could draw inspiration
from the automotive industry. Smart systems such as self-parking and collision prevention
assistants are now embedded in cars to improve driving safety, for example by detecting hazardous
conditions and alerting drivers. These smart systems are enhanced with artificial intelligence (AI)
in auto-pilot systems (e.g. as in Tesla vehicles) (Kessler 2015) and autonomous vehicles (e.g.
Apple self-driving cars) (Harris 2015). Since movement on the road is no less complex than on a
construction site, there are no barriers to the exploration of AI in construction OHS management.
It is thus proposed the “third wave” of construction OHS management, in this study, subscribes to
AI-based solutions. It acknowledges that human beings are not infallible, but rather, show
deficiencies (such as being slower and more error-prone) when compared with AI in processing
information and making prompt actions (Sterman 1989; Reason 2000).
3. SCOs for OHS management
Proposed by Niu et al. (2015), smart construction objects (SCOs) represent a new way of capturing,
processing, and communicating information to support decision making in construction. SCOs are
“construction resources (e.g., machinery, tools, devices, materials, components, and even
temporary or permanent structures) that are made smart by augmenting them with sensing,
processing, and communication abilities so that they have autonomy and awareness, and can
5

interact with the vicinity to enable better decision making” (Niu et al. 2016). Instead of introducing
a completely new system to construction sites, an SCO-enabled management system relies on
construction objects (such as machines, materials and components) already involved in the
construction process. Without compromising their original appearance and function, these objects
are augmented with smart and interconnected properties. For example, a smart excavator may be
able to locate and report its real-time position without demanding extra room while still performing
the excavation job.
The three core properties of SCOs, awareness, communicativeness, and autonomy, refer to SCOs’
abilities in sensing, data exchange, and action-taking, respectively (Niu et al. 2015). Each core
property is further categorized into sub-properties with different functions (elucidated by a triaxial diagram and summative table in Fig. 2), the utilization of which allows the potentials of SCOs
for OHS management to be achieved. For example, by applying activity awareness, SCOs could
help record the number of times and the frequency of machine operations. Comparatively, policy
awareness enables SCOs to detect whether there is a break of limit in loading or other critical
factors. The SCOs’ communicativeness ensures that these conditions are conveyed to people
comprehensively and in a timely manner, either passively or proactively. In addition, depending
on the type of autonomy, SCOs have the potential not only to issue alerts but also to take action in
case of emergencies.

Fig. 2. The core properties of SCOs
6

Together, the smart properties of SCOs offer a new avenue for advancement of construction OHS
management that is not completely departing from existing OHS management methods while
adding AI-based values. From the perspective of the “first wave” of OHS management, using
SCOs is not an abandon of PPE. Since SCOs is primarily making existing construction recourses
smarter, the functions of SCOs is still “lodged in” PPE and other construction objects. Looking at
SCOs from the “soft” OHS strategies, it focuses on the dangerous situations as one of the leading
causes for accidents and injuries. The application of SCOs aims to take active and preventive safeguarding actions when dangerous situations are detected. Nevertheless, rather than comprising a
new, ambitious centralized system with artificial general intelligence, or “full AI”, capable of
performing any human intellectual task (Kurzweil 2005), SCOs could augment construction
resources with a “narrow AI” that equals or exceeds human intelligence with regards to specific
tasks. The rationale and workflow of SCOs will be articulated in details as follows in the SCOenabled OHS management system.
4. The SCO-enabled OHS management system
In this paper, a multi-layered SCO-enabled OHS management system is proposed. The architecture
of the system is shown in Fig. 3. At the shopfloor layer are the construction objects (e.g. precast
facades or machinery) that are augmented into SCOs. A smart core integrating various sensors,
communication modules, and actuators (e.g. GPS, IMU, Bluetooth, and LiDAR) is installed in or
attached to the construction objects, endowing them with the three core SCO properties of
awareness, communicativeness, and autonomy. Dangerous situations to be detected and the SCObased solutions are stored in the respective databases, which are centrally managed in the smart
management platform (SMP). Pre-existing conditions of dangerous situations can be input into the
event database, which could be continuously expanded and updated with newly emerging industryreported events. Based on updated conditions in the database, relevant SCO solutions can be
revised to guide the applications in the top layer. The SMP also incorporates a BIM-oriented
database so as to relate the conditions to ongoing projects and identify the possible impacts of
these conditions on overall project performance. An online monitoring interface is established in
the SMP for visualization purposes with Cesium (ver. 1.24). The smart applications enabled by
the sensing, communicating, and action-taking abilities of SCOs are specified in the application
7

layer. These applications will be designated to sensors and actuators based on the application
scenarios, which are directly executed by SCOs. Each application is also supported by the SMP,
which can provide human decision-makers with visualized data and prompt alerts.

