Contents lists available at ScienceDirect

Safety Science

journal homepage: www.elsevier.com/locate/safety

Building information modeling in combination with real time locationsystems and sensors for safety performance enhancement

Shuang Donga,b,⁎, Heng Lib, Qin Yinb

a East China University of Science and Technology, Chinab Hong Kong Polytechnic University, China

A R T I C L E I N F O

Keywords:PPEMisuseIdentificationAssessmentAutomation

A B S T R A C T

The misuse (including nonuse) of personal protective equipment (PPE) directly catalyzes the changes from in-cidents to critical accidents and diseases. However, to date, the main control over PPE use is through manualvisual inspections, which is time-consuming and more often than not biased. Moreover, the current applicationsof technologies such as RFID have not guaranteed the proper use of PPE on site.

This study introduces a novel approach towards automated remote monitoring and assessing how the PPEsare worn through integrating pressure sensors and positioning technologies. In order to realize such automatedcontrol, potential technologies are reviewed and a wireless network architecture composed of end nodes, re-peaters/checkpoints and coordinators is considered. The real time location system (RTLS) and virtual con-struction are developed for worker’s location tracking to decide whether the worker should wear helmet and givea warning, while the silicone single-point button is designed to show whether the PPE is used properly for furtherbehavior assessment. The process of data synchronization and fusion of location coordinates and pressure data isdescribed in detail and the system is tested in an open area experiment to prove its feasibility.

1. Introduction

Construction has been always notorious as one of the most dan-gerous industries due to its unique nature (Jannadi and Bu-Khamsin,2002), such as outdoor operations, work-at heights, complicated on-siteplants and equipment operation coupled with workers attitudes andbehaviors towards safety (Choudhry and Fang, 2008). In Hong Kong,construction industry is one of pillar industries as it employs millions ofsite workers. But this industry has experienced a shortage of laborsthese years and companies have to reach hands to people elder or withlittle relevant working experience. In this situation, how to ensure thesafety and health of workforces is becoming a more challenging andcomplex task. The misuse of personal protective equipment (PPE) likesafety helmet and respirator contributes directly to the severity of ac-cidents and is closely connected with many occupational diseases likehead injuries, hearing loss, dermatological disease and pneumoconiosis.As a result, it is recognized as one of the twenty non-negotiable unsafebehaviors on work site.

To reduce the misuse behaviors and improve safety performance,the current ways have mainly emphasized three aspects: (1) improvingPPEs in technological aspects; (2) enhancing on-site safety managementprocedures and protective measures; and (3) providing more PPE use

trainings. These technical and organizational actions are useful buthave reached a world class bottleneck: although expenditures in safetymanagement manpower, protective measures and safety trainings havebeen increasing annually, the accident rate has decreased little after theimplementation of compulsory construction industry safety trainingcertificate (education) and Pay for Safety Scheme (punishment). Theirineffectiveness is attributed to: (1) they rely on well trained and highlyexperienced safety observers, (2) subjective observations or surveys areneeded that result in omissions or biases, (3) they do not allow PPEtraceability or real-time monitoring for in-time feedback and (4) out-come-based group level assessment conceals personal performanceunduly.

One of the solutions to this problem are positioning and sensortechnologies because they have the potential to foster better safety andproductivity by tracking construction resources (labor, equipment,materials, etc.) anywhere and anytime (Cheng et al., 2011; Ergen et al.,2007; Torrent and Caldas, 2009). Many related technologies like RFIDand cyber physical system have been employed for PPEs use mon-itoring, but these methods can only detect whether the workers carrythe PPEs with them without any judgment on whether they use thePPEs correctly in the place needed. Therefore, the goal of this study is toinvestigate the better prospect of automatically identifying and

http://dx.doi.org/10.1016/j.ssci.2017.10.011Received 18 April 2017; Received in revised form 26 September 2017; Accepted 17 October 2017

⁎ Corresponding author.E-mail address: rosine.dong@connect.polyu.hk (S. Dong).

Safety Science 102 (2018) 226–237

0925-7535/ © 2017 Elsevier Ltd. All rights reserved.

MARK

assessing PPE misuse behavior on personal level, and providing feed-back sufficiently and quickly to modify the unsafe behavior. In doingthis, the idea is developed in the context of helmet misuse on a work-site. The supporting system has the primary functions of trackingworkers and danger sources, sending our warning signals and assessinghelmet misuse behavior. The warnings and worker responses are re-corded for use in analyzing individual safety performance and pro-viding timely feedback.

The paper is organized as follows. First a detailed background isprovided, describing PPEs use problems on site, the existing methodsfor behavior improvement and insufficiencies. Then potential technol-ogies for location tracking and sensor are provided. A conceptualmethod of positioning and sensor technologies-enhanced PPE usemanagement is then developed, followed by details of the supportingsystem. The application of the system is then demonstrated and verifiedon a live construction site in Hong Kong. Final comments are providedconcerning further research work needed and prospects for extension toother situations involving safety considerations.

2. Background

2.1. Current PPE misuse and management on site

According to Occupational Safety and Health Council (OSHC), PPEmeans any protective equipment that protects users from being exposedto a potentially hazardous environment. PPE can be divided by theprotection parts or functions of human body such as head protection,hand protection, body protection and respiratory protection. Some fa-mous PPE are safety helmets, safety shoes, safety belts or ear muffs. PPEis highly effective to avoid injury if wore and fitted properly whenworkers come into contact with a potentially hazardous situation(Dingus et al., 1993; Long et al., 2013) such as toxic chemicals, electricshock, slippery floor or falling objects. Lack of PPE use is repeatedlynoted as a contributor to occupational injuries (Contreras andBuchanan, 2012). For example, hard hats are designed to protectagainst concussions and traumatic brain injuries caused by strikes to thehead. The nature of construction calls for proper use of hard hat on thejob. In the United States, the construction industry has the highestnumber of workplace head injuries due to the improper/use non-use ofhard hats. Beyond safety problems confusing construction for decades,some occupational chronic diseases, such as hearing loss, dermatolo-gical disease and pneumoconiosis, caused by misuse of PPEs in badenvironment ruin the quality of life in long run. Meanwhile, thecommon phrase “dress for success” applies beyond white-collarworkers, which means clear positive connections have been found be-tween satisfaction of PPE and work clothing, self-efficacy, and overallsatisfaction of trades work (Wagner et al., 2013).

