Fs

Aa

b

a

ARRA

KOSCF

1

aral

ettoIr

tdorppva

0d

Accident Analysis and Prevention 45 (2012) 281– 290

Contents lists available at ScienceDirect

Accident Analysis and Prevention

j ourna l h o mepage: www.elsev ier .com/ locate /aap

uzzy approach for reducing subjectivity in estimating occupational accidenteverity

bel Pintoa,∗, Rita A. Ribeirob, Isabel L. Nunesa

Universidade Nova Lisboa/FCT, Caparica 2829-516, PortugalUninova, Campus UNL/FCT, Caparica 2829-516, Portugal

r t i c l e i n f o

rticle history:eceived 20 May 2011eceived in revised form 20 July 2011ccepted 21 July 2011

eywords:

a b s t r a c t

Quantifying or, more generally, estimating the severity of the possible consequences of occupationalaccidents is a decisive step in any occupational risk assessment process. Because of the lack of his-toric information (accident data collection and recording are incipient and insufficient, particularly inconstruction) and the lack of practical tools in the construction industry, the estimation/quantificationof occupational accident severity is a notably arbitrary process rather than a systematic and rigorous

ccupational accidentseverityonstructionuzzy sets

assessment.This work proposes several severity functions (based on a safety risk assessment) to represent biome-

chanical knowledge with the aim of determining the severity level of occupational accidents in theconstruction industry and, consequently, improving occupational risk assessment quality. We followa fuzzy approach because it makes it possible to capture and represent imprecise knowledge in a simpleand understandable way for users and specialists.

. Introduction

In an occupational safety risk assessment (OSRA), assessing risklways has an element of subjectivity. A bad judgement about aisk level will result in false alarms and inappropriate preventivections or, in the worst cases, no action at all. Defining the severityevel is a component of judging occupational risks.

According to Faber and Stewart (2003), risk is defined as thexpected consequences associated with a given event. In occupa-ional safety, a work accident is an event with consequences; risk is,herefore, the combination of the probability that this accident willccur and the severity of consequences arising from the accident.n our work, we use possibility instead of probability because it isather difficult to make estimates with ill-defined data.

Thus, the notion of severity is useful for understanding occupa-ional risks and mitigating them. However, as soon as one delveseeper into the question in terms of specific cases, it becomes obvi-us that it is rather difficult to estimate the severity factors andespective severity levels because of (1) the existing multiplicity ofossible consequences for a given accident and (2) the diversity of

oints of view regarding possible severity assessments (potentialictims, actual victims, medical staff (physicians), workers, man-gerial staff and safety managers). The apparently simple and easily

∗ Corresponding author.E-mail address: abel.fnpinto@gmail.com (A. Pinto).

001-4575/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.oi:10.1016/j.aap.2011.07.015

© 2011 Elsevier Ltd. All rights reserved.

applicable notion of conducting a severity assessment of an acci-dent becomes complex in practice because of the assessment’simprecision and strong dependence on the analyst’s perception.

How does one estimate severity in a proactive analysis withaccuracy and in a practical way in the construction industry? Amethod to estimate the level of the expected severity of poten-tial work accidents would be of undeniable value to constructioncompanies seeking to improve their understanding of such events.However, because an occupational accident at a construction sitecould cause multiple consequences of highly variable severity, it isa difficult situation to model.

Hence, the definition of fuzzy severity functions (also called thefuzzification process) to express imprecise severity work accidentconsequences is the main objective of this work. Fuzzy member-ship functions (Ross, 2004; Zadeh, 1965) allow easy normalizationand uniformization of all data; therefore, we can then use them asa measure to assess the work accident severity expected level atthe construction sites. Thus, with this fuzzification process, we arecapable of handling uncertainty when estimating work accidentseverity levels and of offering a user-friendly assessment method.