Fig. 3. The architecture of the SCO-enabled OHS management system
A generic SCO workflow in dealing with the dangerous situations is outlined Fig. 4. It is similar
to the logic behind the software that makes the OHS management system operable. The workflow
is “generic” in the sense that it is expected to be sufficiently inclusive to embrace all sorts of typical
scenarios in construction OHS management. The conditions of dangerous situations (e.g. hoisting
materials) that may induce accidents or injuries are constantly sensed using SCO awareness. For
situation, there will be a series of pre-set conditions against which to gauge whether the condition
hits a threshold or not. If not, the SCOs will continue sensing. When an condition sensed by the
8

SCO is diagnosed as dangerous, respective communicativeness and autonomy solutions will be
triggered. In the SCO-based OHS management system, each SCO-based solution is assigned a set
of communicativeness and autonomy sub-modules. The communicativeness sub-modules will
communicate the diagnosed situation to the SMP, searching suitable autonomy sub-modules. Clear
rule-based decisions such as halt or force quit can be autonomously made by SCOs without
necessarily involving human decision-makers in the loop. Where no active autonomy is available,
passive autonomy will be triggered to alert human decision-makers. Records of emerging
conditions are constantly logged and pushed to the SMP, assisting further data analysis. Compared
with the human decision-making process, SCO awareness and decision-making can occur
instantaneously, making the subsequent SCO-enabled reaction concurrent or near concurrent. This
SCO-enabled concurrence, vis-à-vis most prevailing “ex-ante” training or “ex-post” analyses,
could more effectively prevent dangerous situations from developing into serious accidents.

Fig. 4. A generic SCO-enabled OHS management workflow
A few examples of how dangerous situations are managed according to the workflow are provided
in Table 1. While by no means exhaustive, these are based on most-commonly occurring and mostoften addressed hazardous events that are prone to deteriorate into accidents identified from the
literature (Cambraia et al. 2010; Green and Tominack 2012; Wu et al. 2010; Yang et al. 2012) and
reports (OSHA 2011a, 2011b, 2011c, and 2011d). Against each listed event, the potential SCObased solutions that can be deployed and the associated workflow are demonstrated and explained
as follows.
Table 1. The examples of events to be managed by the SCO-based OHS management system
9

Events to manage

Pre-set condition

Dangerous

for awareness

situations

Distance detection

Distance ≤

Policy

maintain safe

between worker

buffer

awareness

distance between

and restricted area

distance

(a) Failure to

SCOs
Smart PPE

Awareness

Communicativeness

Autonomy

Triaxial diagram of
SCOs

Information push

Passive
autonomy

on-foot worker and
restricted area
(b) Failure to

Smart PPE and

Distance detection

Distance ≤

Policy

maintain safe

moving parts of

between worker

buffer

awareness

distance from parts

machine/vehicle

and moving parts

distance

(c) Incorrect

Smart machine

Critical factor

Factor value

Mixed

operation/Improper

and equipment

sensing and

≥ threshold

awareness

use of machine for

operation process

± buffer

(policy

critical procedures

detection

range;

awareness

incorrect

and process

operation

awareness)