PPE misuse is often neglected because current assessment is mainlyfocused on visible outcomes such as critical injuries and accidents, andit is hard to identify hazardous behaviors in time (SWA, 2013). Un-doubtedly, using PPE is a factor which would be positively correlated tosafety performance on construction sites and became one of the mostimportant factors affecting safety performance (Sawacha et al., 1999).PPE misuse records are mainly kept by self-reporting, which is inhibitedby a blame culture for error, time-consuming paperwork, and lack offeedback on how the information reported has been used (Van DerSchaaf and Kanse, 2004). To solve these problems, current activitiesmainly involve modifying PPE use behavior through safety regulationsand training (Kaskutas et al., 2013), and improving safety attitudethrough better organizational safety culture (Fung et al., 2012). Thesemethods are useful but do have disadvantageous such as:

(1) being unable to remedy the limitations of human vision and abilityto detect all surrounding danger sources;

(2) largely relying on wandering inspection and lagged (outcome)measurement, which fails to provide feedback to change unsafe

behaviors in time. The main reasons resulting in failed protectionare located in: disability of PPEs to fulfill a nominal safety function(Mäkinen and Mustonen, 2003), or unexpected behaviors in ha-zardous situation (Choudhry and Fang, 2008). As summarized inFig. 1, the PPE-oriented researches are devoted to technical im-provement for better protective and reminding function.

Many international safety and health organizations like OSHA(2003) made a detailed and comprehensive list of protective functionsfor multiple PPEs in construction. For example, the filter lenses forprotection against radiant energy should be selected and examined interms of plate thickness and minimum protective shade according todifferent electrode size and arc current. What’s more, Mäkinen andMustonen (2003) analyzed 25 electric arc accidents that occurred inFinland in 1996–1999 to find out the different technical requirementsof protective clothing. And there are 12 factors that affect the accept-ability of head protection at work including weather protection,thermal properties, tactile properties, absorptivity/permeability andmass distribution et al. (Hsu et al., 2000). To fulfill these requirements,many methods are implemented, such as helmet made by compositematerials with thermoplastic matrices and a reinforcement of naturalfibers (Murali and Nagarani, 2013). Except for the detailed technicalimprovement to ensure better protective function, many other factorslike the way of warning and marking also play very important roles inpromoting PPE use. As proven by Dingus et al. (1993), a warning whichcontains information pertaining to the specific consequences of usingthe product, and a warning situated in such a way that the consumermust interact with the label (increasing label noticeability) can improvesafety performance effectively.

With regard to another reason leading to protection failure, themisuse of PPEs is identified as a serious and unforgivable unsafe be-havior on worksite. According to the decision making process, thefactors influencing use of PPE and consists of three primary branches:perceptions of hazards and risks, “barriers” to PPE use, and enforce-ment and reinforcement (Lombardi et al., 2009). Many regulations areestablished to minimize the cost of compliance like the contractors areresponsible to provide qualified PPEs for workers. To improve thesafety perception and changing attitude, training is always the popularway both in academy and practice (Mäkinen and Mustonen, 2003).During the implementation phase, the behaviors are inspected and as-sessed for feedback to correct the unsafe behavior finally. To realizeefficient behavior inspection, many technologies like Mobile passiveRadio Frequency Identification (RFID) was applied to perform auto-matic site access, time recording, and completeness control (Kelm et al.,2013). And there was cyber physical system set up for real-time PPEsmonitoring by keeping the PPEs in close range (Barro-Torres et al.,2012). But we identify three weaknesses of these similar methods andtechnologies: (1) these methods cannot ensure the proper use of PPEs,(2) do not take conditions into consideration automatically where dif-ferent kinds of PPEs are needed, and (3), the wearable devices appliedare relatively heavy and difficult to apply in real construction sites.

PPE-Oriented Behavior-Oriented

• Protective function• Warning function

• Marking function

• Regulations• Training

• Inspection

Automated PPEs use

Fig. 1. The current methods to improve PPE use.

S. Dong et al. Safety Science 102 (2018) 226–237

227

2.2. Difficulties in PPEs use management

Although many factors may affect the outcome of traditionalmethods, the difficulty is largely attributed to inaccurate assessmentand inefficient inspection. Occupational health and safety (OHS) hastraditionally been measured by outcomes such as accidents, injuries,illnesses and diseases (Arezes and Miguel, 2003). This lagged mea-surement still prevails in many industries since it is relatively easy tocollect data, easily understood, objective and valid (Lingard et al.,2011). However, these “after the fact” indicators limit the opportunityfor prevention and correction in time. As a result, these retrospectiveindicators are not an accurate representation of construction workersafety.

As human behavior is the key factor leading to accidents (HSE,2002), many researches treat safety issues in a more proactive way(Blewett, 1994; Council, 2005; Wales, 2001). These personal perfor-mance indicators are derived from hazardous behavior during worktime and involve safety compliance as well as supporting participation.Safety compliance indicators refer to following the safety regulationsand plans, which constitute a key part of the traditional methodsmethodology.

Safety performance in traditional methods is calculated by thepercentage of safe behavior of all observed behavior (Choudhry, 2012).This percentage can provide a good reflection of safety at the group orproject level, but conceals personal/individual safety. The lack of per-sonal level assessment compromises traditional methods ability to im-prove everyone’s safety behavior. Furthermore, the observed behavioris judged by “all or nothing” normality (Wiegand, 2007) and cannotreveal the whole process involved in the behavior.