Information push

Active
autonomy

of machine/vehicle

Information push

Mixed
autonomy

procedures
(d) Failure to

Smart

Checking the total

Time /

Activity

Mixed

Mixed

check/maintain

equipment

time / frequency

frequency ≥

awareness

communicativeness

autonomy

of usage

threshold

equipment on time

10

(e) Critical

Smart PPE

Sensing the

Factor value

Policy

environmental

critical

≥ threshold

awareness

factors beyond

environmental

± buffer

human-bearing

factors

range

Information push

Passive
autonomy

threshold

11

(a) Failure to maintain safe distance between on-foot worker and restricted area
Most dangerous situations related to falls, electrocution, and “caught-in between” events are
associated with workers getting too close to hazardous areas such as edges at high levels, trenches
without shoring, and working radii of derricks or cranes. In these circumstances, the personal
protection equipment (PPE) of workers, such as safety helmets, could be made into SCOs able to
sense the real-time location of these workers at all times. Applying SCO policy awareness,
geographical location is set as a threshold with a buffer range in the periphery. When a worker
steps into the buffer range, the smart PPE item issues an alarm via passive autonomy, alerting the
worker so that he/she proceeds no further.
(b) Failure to maintain safe distance from moving parts of machines/vehicles
Occupational Safety and Health Administration (OSHA) reports reveal that workers can easily be
struck when passing a machine/vehicle operation without keeping a safe distance, whether due to
carelessness of the worker or the operator. If a worker stands within the swing range of a moving
part of a machine, he/she can be caught between the machine and a solid object, such a wall or
another piece of equipment. To manage such scenarios, both workers’ PPE and the moving parts
of machines/vehicles can be transformed into SCOs able to constantly calculate the distance
between them. These machines/vehicles can be augmented with electrical brakes which activate
when the SCOs detect border-crossing passers-by, thereby preventing accidents.
(c) Incorrect operation/ Improper use of machines for critical procedures
Turning construction equipment and machines into SCOs enables prevention of their incorrect
operation and improper use. Mixed awareness, mixed autonomy and information push can be
applied to cover a diverse range of dangerous situations. For example, loading capacity, rotation
angle, and lifting height of a tower crane can be set as policy awareness thresholds to prevent
overloading or hoisting in multiple directions simultaneously. For non-critical procedures, the
operator can be alerted via passive autonomy; at the same time, standard procedure instructions
can be pushed to the operator. In the case of critical procedures, equipment can be compulsorily
locked or turned off until the necessary corrections are made.

12

(d) Failure to check/maintain equipment on time
Failure to undertake regular examination and maintenance of equipment, especially of heavy
machinery, has significant safety and cost implications for construction. When items of equipment
are turned into SCOs, activity awareness can sense and assist in the precise recording of each
activity related to their use or handling, such as picking up, turning on, and operating. A typical
case of activity awareness is presented in Fitton et al. (2008), where a pay-per-use function was
enabled by sensors in road patching machines. For regular examination and maintenance purposes,
a mixed communicativeness is chosen. Here, the SCOs actively push information at regular
intervals, while the machine use record can be pulled out manually when needed. Alerts are made
via passive autonomy when maintenance is required based on handling time. If no subsequent
maintenance is undertaken, the SCOs will use active autonomy to intervene by forcing users off
the equipment or locking it into standby mode.
(e) Critical environmental factors
Proposals for environment-based construction OHS management solutions have been made since
SCOs were first discussed in Niu et al. (2015). SCOs enable monitoring of critical environmental
factors that are hazardous to workers or machine operations. Monitoring non-perceptible factors
such as toxic vapours, for example, can reduce the occurrence of diseases such as pneumoconiosis
or asbestos-related lung cancer. For critical environmental factors, maximum human-bearing
thresholds can be input into smart tools and PPE. Augmented with policy awareness, these SCOs
can sense environmental conditions and, if conditions are below the threshold, perform
information push to the management platform for monitoring. If the threshold is crossed, the SCOs
can use passive autonomy to alert workers.
5. Demonstration and validation
5.1 Background
To demonstrate and validate the proposed SCO-enabled OHS management system, the operation
of tower cranes was explored. Tower cranes hoist and transport a variety of loads near and above
construction workers, often working in crowded conditions and occasionally with overlapping
work zones. The use of tower cranes can increase safety risks on sites that are already inherently
hazardous (Shapira and Lyachin 2009, Raviv and Shapira 2018), as well as threatening pedestrians
13

(Shepherd et al. 2000). Estimates suggest that cranes are involved in up to one-third of all
construction and maintenance fatalities (Neitzel et al. 2001); therefore, the importance of tower
crane management in improving overall construction safety performance cannot be overemphasized.
Prevailing OHS management practice in tower crane operations is highly dependent on individual
experience rather than scientific evidence. While experience is extremely important in construction,
overconfidence in this experience means that evidence-based decision making is lacking. Ongoing
tower crane operation conditions are reported and recorded by contractors sporadically, if at all.
Although some studies have used data obtained from statistical reports as a reference for accident
prevention (e.g. Chi and Han 2013, Tsang et al. 2017), such data may be unreliable due countless
unreported incidents; in addition, such statistics are unable to provide information on root causes ,
as well as being questionable predictors of accidents (Shapira and Lyachin 2009). Post-accident
analysis also has limited power in preventing recurrence. The proposed SCO-enabled OHS
management system offers a means of capturing and recording more reliable, real-time or near
real-time, comprehensive data covering target conditions in tower crane operations. It also
provides impetus for AI applications providing in-time mitigation of dangerous situations in
construction OHS management.
5.2 Field test
Discussions with construction managers revealed five commonly occurring dangerous situations
related to tower crane operation (see Table 2), all of which could lead to serious accidents if not
handled properly. Hook over-height could cause equipment damage when hoisting heavy loads, or
in extreme cases tip the crane. Crossing of the jib and trolley into restricted areas may result in
collisions with surrounding machinery, buildings, or people working at heights. Unbalanced
hoisting and lifting heavy weights over dynamic restricted areas (e.g., personnel work zones, areas
containing assets and equipment) are both serious, dangerous situations which could easily cause
objects to fall as loads become out of control. The conditions to be sensed and criteria for
alert/action for each dangerous situation in our field test are listed in Table 2.

14

Table 2. Dangerous situations and related tower crane operation conditions managed in the field
test
Dangerous situations

Real-time data of conditions

Criteria for alert/action

(1) Hook over-height

Height of hook

Hooking height ≥ height threshold

(2) Jib/Trolley/Load

Slewing angle of jib, distance of

Jib slewing angle entering a

crossing pre-set restricted

trolley, swing motions of load

constant range of angles

(3) Jib/Trolley/Load

Slewing angle of jib, distance of

Jib slewing angle entering a

crossing dynamic restricted

trolley, swing motions of load and

constant range of angles, and heavy

areas

its geo-position in relation to

load moving over dynamic

moving personnel and vehicles in

restricted zones of personnel and

the zone

vehicles

Motions of jib, trolley, and hook

Simultaneous motions of jib,

areas

(4) Unbalanced hoisting

trolley, and hook
(5) Over-swing of load

Swing motions of beam

Swing angle ≥ swing threshold

Several smart cores were developed for the field test, each consisting of a microcontroller, an
inertial measurement unit (IMU), a GPS module, a barometer, an anemometer, and a global system
for mobile communication (GSM) module. These smart cores were mounted to the key
components of a tower crane and the hoisted object to make them smart. No prior knowledge
existed regarding where to mount the smart cores, or what to collect to sufficiently capture tower
crane operations and subsequently identify dangerous situations for alert and intervention purposes.
Therefore, this process was discussed with site managers and conducted through trial and error.
Figure 5 shows a feasible installation scheme using a smart core adopted, without suggesting it is
the only and best scheme to do so. The figure shows the smart core installation positions, while
the table shows what data is collected through which sensing modules for monitoring and
diagnosing specific dangerous situation. At this point in the field test, the conventional tower crane
and its materials had been turned into SCOs through the use of smart cores (c.f. Fig. 3) and it could
now function with extra smartness through awareness, communicativeness, and autonomy.

15

Sensing Module

Monitoring Index/Data

Indicators for dangerous
events

GPS

 a) Geographical location

Distance of trolley

Accelerometer (IMU)

 b) Acceleration

Slewing angle of jib

Magnetometer

 c) Magnetic angle

Barometers

 d) Barometric pressure

Heading angle of beam

Gyroscope (IMU)

 e) Roll, pitch, and yaw angles

Swing motions of beam

 Height of hook

Fig. 5. An illustration of the smart mounted to the tower crane
The tower crane was in use on the site of a high-rise residential development project in the New
Territories, Hong Kong. The smart cores collected and updated information on the real-time
operation conditions of the tower crane and the materials hoisted (i.e. four precast beams) every 3
seconds throughout the operation. This formed a big data set, 1,270 sets of well-structured records,
an excerpt of which is shown in Fig. 6.

16

Fig. 6. Sample data captured by the smart cores
A smart management platform (SMP) (c.f. Fig. 3) was developed to visualize the operations and
the conditions of the smart tower crane in a real-time manner. As shown in Fig. 7, the SMP has a
graphic user interface (GUI). The background is a cyber construction site reproduced from the real
site using a WebGL engine Cesium and Microsoft Bing Map. The building information model was
obtained and reproduced in the cyber system. A 3D tower crane model was created as the “cyber
twin” of the target crane positioned properly on the site to illustrate the real-time operations of the
crane. Based on the live data returned by the smart cores, the SMP could reproduce and visualize
the motions of the target tower crane simultaneously, with additional aerial and front views for
easier perception.