Another issue concerns the inefficiency of behavior inspection insafety management. With current practice, trained observers or safetysupervisors are responsible for safety behavior inspection based onsafety plans and operation regulations (Zhang and Fang, 2013). Thistime-consuming activity largely depends on the supervisor’s safetyknowledge and experience, which often results in omissions or biases.

Except for lacking of systematic personal safety performance as-sessment, a further issue concerns the inefficient behavior inspection insafety management. In present practice, trained observers or safetysupervisors are responsible for safety behavior inspection based onsafety plans and operation regulations (Zhang and Fang, 2013). Thistime consuming activity is largely depended on supervisor’s safetyknowledge and experience which often results in omission or bias.Consequently, visualization and tracking techniques emerge as pro-mising remedies in safety monitoring (Han and Lee, 2013) rather thansafety assessment yet.

In general, therefore, mass application of existing traditionalmethods is not possible for construction work as it mainly applies post-mortem analysis at the group or project level due to the lack of a meansof quickly and objectively collecting real-life behavioral data from sites.

2.3. Potential technologies for location tracking and sensing

Following their successful application in manufacturing (Breweret al., 1999) and traffic management (Wang et al., 2004), a wide rangeof positioning technologies have become available with the potentialfor improving on-site management (Carbonari et al., 2011). Of these areRadio Frequency Identification Devices (RFID), Laser Detection andRanging (LADAR), Vision Cameras (VC), Audio Technology, RadioDetection and Ranging (RADAR), Global Positioning Systems (GPS),Ultra Wide Band (UWB), Inertial Measurement Unit (IMU) and infraredheat and magnetic sensors (Teizer et al., 2008). Many popular tech-nologies are able to solve problems in material flows, equipment useand movement. For example, Grau et al. (2009) compare the automatedidentification and localization of engineered components with tradi-tional manual methods on industrial sites and demonstrate significantproductivity gains. Yang et al. (2012) also illustrate the use of

surveillance cameras and discrete-state inference algorithms for asses-sing tower crane activities during the course of a workday.

In terms of safety management, the identification of accident pre-cursors, training and inspection are three main aspects involved inimproving safety behavior. For instance, Teizer et al. (2008) record themovement of workers by UWB and use a combination of convex hulland shortest path algorithms to identify obstacles and dangers ac-cording to the frequency with which workers cross their path. Theythen use emerging radio frequency (RF) remote sensor and actuatingtechnology to improve construction safety by pro-active real-timewarning workers-on-foot and plant operatives when they become tooclose to each other (Teizer et al., 2010). Carbonari et al. (2011) havealso established a new advanced hardware/software system to performreal time tracking of workers’ routes and prevent workers being inhazardous situations by a virtual fencing tool. Jebelli et al. (2016) haveapplied IMU to measure workers’ fall risk in stationary postures. Foreducation and training, Teizer et al. (2013) integrate real-time locationtracking and three-dimensional immersive data visualization technol-ogies to train and assess the operations of steel-erection tasks.

However, the success of these methods is often comprised by pooraccuracy, high cost, complex deployment and lacking validation by onsite case studies. For example, the commonly used wireless devices forobstacle avoidance require tags (small hardware devices designed to bemounted on helmets and moving objects) on every individual resourceon a site (human, material, and equipment), which is vulnerable tounforeseen events involving mistakenly untagged resources (Teizeret al., 2007). Failures also derive from limited signal strength throughobstructions, the unavailability of GPS satellites or losing contact with abase station to determine precise locations, the high cost of tags, etc.(Teizer et al., 2010). As a more accurate positioning technology, IMUhas been approved to be an effective tool for indoor location trackingand good at relatively short distance detection (Ibrahim and Moselhi,2016) rather than long distance movement. At the same time, the dy-namic and evolving environments of construction projects require fur-ther amendments to the technologies trialed in the laboratory(Carbonari et al., 2011). Moreover, although monitoring materials islargely considered to be adequate, continuously monitoring labor is lessso (Navon and Sacks, 2007) and more effective approaches are urgentlyneeded (Teizer et al., 2008).

3. Methodology

3.1. PPE misuse identification and assessment framework

Fig. 2 contains the general methodology of identifying and assessingPPE misuse behavior. All these PPE misuse behaviors are reminded bywarnings and assessed according to the later responses represented.Only if the worker enters the danger zone without PPE, the alarm willbe given for reminding. Then if the worker still doesn’t wear the PPEafter a certain response time, this behavior will be recognized as amisuse behavior.

At the beginning of the decision cycle, danger zone where need towear PPE is identified based on a full discussion with experiencedproject managers and safety officers such that:

≜ = ∈ ⋂ ⩽ = = …=F x x x R N X n X x x x i m{( , , ) | }; ( , , ), 1,2, ,im

i i1 2 33

1 1 2 3

(1)

where (x1, x2, x3) are the coordinates of points in the danger zone, andthe danger zones can be designed as a space, a plane, a line or a dot.Workers are informed through training of the danger zones and thistraining can be used to improve safety by directing attention to PPEmisuse.

In the inspection and assessment phase, traditional manual ob-servations and subjective judgments are substituted by automatedwarning and response assessment:

S. Dong et al. Safety Science 102 (2018) 226–237

228

= ⎧⎨⎩

∈g Y t

Y Felse( , )

1;0;

(2)

where the worker’s real-time location Y is measured by positioning andsending technologies and recorded in a database. If the worker enters a

danger zone ( ∈Y F), a reminding warning rings out as in-time feedbackto workers. After this warning, the unqualified behavior is identified by:

+ = ⎧⎨⎩

∈ =g Z t t

Z F P Zelse( , Δ )

1; , ( ) 00;

(3)

where tΔ denotes the response time after the warning. In this phase,there are two kinds of corresponding activities: (a) if the worker is stilllocated in danger zone ( ∈Z F ) without PPE ( =P Z( ) 0), this behavior isrecognized as a misuse response; but (b) if the worker leaves the dangerzone ( ∉Z F ) or takes on the PPE ( =P Z( ) 1), this response is regardedas a safe behavior.