17

Fig. 7. The smart management platform (SMP) for tower crane safety management
In parallel with the cyber tower crane operation is a visualisation of the big data transmitted back
from the smart cores. Fig. 8 illustrates the visualized dataset for the field test. To remap the status
of the crane and to identify the dangerous situation, a finite-state machine (FSM) model, with the
six states idling, hoisting, slewing, hovering, installation, and resetting, was developed. The change
from one status to another required one (or more) speed (or angular speed or velocity) surpassing
the threshold(s) pre-defined. For example, the state changed from “idling” to “hosting” at second
8 (S8) when hook velocity > 0.2 m/s. Some state changes are not directly reversible. For example,
after changing to “hoisting,” the state remained during s16~s24 even though the hook had stopped
elevating (see Figure 8). The detections of dangerous situations were also based on the velocities
or angular speeds. For example, the criterion of identifying unbalanced hoisting was jib angular
speed > 0.3°/s, trolley speed > 0.15 m/s, and hook speed > 0.2 m/s simultaneously. During the
installation of beam B1-46 (14:00:07 to 14:07:11, 7 November 2016), the two dangerous situations
of unbalanced hoisting and load crossing dynamic restricted zone were sensed and alerts were sent
directly to the on-site operator and site manager via text SMS. When referring the identified status
of crane and dangerous situations back to Fig. 8, users can observe the events with a highlighted
focus. The parallel records of jib, trolley, and hook motions, and the heading direction and swing
angle of the beam, can reveal to a safety manager the exact motions of both the crane and the beam.

18

Fig. 8. Visualized action patterns and alerts of dangerous situations
5.3 Lab test
While passive autonomy was successfully achieved in the field test, the exercise of active
autonomy, such as execution of a halt action, was untested. After several rounds of negotiation
with the cooperating construction company, the active autonomous control was still perceived as
non-compliant with existing codes of practice (Irani and Kamal, 2014). Hence, a further test was
conducted in a controlled lab environment to demonstrate and validate the feasibility of active
autonomy. A model tower crane capable of emulating actual tower crane movement was
assembled with LEGO®. A servo motor was used to control its movements both clockwise and
anti-clockwise and at different speeds. The same smart core used in the field test was attached to
the main jib of the model tower crane to control its motions so as to prevent dangerous situations
developing into accidents.
The lab test focused on one specific dangerous situation identified in the field test: the jib crossing

a restricted area. The restricted area is for on-foot workers to safely work within or passers-by to
walk through. In the lab test, if the jib of the smart tower crane moved into the buffer range of the
19

restricted area, the smart tower crane was to autonomously halt the operation. A cyber 3D tower
crane was developed in an online monitoring interface and linked to the LEGO® tower crane to
visualize the crane motion in real-time. The restricted area pre-determined in the LEGO® model
was also converted to the online monitoring interface at the same scale. The LEGO tower crane
was initially set out of the restricted area and moved steadily towards the restricted area (Fig. 9a).
When the jib was out of the restricted area, it operated normally with no alert triggered. On
touching the buffer range at one side of the restricted area, the policy awareness of the smart core
diagnosed the condition as a dangerous situation, triggering the amber alert in the SMP (Fig. 9b).
When the jib was entirely within the buffer range, the alert was continuously triggered as the
dangerous situation was not resolved. When the jib touched the boundary of the restricted area
when swinging across the buffer range, the smart core instantly reacted by pausing the motor,
stopping the jib motion (Fig. 9c), and a red alert was triggered.

a

b

c

Fig. 9 The active autonomy lab test for the SCO-enabled OHS management system
20

In order to test the precision of the halt reaction, the rotation angles sensed by the smart core were
compared with the actual rotation angles. These were obtained by a rotary sensor tied to the motor
throughout the whole process. The degree of difference was measured using the root mean squared
error (RMSE) as shown in Equation (1).
1

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 = � ∑𝑛𝑛𝑖𝑖=1(𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖 − 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑖𝑖 )2
𝑛𝑛

(1)

Data from the gyroscope, accelerometer and the magnetometer were fused through the Kalman
filter. Based on 87 sets of data, the value of RMSE was 1.76 degree, indicating that the reaction of
the smart tower crane was ±1.76 degree ahead of or lagging the movement of the jib; acceptable
in the controlled lab environment. The lab test supplements the field test by demonstrating the
automatic control potential of SCOs. If improper operation is yet to be manually stopped, the smart
core can autonomously control the dangerous condition, thus preventing it from developing into a
serious accident.
6. Discussion
SCOs, with their core smart properties of awareness, communicativeness, and autonomy, present
a new opportunity to improve OHS management in the construction industry. Compared with
traditional OHS management systems that collect and analyse after-accident data, the SCOenabled OHS management system inherits several advantageous functions from existing proactive
technologies (Fang et al 2016, Teizer et al. 2010, Yang et al. 2010) such as worksite monitoring,
hazard detection, alerting, and data visualization. Unlike those of traditional OHS management
systems, these functions offer round-the-clock monitoring and objective record taking. Another
important distinguishing feature of the SCO-enabled OHS management system is its autonomy.
The system has potential to prevent dangerous situations from developing into fatal accidents by
taking active and prompt actions in conditions that would overload human thinking and reacting
abilities. Existing studies on construction OHS management have investigated just one or two
specific SCO properties, such as policy awareness empowered by a ranged-based sensors network,
or passive autonomy for issuing alerts. This study, however, has shown that the panoramic and
interconnected smart properties of SCOs can not only objectively identify the dangerous situations
but also deal with them promptly.
21

Several innovations are offered by this study. Firstly, SCOs present a new way to integrate in a
single management system with the monitoring, identification, and visualization of dangerous
situations, as well as alerting and autonomous action-taking functions. Although its architecture is
multi-layered and its operation processes may seem complicated, the SCO-enabled OHS
management system can be encapsulated into one or more smart cores and executed instantly using
their computational power. The smart core has customizable functions, and it can be mounted to
and demounted from existing construction objects. Secondly, this approach aims not to alter
existing functionalities of construction objects, but to make them smarter with the introduction of
SCOs. This can be achieved with minimal interference with existing construction processes. There
are plenty of cases where researchers or consultants have introduced new, grand smart systems,
which have proven futile due to requirements placed upon construction personnel to cater these
systems (Woudhuysen and Abley 2004). The system proposed in this paper aligns with the
argument that a successful smart construction system is the one that causes the least interruption
to accepted processes (Niu et al. 2016). The automotive industry provides a parallel example with
its successful implementation of smart systems in vehicles.
This study has begun the work of introducing data mining, pattern recognition, machine learning,
and artificial intelligence (AI) to construction OHS management. By mining the large amounts of
relevant data collected, AI can be developed. Too often in construction, data is scattered across
systems (Cheng and Teizer 2013). Deviations among emerging technologies make data
consolidation difficult, limiting their potential to provide substantive information, which can
support smarter decision making. The SCO-enabled OHS management system integrates various
data/information islands in the same platform and makes good use of big data. These data collected
can be further analysed for worker behaviour patterns, which could in turn contribute to safety
training, and endeavours fostering a safety culture and climate.
It may be claimed that the smart system proposed here is too ambitious and impractical, especially
given that the construction industry has long been regarded a notorious “laggard” in technology
development and adoption (Liu et al. 2018). In developing an SCO-enabled smart system, the
intention is not to persuade construction industry personnel to relinquish the existing protections
22

offered by PPE and the safety climate they have cultivated. Rather, the system is intended to
provide an extra layer of protection where existing protections fail. Finding a balance between
traditional fragmented management and “full AI” that supersedes human beings is a delicate matter.
What this study proposes, however, is a “narrow AI” that equals or exceeds human intelligence
for certain tasks.
While this research provides an innovative, operable system to enhance OHS management by
focusing on dealing with the dangerous situations, readers are reminded that it does not aim to
introduce a technical solution per se. Rather, it aims to promote an ideological shift. An essential
purpose of this paper is to urge researchers and practitioners to go beyond the improvements
offered by the traditional first and second waves of construction OHS management to explore AI
as the third wave. Although the AI provided by SCOs is rudimentary, this should not prevent us
from devoting greater efforts to this promising area. Dating back to twenty years ago, the accidentavoiding car with the intelligent cruise control system is envisioned less realistically as a grand
development direction of AI in Reddy’s (1996) work. Despite continuing scepticism regarding the
development of AI, the introduction of autopilot cars to the market has proven that smart systems
are by no means mere fantasy.
The fact that the autonomy of the smart system could only be tested in a lab environment is not
considered by the authors to be a limitation of this study. On the contrary, it vividly reveals the
difficulties and resistance AI would encounter in the practical world of construction, and
encourages us to devise robust AI solutions as a means of convincing practitioners. Solutions to
technological hurdles such as scalability, endurance, and replacement of smart systems should be
investigated and tested rigorously, particularly in the case of systems designed to manage the safety
and health of workers. When in the future AI reaches technical maturity, bigger challenges may
lie ahead in navigating codes of practice, cultural norms, and ethical concerns. Now, however, is
the moment to explore AI to achieve smarter and safer construction.
7. Conclusions
This research offers an in-depth exploration of smart construction objects (SCOs) focusing on their
smart abilities in construction occupation health and safety (OHS) management. Deviating from
23