The warnings, time, coordinates and pressure data are collected andrecorded for further statistical analysis including PPE misuse at dif-ferent times, by different workers, and in different danger zones.Formal feedback and interventions regarding these outcomes are pro-vided to workers to modify their intrusion behaviors, and are alsoprovided to safety managers as a reference to goal setting in the nextround. The safety performance measured in different rounds can becompared both at the individual and group level to judge whether thePPE misuse behaviors have been effectively modified.

3.2. Data synchronization

Since multiple technologies are employed to collect heterogeneousdata sources which have different levels of detail, data collection rates,data representations, and time reference systems, data synchronizationand fusion are needed for further data analysis. In this study, the datafusion method is adopted and revised from the research of (Cheng et al.,2012), which effectively synchronized the physiological status and lo-cation data.

For three kinds of data shown in Fig. 3, video time is regarded as theground truth, and it is assumed that the propagation of time differenceconsists of two parts: initial time shift and continuous time shift:

= + = −t t αt t t tΔ Δ ; Δi sensor i video i sensor i0 , , (1)

where tΔ i means time lag between sensor and video when a specificevent i occurs. An event refers to taking on/off PPE or moving betweenfree and danger zones. Term tvideo i, is the video time when event i (e.g.,enter a danger zone without PPE) is observed. While term tvideo i, is the

Location data

Alert

PPE pressure = 0?Yes

Identified as misuse behavior

Identified as safe behavior

After response time

In danger zone?

NoYes

Still in danger zone?No

YesNo

Fig. 2. The process of PPE use behavior warning and assessment.

Fig. 3. Time lines of multiple data.

S. Dong et al. Safety Science 102 (2018) 226–237

229

event i time recorded by sensor’s clock. When i=0, ti refers to theinitial status of location system and sensor when they start recordingdata, then tΔ 0 represents the initial time shift between sensor and videorecordings. In addition, αtsensor refers to the built-in drifting time ofsensor, where α is time adjustment factor correcting the second fromsensor to be equal to the video. A positive α means the sensor clock runsslower than video clock while the negative one indicates the oppositesituation. To calculate the parameters α and tΔ 0. The linear time lagpropagation algorithm is applied on a set of random events, which isshown as:

=∑ − − −

∑ −= −α

t t t t tt t

t t αt( )[Δ ( )]

( );sensor i sensor i video sensor

sensor i sensori sensor

,

,2 0

(2)

Once the time lag propagation parameters are determined, the timeof sensor and video can be synchronized:

= + + = −t t α t ε t t^ (1 ) ; ^j sensor j video j j0 , , (3)

where t̂j predicts time on the corresponding video timeline. Event j isrecorded from the sensor at sensor time tsensor j, and ε means predictederror. When the sample size of testing event j is large enough, ε followsGaussian distribution. And the likelihood function is calculated by theBayesian approach:

= ∗ ∗∗

A L S P A L P A S P A L SP A L P A S

( | ) ( | ) ( | ) ( | )( | ) ( | )i new new

i new i new i old old

i old i oldL

where Ai is the observation status i; Lnew and Snew are the new data from

Pressure Sensing

Web Server

Real-time Location Engine

Real-time Location Network

Application Server (Virtual Construction Engine)

Database

Data Fusion

User Client

Fig. 4. General architecture of PUMS.

Anchor

End node

Repeater

Checkpoint

Router

Coordinator

Fig. 5. System deployment.

S. Dong et al. Safety Science 102 (2018) 226–237

230

location system and sensor respectively, while Lold and Sold are the olddata. P A L S( | )i old old is the estimation in previous data synchronizationmodel.

3.3. System architecture and applied technologies

3.3.1. General architectureThis multi-user on-line supporting system, which is named as PPEs

Use Management System (PUMS), consists of three parts: Real TimeLocation System, Virtual Construction System and PPEs Sensing System.Real Time Location System applies tags and reference anchors in de-tecting and sending the ranging information through wireless signals.

And Virtual Construction System is responsible for measuring the re-lative 3D positions of workers and their surrounding danger sources/zones, and recording real-time 3D movements of workers and movingequipment. Meanwhile, PPEs Sensing System is developed for sensing,collecting, transferring and saving the PPE use status. If deemed ne-cessary by PUMS, warnings will be sent to alert workers through tagsinstalled on their helmets. In order to access the latest virtual con-struction models conveniently and timely, PUMS adopts a typical three-tier web-based application structure illustrated in Fig. 4 composed ofpresentation layer, business layer and data layer.

The PUMS comes true by help of four main parts shown in Fig. 5:End nodes, repeater/checkpoint and coordinator. End nodes are thecritical part of the system which are worn by workers and responsiblefor gathering information about PPEs and location. These devices arecomposed by a central unit microcontroller to regulate the behavior ofthe device, a pressure button sensor for pressure information collectionand a radio module for location detecting and transmitting information.After the data is collected, the repeater or checkpoint will help the endnodes connect with coordinator wirelessly. At last, the coordinator willcollect, store and synchronize the data from pressure sensor and loca-tion. It is also responsible for node configuration and activating alarm.

3.3.2. Location system and virtual constructionReal time location network is the most important part of Real Time

Fig. 6. Location system deployment.

Fig. 7. Silicone single-point sensor.

Silicone single-point buttons

Cache Read

On/Off

PPEs sensing systemCheckpoint established at site gates

Database

Web GUI/mobile apps

PPEsmonitoring

services

logicBusiness

Fig. 8. Sensing system architecture.

S. Dong et al. Safety Science 102 (2018) 226–237

231

Location System, for ranging and real time location engine for re-cording and calculating. It is constructed by tags which are smallhardware devices designed to be mounted on helmets and moving ob-jects, and anchors to be fixed at some static positions as referencepoints. In order to be applied in construction industry, the trackingtechnology should meet the criteria in terms of cost and maintenance,device form factor, scalability, data update rate, and social impact(Cheng et al., 2011). As a result, many positioning technologies areinvestigated and Chirp Spread Spectrum (CSS) technology is employedfor ranging, which estimates physical distance between two devices byTime of Flight (TOF) of radio frequency signals. CSS is a spread spec-trum technique defined in the standard IEEE %2.15.4a (Cho and Kim,2010) and uses wideband linear frequency modulated chirp pulses toencode information which is relatively less time-consuming, robustagainst disturbances, against multipath fading, low power consumptionand easy to implement in silicon.