traditional research on OHS management using safety technologies or developing a safety culture,
this research argues for artificial intelligence (AI) in improving the stagnant OHS management in
construction. By augmenting existing construction resources with core smart properties including
awareness, autonomy, and communicativeness, SCOs represent an integrated means of monitoring,
visualizing, alerting, and action taking in the management of dangerous situations. Targeting the
operation of a tower crane, the SCO-enabled OHS management framework and system were
validated in a lab experiment. The results of this experiment demonstrate the feasibility of applying
the proposed system to on-site practice.
The research makes several practical and theoretical contributions. Firstly, by referring to the
example in this research, the SCO-enabled OHS management framework can be extended and
applied to other smart technology-enabled OHS management systems to develop management
strategies. The multilayer architecture of the system developed in this study provides clear
direction and sufficient detail for other researchers interested in replicating this work. Theoretically,
while acknowledging the merits of traditional PPE and human-based OHS management strategies,
this research seeks a united front on smart technology-enabled OHS management systems by
drawing attention to the deficiencies of traditional strategies, specifically in provision of proactive
monitoring and real-time alerts. Beyond the monitoring and alerting functions supported by
existing OHS management systems, this research argues for SCO autonomy as a new dimension
which can prevent dangerous situations from becoming fatal accidents in a timely manner. When
proposing the AI-based solution as the direction of the “third wave” of construction OHS
management, this study aims to emphasize the differences and potential values that could be
brought about by SCOs. This study not only provides a sound theoretical foundation for efforts to
proactively manage dangerous situations, but also concludes that future research efforts should be
devoted to the achievement of smarter construction, incorporating AI in particular, to reduce major
accidents. By exploring the AI offered by SCOs, there are opportunities and challenges in steering
the construction industry toward a smarter and safer future.
References
Abdelhamid, T. S., and Everett, J. G. (2000). Identifying root causes of construction
accidents. Journal of Construction Engineering and Management, 126(1), 52-60.
24

Cambraia, F. B., Saurin, T. A., and Formoso, C. T. (2010). Identification, analysis and
dissemination of information on near misses: A case study in the construction industry. Safety
Science, 48(1), 91-99.
Cheng, T., and Teizer, J. (2013). Real-time resource location data collection and visualization
technology for construction safety and activity monitoring applications. Automation in
Construction, 34, 3-15.
Chi, S. and Han, S. (2013). Analyses of systems theory for construction accident prevention with
specific

reference

to

OSHA

accident

reports. International

Journal

of

Project

Management, 31(7), 1027-1041.
Choudhry, R. M., and Fang, D. (2008). Why operatives engage in unsafe work behavior:
Investigating factors on construction sites. Safety Science, 46(4), 566-584.
DeJoy, D. M. (1990). Toward a comprehensive human factors model of workplace accident
causation. Professional Safety, 35(5), 11.
Fang, D., Jiang, Z., Zhang, M., and Wang, H. (2015). An experimental method to study the effect
of fatigue on construction workers’ safety performance. Safety Science, 73, 80-91.
Fang, Y., Cho, Y. K., and Chen, J. (2016). A framework for real-time pro-active safety assistance
for mobile crane lifting operations. Automation in Construction, 72, 367-379.
Fitton, D., Sundramoorthy, V., Kortuem, G., Brown, J., Efstratiou, C., Finney, J., and Davies, N.
(2008). Exploring the design of pay-per-use objects in the construction domain. In European
Conference on Smart Sensing and Context (pp. 192-205). Springer Berlin Heidelberg.
Flanagan, R., Jewell, C., Lu, W., and Pekericli, K. (2014). Auto-ID — Bridging the Physical and
the Digital on Construction Projects. Chartered Institute of Building.
Green, L., and Tominack, G. (2012). Real-time proactive safety in construction. Power, 156(1),
62-65.
Hadikusumo, B. H. W., and Rowlinson, S. (2002). Integration of virtually real construction model
and design-for-safety-process database. Automation in Construction, 11(5), 501-509.
Hahn, S. E., and Murphy, L. R. (2008). A short scale for measuring safety climate. Safety
Science, 46(7), 1047-1066.
Harris, M. (2015). Documents confirm Apple is building self-driving car. The Guardian, 14.
Heinrich, H. W. (1941). Industrial Accident Prevention. A Scientific Approach. (Second Edition).

25

Hinze, J., Hallowell, M., and Baud, K. (2013). Construction-safety best practices and relationships
to safety performance. Journal of Construction Engineering and Management, 139(10),
04013006.
Holt, A. S. J. (2008). Principles of construction safety. John Wiley and Sons.
International Labour Office (ILO), 2001. Guidelines on Occupational Safety and Health
Management Systems. MEOSH/2001/2(Rev.). International Labour Office, Geneva.
Irani, Z., and Kamal, M. M. (2014). Intelligent systems research in the construction
industry. Expert Systems with Applications, 41(4), 934-950.
Kessler, A. M. (2015). Elon Musk Says Self-Driving Tesla Cars Will Be in the US by Summer. The
New York Times, B1.
Kurzweil, R. (2005). The singularity is near: When humans transcend biology. Penguin.
Lu, W., Huang, G. Q., and Li, H. (2011). Scenarios for applying RFID technology in construction
project management. Automation in Construction, 20(2), 101-106.
Mohamed, S. (2003). Scorecard approach to benchmarking organizational safety culture in
construction. Journal of Construction Engineering and Management, 129(1), 80-88.
National Safety Council. (2015). Injury Facts®, 2015 Edition.
Neitzel, R. L., Seixas, N. S., and Ren, K. K. (2001). A review of crane safety in the construction
industry. Applied Occupational and Environmental Hygiene, 16(12), 1106-1117.
Niu, Y., Lu, W., Chen, K., Huang, G. G., and Anumba, C. (2015). Smart construction
objects. Journal of Computing in Civil Engineering, 30(4), 04015070.
Niu, Y., Lu, W., Liu, D., Chen, K., Anumba, C., and Huang, G. G. (2016). An SCO-Enabled
Logistics and Supply Chain–Management System in Construction. Journal of Construction
Engineering and Management, 04016103.
OSHA Directorate of Training and Education. (2011a). Construction Focus Four: Caught-In or Between Hazards, OHSA Training Institute.
OSHA Directorate of Training and Education. (2011b). Construction Focus Four: Electrocution
Hazards, OHSA Training Institute.
OSHA Directorate of Training and Education. (2011c). Construction Focus Four: Fall Hazards,
OHSA Training Institute.
OSHA Directorate of Training and Education. (2011d). Construction Focus Four: Struck-by
Hazards, OHSA Training Institute.
26

Pearce, D.W. (2003), Environment and business: socially responsible but privately profitable? In
J. Hirst (ed). The Challenge of Change: Fifty Years of Business Economics, 54-65, London:
Profile Books.
Petersen, D. (1984). Human-error reduction and safety management. STPM Press, New York.
Raheem, A. A., and Hinze, J. W. (2014). “Disparity between construction safety standards: A
global analysis.” Safety Science, 70, 276-287.
Raviv, G., and Shapira, A. (2018). Systematic approach to crane-related near-miss analysis in the
construction industry. International Journal of Construction Management, 18(4), 310-310.
Reason, J. (2000). Human error: models and management. BMJ, 320(7237), 768-770.
Reddy, R. (1996). The challenge of artificial intelligence. Computer, 29(10), 86-98.
Shapira, A., and Lyachin, B. (2009). Identification and analysis of factors affecting safety on
construction sites with tower cranes. Journal of Construction Engineering and
Management, 135(1), 24-33.
Shepherd, G.W., Kahlera, R.J., Cross, J., 2000. Crane fatalities – a taxonomic analysis. Safety
Science 36 (2), 83–93.
Simon, H. A. (1976). Administrative behavior, 3rd Edition. Free Press, New York.
Somavia, J. (2005). Facts on safety at work. International Labor Office (ILO), Technical Report.
Sterman, J. D. (1989). Modeling managerial behavior: Misperceptions of feedback in a dynamic
decision making experiment. Management Science, 35(3), 321-339.
Teizer, J., Allread, B. S., Fullerton, C. E., and Hinze, J. (2010). Autonomous pro-active real-time
construction worker and equipment operator proximity safety alert system. Automation in
Construction, 19(5), 630-640.
Tsang, Y. T., Fung, I. W., Tam, V. W., Sing, C. P., and Lu, C. T. (2017). Development of an
accident modelling in the Hong Kong construction industry. International Journal of
Construction Management, 17(2), 124-131.
Woudhuysen, J., and Abley, I. (2004). Why is construction so backward? Wiley Academy.
Wu, W., Gibb, A. G., and Li, Q. (2010). Accident precursors and near misses on construction sites:
An investigative tool to derive information from accident databases. Safety science, 48(7),
845-858.

27

Yang, H., Chew, D. A., Wu, W., Zhou, Z., and Li, Q. (2012). Design and implementation of an
identification system in construction site safety for proactive accident prevention. Accident
Analysis and Prevention, 48, 193-203.
Yuan, X., Anumba, C. J., and Parfitt, M. K. (2016). Cyber-physical systems for temporary
structure monitoring. Automation in Construction, 66, 1-14.

28