To enable site managers to monitor intrusion behavior compre-hensively and timely, virtual construction technology is integrated withRTLS as the location-based virtual construction system which allowsimmediate synchronization between virtual building models and rea-listic construction situation. This location-based virtual constructioncomes true with the help of Location Engine, SmartFoxServer and Unityrevealed in Fig. 6. The location engine is named as PSMS Site andprogrammed to calculate tags’ positions and then send the positioninformation to application server. It also takes charge of relayingwarning signals to Location Network as a sound or vibration trigger.The warning to danger is generated by comparing tags’ coordinateswith those in dangerous zones marked in 3D model, and this compar-ison process is programed according to Java.awt.geom and Polygon2Dalgorithms. The dangerous zones are marked on virtual model and thedangerous coordinates are called by application server from database.All the programming are developed on SmartFoxServer which is amassive multiplayer game server providing a comprehensive platformfor rapidly developing multi-user applications and games with main-stream programming technologies. To be a multi-user system, a ServerObject Extension is also developed in application server to drive andsynchronize all the User Clients in terms of the real construction si-tuations. Last but not least, Unity is used to build the User Client forvisualizing construction progresses, defining static and dynamic dan-gers, as an integrated authoring tool for creating 3D video games orother interactive content such as architectural visualizations or real-time 3D animations.

The connection and information transformation between the net-work and location engine on computer come true by the application ofthe NanoLOC TRX transceiver which is based on CSS and the onlywireless devices using CSS with real time location and RFID abilities.

3.3.3. Sensing systemBased on price/responsiveness ratios and their capacity to resist

harsh conditions of construction, silicone single-point sensor is initiallyselected to detect if PPE is being worn by the site workers. Siliconesingle-point sensors, as shown in Fig. 7, are widely used in the pro-duction of keypads. Silicone rubber with raised dome protrusions gives

a controlled collapse when pressure is applied. When a conductivecarbon loaded silicone disc moulded under the domes bridges theprinted open circuit design on the printed circuit board underneath,switching occurs (Starke, 2010). Silicone-point single button designwith fully sealed, waterproof, dustproof, anti-oil, and anti-acid attri-butes, can be used in any harsh environment, and can be sterilized andcleaned by water. Silicone single-point buttons are adopted to detect ifPPEs are on workers by sensing the pressure.

PPEs Sensing System is developed for integrating sensor technolo-gies and wireless communication, as indicated in Fig. 8. Sensors auto-matically track real-time behavioral data on whether real-time beha-vioral data indicating whether workers are wearing the required PPEs.Through wireless communication devices such as Bluetooth technology,these data are recorded in the database for analyzing behavioral pat-terns, as well as serving as the track records of safety performances ofworkers. These data can be used to reinforce safety training, as well as abasis for developing a bonus/penalty scheme to correct the unsafe be-havior of workers.

4. Experiment

In order to demonstrate the feasibility and validity of this proposedautomated PPE use management method, it is necessary to test this on-line supporting system in an on-site environment. Since PPE misusebehaviors involve multiple scenarios and complicated operations, thisstudy selects safety helmet in the primary experiment because the headis the most critical area of a human body and severe trauma to the headcan lead to death or long-term disability. Helmets not only prevent theskull from being perforated but also dampen the force of the impactobject transmitted to the wearer (Long et al., 2013). It can be effectivein reducing both head accelerations and compressive neck forces forlarge construction objects in vertical impacts (Suderman et al., 2014).

Through this experiment, three main aspects will be tested, whichare (a) whether the whole automated methodology can be realized, (b)how accurate the location tracking and warning function is, and (c)how accurate the PPE use detection function is.

4.1. Experiment setting

The experiment was conducted at a residential project in Shenzhen,China. This open test site is in the land leveling process occupied bymany heavy machines and moving objectives, which is incident-prone.At first, the research staffs trained the managers and workers on thefunctions of the system and how to properly use it in practice. In thisexperiment, two senior safety managers and project manager partici-pated and were responsible for identifying the danger zones and pro-viding the workers’ information. Under the guide of two research staffs,the senior safety managers and project manager input the workers in-formation into the system and matched their information with the tagnumber separately. Two workers of cement subcontractor, named asWorker A and Worker B, shown in figure, were selected as the trackingobjectives during a normal work day (from 9 AM to 5 PM) The devicesneeded in this experiment included two tags, six anchors, one router,

Fig. 9. The endnote devices and tag carriers.

S. Dong et al. Safety Science 102 (2018) 226–237

232

two pressure sensor-mounted helmets and one receiver, which is alsoshown in Fig. 9.

To perform this comparative experiment, two research staffs wereresponsible for supervising workers’ behavior as the reference data.

After the system was in use, they adjusted their timers to the time of thesystem, and then they would observe those two workers. They markeddown the time when they entered and got out the danger zones, thedanger zone number, whether the warning was automatically given,

Fig. 10. Danger zones identification and experiment de-ployment.

S. Dong et al. Safety Science 102 (2018) 226–237

233

and the time when the workers took off and put on their helmets.According to The US Department of Labor’s Occupational Safety and

Health Administration (OSHA) requires employees to wear head pro-tection if: objects might fall from above and strike them on the head;they might bump their heads against fixed objects; or, there is a pos-sibility of accidental head contact with electrical hazards. In this case,two danger zones were identified: Excavation Zone 1 (EZ1) andExcavation Zone 2 (EZ 2), where two heavy excavators were workingand the safety helmet was required strictly. The parameters of thedanger zones were set up on the virtual model of this project inFig. 10(a) and (b), including the danger type and the shape, radius andlocation of the danger zone. And then the anchors were added on thevirtual model with their coordinates as location references likeFig. 10(c). For real time tracking, tags carried by the two workers wereutilized to track the location of workers. The tags were then matchedwith the personal information of the workers, such as work type andname in Fig. 10(d). Through careful calculation and prolonged discus-sion among those involved, the response time to the warning signal wasset as 3 s.

4.2. Data analysis

After the hardware was deployed on experiment site and parameterswere set up in system, the real time location of the work site was cal-culated and synchronized on the virtual map as shown in Fig. 1(a),where the virtual object in circle indicated the movement of Worker A.Meanwhile, the status of helmet use was synchronized and recordedautomatically, which is illustrated as Fig. 11(b). The synchronizedmovement, the real-time coordinates of both the tag carriers, andhelmet use status were recorded in the database as video, (X, Y, Z) andon/off time respectively. Since the signals can be distorted by occa-sional outliers, a Robust Kalman Filter was applied to reject outliermeasurements shown Fig. 12. This step proves the methodologyadopted in this system can be realized.

As to the two accuracy tests, compared with the reference data

collected by the two research staffs, there was none unmatched recordsin the system. The warning was given automatically when the workersentered the danger zones and the danger zone number was correct. Andevery time when the workers put on or take off the helmets, the be-havior was recorded in the system as shown in Fig. 11(b), and the re-corded time were the same with the reference data.

Since there had been no similar cases previously, two representativescenarios were chosen as examples for the response analysis. As in-dicated by the results shown in Fig. 13, the first misuse warning wastriggered by being in danger zone without helmet and the carrier ig-nored the danger warning. This was identified by the system and re-corded as unsafe behavior. In another case, the carrier put on thehelmet within 3 s of the warning and this it was therefore not recordedas a misuse behavior.

In total, there were 2885 points in danger zones, 9 warnings and 12helmet on/off events during the trial time and were recorded in data-base for further personal safety performance assessment. Specifically,Worker A entered danger zones 4 times totally, twice in EZ 1 and othertwice in EZ 2. Also Worker B entered danger zones for 5 times totally,once in EZ 1 and 4 times in EZ 2. The over many entrances in EZ 2 forWorker B were explained by the fact that Worker B was mainly incharge of the excavator in EZ 2.

As to the durations in danger zones for different workers, Fig. 14illustrated that Worker A stayed in EZ 1 for 353 and 334 s respectively,while in EZ 2 for 80 and 449 s each time. However, Worker B stayed inEN 1 for 274 s, but four times in EZ 2 last 871, 42, 501 and 45 s re-spectively. As shown in Fig. 14, Worker B stayed much longer in dangerzones than Worker A, which may indicate that B’ job was more dan-gerous than A. In terms of different danger zones, EZ 2 was associatedwith more entrances and longer durations than EZ 1, because the workin EZ 1 just started and needed more coordination and help fromWorker A and B.

In this experiment, danger zone entrance dose not equal to safetyviolation because the two participants were needed to work in thedanger zones. Only if they did not use their safety helmets in danger

Fig. 10. (continued)

S. Dong et al. Safety Science 102 (2018) 226–237

234

zones, they were recognized as safety rule breakers. There were in sum14 safety helmet on/off records, which included 6 times for Worker Aand 8 times for Worker B. After the fusion of location tracking data andsafety helmet on/off data, the safety helmet usage status are shown inTable 1 below. Only 9 safety misuse records were related to dangerzone entrances. Since all the durations in danger zones were longerthan 3 s, then if the helmet status at 3 s after entrance was off, the

Fig. 11. The synchronization and visualization of locationsand pressures.

Fig. 12. The Robust Kalman filter example.

(a)

(b)

Warning +

Response time

Warning +

Response time

Fig. 13. Helmet misuse behavior assessment.

S. Dong et al. Safety Science 102 (2018) 226–237

235

violation was positive. As shown in the list, Worker A entered in dangerzones without a helmet for 3 times and then took on twice afterwarnings. Worker B had 2 safety violations, one for ignored thewarning, and the other for took off the helmet in danger zone.

5. Conclusions

Traditional PPEs use management methods have failed to be widelyeffective because it is highly dependent on a manual and experiencedinspection process, and lacks accurate personal assessment and timelyfeedback. This paper solves this problem by providing an effective ap-proach to automatically identifying PPEs misuse behaviors with in-tegrating positioning technology and pressure sensor, and assesses thepersonal safety performance of workers according their response todanger warnings. This involved the development of a supporting multi-user platform to obtain the real-time position of workers in relation tovirtual hazardous zones. An on-site experiment study was conductedthat verified its ability to identify PPEs misuse behavior in specificcondition, issue timely warnings and capture worker responses. Thewarning and response data were then analyzed to assess individualsafety performance and locations over time for effective safety behaviorimprovement. One major limitation of this study is that only a fewconstruction sites are available for the experiment. It is unfortunate thatwe can only obtain such an amount of data for analysis. The mainreason leading to the practical problem is that most workers and

Fig. 14. Duration analysis results.

Table 1Safety helmet usage status list.

Worker Danger zone Helmet status whenwarning

Helmet status in3 s

Violation

A EZ 2 Off Off YesA EZ 1 Off On NoA EZ 1 On On NoA EZ 2 Off On NoB EZ 2 On On NoB EZ 1 Off On NoB EZ 2 On Off YesB EZ 2 Off On NoB EZ 2 Off Off Yes

S. Dong et al. Safety Science 102 (2018) 226–237

236

managers are worried that they need extra time to learn and use thenew system, which may delay the project schedule. We are aware thatthe small experiment data set may possibly influence the test results.For example, this experiment was conducted on an open site at a sunnyday, we are not sure whether the work environment or weather situa-tion would reduce the system accuracy. Then, we intend to continue ourexperiment on more construction sites and implement this study withmore data for analysis in the future. Future research will investigatevarious personal factors leading to PPE misuse.

Acknowledgements

The work was supported by the Innovation and TechnologyCommission (ITC), Hong Kong Special Administrative Region, underthe project “Location-based Technologies for Asset Tracking and RiskManagement” (ITP/036/12LP), a General Research Grant titled“Proactively Monitoring Construction Progress by Integrating 3D Laser-scanning and BIM” from the Research Grants Council of Hong Kong(Reference No. PolyU 152093/14E), and the National Natural ScienceFoundation of China (Grant No. 71390523). The authors gratefullyacknowledge the Department of Building and Real Estate at Hong KongPolytechnic University and the Research Institute of ComplexEngineering &Management at Tongji University for providing supportsto conduct this research.

References

Arezes, P.M., Miguel, A.S., 2003. The role of safety culture in safety performance mea-surement. Meas. Bus. Excell. 7 (4), 20–28. http://dx.doi.org/10.1108/13683040310509287.

Barro-Torres, S., Fernández-Caramés, T.M., Pérez-Iglesias, H.J., Escudero, C.J., 2012.Real-time personal protective equipment monitoring system. Comput. Commun. 36(1), 42–50. http://dx.doi.org/10.1016/j.comcom.2012.01.005.

Blewett, V., 1994. Beyond lost time injuries: positive performance indicators for OHS,summary paper. Positive Performance Indicators: Beyond Lost Time Injuries: Part1—Issues 1–55.

Brewer, A., Sloan, N., Landers, T.L., 1999. Intelligent tracking in manufacturing. J. Intell.Manuf. 10 (3–4), 245–250. http://dx.doi.org/10.1023/A:1008995707211.

Carbonari, A., Giretti, A., Naticchia, B., 2011. A proactive system for real-time safetymanagement in construction sites. Autom. Constr. 20 (6), 686–698. http://dx.doi.org/10.1016/j.autcon.2011.04.019.

Cheng, T., Migliaccio, G.C., Teizer, J., Gatti, U.C., 2012. Data fusion of real-time locationsensing and physiological status monitoring for ergonomics analysis of constructionworkers. J. Comput. Civ. Eng. 27 (3), 320–335. http://dx.doi.org/10.1061/(ASCE)CP.1943-5487.0000222.

Cheng, T., Venugopal, M., Teizer, J., Vela, P., 2011. Performance evaluation of ultrawideband technology for construction resource location tracking in harsh environ-ments. Autom. Constr. 20 (8), 1173–1184. http://dx.doi.org/10.1016/j.autcon.2011.05.001.

Cho, H., Kim, S.W., 2010. Mobile robot localization using biased chirp-spread-spectrumranging. IEEE Trans. Ind. Electron. 57 (8), 2826–2835. http://dx.doi.org/10.1109/TIE.2009.2036633.

Choudhry, R.M., 2012. Implementation of BBS and the impact of site-level commitment.J. Prof. Issues Eng. Educ. Pract. 138 (4), 296–304. http://dx.doi.org/10.1061/(ASCE)EI.1943-5541.0000111.

Choudhry, R.M., Fang, D., 2008. Why operatives engage in unsafe work behavior:Investigating factors on construction sites. Saf. Sci. 46 (4), 566–584. http://dx.doi.org/10.1016/j.ssci.2007.06.027.

Contreras, E., Buchanan, S., 2012. Piloting a personal protection equipment distributionprogram among Chicago day laborers. Am. J. Ind. Med. 55 (2), 159–166. http://dx.doi.org/10.1002/ajim.21996.

Council, D.o.E. a. W.R.O. o. t. A. S. a. C., 2005. Guidance on the use of positive perfor-mance indicators to improve workplace health and safety.

Dingus, T.A., Wreggit, S.S., Hathaway, J.A., 1993. Warning variables affecting personalprotective equipment use. Saf. Sci. 16 (5), 655–673. http://dx.doi.org/10.1016/0925-7535(93)90029-D.

Ergen, E., Akinci, B., Sacks, R., 2007. Tracking and locating components in a precaststorage yard utilizing radio frequency identification technology and GPS. Autom.Constr. 16 (3), 354–367. http://dx.doi.org/10.1016/j.autcon.2006.07.004.

Fung, I.W.H., Lo, T.Y., Tung, K.C.F., 2012. Towards a better reliability of risk assessment:Development of a qualitative & quantitative risk evaluation model (Q 2REM) fordifferent trades of construction works in Hong Kong. Accid. Anal. Prev. 48, 167–184.http://dx.doi.org/10.1016/j.aap.2011.05.011.

Grau, D., Caldas, C.H., Haas, C.T., Goodrum, P.M., Gong, J., 2009. Assessing the impact ofmaterials tracking technologies on construction craft productivity. Autom. Constr. 18(7), 903–911. http://dx.doi.org/10.1016/j.autcon.2009.04.001.

Han, S., Lee, S., 2013. A vision-based motion capture and recognition framework forbehavior-based safety management. Autom. Constr. 35, 131–141. http://dx.doi.org/

10.1016/j.autcon.2013.05.001.HSE, 2002. Strategies to promote safe behavior as part of a health and safety management

system, vol. 430/2002. UK.Hsu, Y.-L., Tai, C.-Y., Chen, T.-C., 2000. Improving thermal properties of industrial safety

helmets. Int. J. Ind. Ergon. 26 (1), 109–117. http://dx.doi.org/10.1016/S0169-8141(99)00058-X.

Ibrahim, M., Moselhi, O., 2016. Inertial measurement unit based indoor localization forconstruction applications. Autom. Constr. 71, 13–20.

Jannadi, O.A., Bu-Khamsin, M.S., 2002. Safety factors considered by industrial con-tractors in Saudi Arabia. Build. Environ. 37 (5), 539–547. http://dx.doi.org/10.1016/S0360-1323(01)00056-7.

Jebelli, H., Ahn, C.R., Stentz, T.L., 2016. Fall risk analysis of construction workers usinginertial measurement units. Saf. Sci. 84, 161–170.

Kaskutas, V., Dale, A.M., Lipscomb, H., Evanoff, B., 2013. Fall prevention and safetycommunication training for foremen: report of a pilot project designed to improveresidential construction safety. J. Saf. Res. 44 (1), 111–118. http://dx.doi.org/10.1016/j.jsr.2012.08.020.

Kelm, A., Laußat, L., Meins-Becker, A., Platz, D., Khazaee, M.J., Costin, A.M., Helmus, M.,Teizer, J., 2013. Mobile passive Radio Frequency Identification (RFID) portal forautomated and rapid control of Personal Protective Equipment (PPE) on constructionsites. Autom. Constr. 36, 38–52. http://dx.doi.org/10.1016/j.autcon.2013.08.009.

Lingard, H., Wakefield, R., Cashin, P., 2011. The development and testing of a hier-archical measure of project OHS performance. Eng. Constr. Archit. Manage. 18 (1),30–49. http://dx.doi.org/10.1108/09699981111098676.

Lombardi, D.A., Verma, S.K., Brennan, M.J., Perry, M.J., 2009. Factors influencing workeruse of personal protective eyewear. Accid. Anal. Prev. 41 (4), 755–762. http://dx.doi.org/10.1016/j.aap.2009.03.017.

Long, J., Yang, J., Lei, Z., Liang, D., 2013. Simulation-based assessment for constructionhelmets. Comput. Methods Biomech. Biomed. Eng. 18 (1), 24–37. http://dx.doi.org/10.1080/10255842.2013.774382.

Mäkinen, H., Mustonen, S., 2003. Features of electric arc accidents and use of protectiveclothing in Finland. Saf. Sci. 41 (9), 791–801. http://dx.doi.org/10.1016/S0925-7535(02)00034-6.

Murali, B., Nagarani, J., 2013. Design and fabrication of construction helmet by usinghybrid composite material. Paper Presented at the 2013 International Conference onEnergy Efficient Technologies for Sustainability, ICEETS 2013, Nagercoil. http://dx.doi.org/10.1109/ICEETS.2013.6533372.

Navon, R., Sacks, R., 2007. Assessing research issues in automated project performancecontrol (APPC). Autom. Constr. 16 (4), 474–484. http://dx.doi.org/10.1016/j.autcon.2006.08.001.

Sawacha, E., Naoum, S., Fong, D., 1999. Factors affecting safety performance on con-struction sites. Int. J. Project Manage. 17 (5), 309–315. http://dx.doi.org/10.1016/S0263-7863(98)00042-8.

Suderman, B.L., Hoover, R.W., Ching, R.P., Scher, I.S., 2014. The effect of hardhats onhead and neck response to vertical impacts from large construction objects. Accid.Anal. Prev. 73, 116–124. http://dx.doi.org/10.1016/j.aap.2014.08.011.

SWA, 2013. Issues in the measurement and reporting of work health and safety perfor-mance: a review.

Teizer, J., Allread, B.S., Fullerton, C.E., Hinze, J., 2010. Autonomous pro-active real-timeconstruction worker and equipment operator proximity safety alert system. Autom.Constr. 19 (5), 630–640. http://dx.doi.org/10.1016/j.autcon.2010.02.009.

Teizer, J., Caldas, C.H., Haas, C.T., 2007. Real-time three-dimensional occupancy gridmodeling for the detection and tracking of construction resources. J. Constr. Eng.Manage. 133 (11), 880–888. http://dx.doi.org/10.1061/(ASCE)0733-9364(2007)133:11(880).

Teizer, J., Cheng, T., Fang, Y., 2013. Location tracking and data visualization technologyto advance construction ironworkers' education and training in safety and pro-ductivity. Autom. Constr. 35, 53–68. http://dx.doi.org/10.1016/j.autcon.2013.03.004.

Teizer, J., Mantripragada, U., Venugopal, M., Walia, A., 2008. Analyzing the travel pat-terns of construction workers. Paper Presented at the Proceedings of the 25thInternational Symposium on Automation and Robotics in Construction, Vilnius,Lithuania.

Torrent, D.G., Caldas, C.H., 2009. Methodology for automating the identification andlocalization of construction components on industrial projects. J. Comput. Civ. Eng.23 (1), 3–13. http://dx.doi.org/10.1061/(ASCE)0887-3801(2009) 23:1(3).

Van Der Schaaf, T., Kanse, L., 2004. Biases in incident reporting databases: an empiricalstudy in the chemical process industry. Saf. Sci. 42 (1), 57–67. http://dx.doi.org/10.1016/S0925-7535(03)00023-7.

Wagner, H., Kim, A.J., Gordon, L., 2013. Relationship between personal protectiveequipment, self-efficacy, and job satisfaction of women in the building trades. J.Constr. Eng. Manage. 139 (10). http://dx.doi.org/10.1061/(ASCE)CO.1943-7862.0000739.

Wales, W.N.S., 2001. Safety meter positive performance measurement tool (RetrievedFebruary, 22: 2009).

Wang, A.-P., Chen, J.-C., Hsu, P.-L., 2004. Intelligent CAN-based automotive collisionavoidance warning system. In: IEEE International Conference on Networking, Sensingand Control.

Wiegand, D.M., 2007. Exploring the role of emotional intelligence in behavior-basedsafety coaching. J. Saf. Res. 38 (4), 391–398. http://dx.doi.org/10.1016/j.jsr.2007.03.010.

Yang, J., Vela, P., Teizer, J., Shi, Z., 2012. Vision-based tower crane tracking for un-derstanding construction activity. J. Comput. Civ. Eng. 28 (1), 103–112. http://dx.doi.org/10.1061/(ASCE)CP.1943-5487.0000242.

Zhang, M., Fang, D., 2013. A continuous behavior-based safety strategy for persistentsafety improvement in construction industry. Autom. Constr. 34, 101–107. http://dx.doi.org/10.1016/j.autcon.2012.10.019.

S. Dong et al. Safety Science 102 (2018) 226–237

237