2. Motivation for severity measurements

To capture the multifaceted nature of accident consequencesin the construction industry, a more complete approach wouldinvolve simultaneous consideration of several measurement fac-tors. The measurement is no longer simple but is multifactorial

2 s and

imsj

baoe

ntWiotro(ei

p

adste

o1qi

3

lpii

wtaouihboi

tvirdist

ac

82 A. Pinto et al. / Accident Analysi

nstead, and possible correlations between the different factorsust be taken into account if the intricate nature of consequence

everity is to be correctly addressed and not reduced to a simpleuxtaposition of factors (Cuny and Lejeune, 1999).

The ultimate aim of this study is to provide a value that cane used as a satisfactory overall measurement of the severity ofn accident, and the various factors cannot a priori be consideredf equal importance. It is, therefore, necessary to use collectivexpertise to define which factors are relevant in practice.

Related works (Cuny and Lejeune, 1999) used such factors asumber of days off work and degree of permanent disablemento establish precise levels of the severity scale. Gennarelli and

odzin (2006) pointed out several dimensions of severity, includ-ng threat to life, mortality (theoretical, expected, actual), amountf energy dissipated/absorbed, hospitalization (need for ICU, LOS),reatment cost, treatment complexity, length of treatment, tempo-ary and permanent disability, permanent impairment and qualityf life, but without establishing a measurement scale. Other authorsAneziris et al., 2008) proposed or used “coarse” scales to assess lev-ls of consequence severity: lethal injuries, nonlethal permanentnjuries, and recoverable injuries.

From all of these dimensions, only the amount of energy dissi-ated/absorbed may be useful in a proactive analysis.

In this study, we sought to qualitatively assess the occupationalccident severity from predictors related to the amount of energyissipated/absorbed that can be evaluated in situ, such as heights,peeds, weights, and morphology of moving vehicles. We also usedhe biomechanical limits of the human body as determined in sev-ral studies.

To capture and represent the severity levels of the most relevantccupational work accidents, we used fuzzy set theory (FST) (Zadeh,965) because it is a flexible and versatile method to express bothualitative and ill-defined borders. More details about FST suitabil-

ty are discussed in the next section.

. Rational motivation for using FST

In the real world, vagueness and ambiguity exist because of theimitations of our language and other factors, such as context anderception. Closely related to this ambiguity is the question of lex-

cal imprecision in natural language; when expressing knowledge,ndividuals would rather use words than numbers.

OSRA deals with uncertain situations, that is, situations in whiche do not have complete and accurate knowledge about the sys-

em state, such as estimate severity consequences of occupationalccidents. The vagueness and uncertainty underlying the processesf severity estimation are not truly stochastic. Usually, stochasticncertainty stems from the lack of complete knowledge concern-

ng the future state of a system. Moreover, stochastic uncertaintyas to do with the randomness of the future and is usually treatedy theories of probability and statistics. However, this type of eventr statement (handled by the theories of probability and statistics)s necessarily well defined.

Our view in this work is that the nature of the work in construc-ion sites reveals uncertainties that derive from imprecise and/orague concepts and limits, not from stochastic events. On-sitenspections generally use linguistic expressions rather than met-ics to assess the safety risks. Additionally, legal records, statisticalata and site documentation produced by companies are generally

nsufficient for determining risks. These facts increase the impreci-ion and inaccuracies of the OSRA process, and this imprecision is

he reason why we use a fuzzy approach.

For systems in which imprecise and inaccurate information isvailable, fuzzy concepts and techniques provide suitable ways toollect observed input data and represent it in a uniform and scal-

Prevention 45 (2012) 281– 290

able way (Ross, 2004). FST was formulated in 1965 by Lotfi Zadeh(Zadeh, 1965), and it provides a mathematical framework for thesystematic treatment of vagueness and imprecision. More specifi-cally, FST may be viewed as an attempt to formalize two remarkablehuman capabilities: (1) the capability to reason and make ratio-nal decisions in an environment of imprecision, uncertainty, andincompleteness of information (environments of imperfect infor-mation) and (2) the capability to perform a wide variety of physicaland mental tasks without any measurements (at least in a quanti-tative way).

A fuzzy set presents a boundary with a gradual contour in con-trast with classical crisp sets, which present a discrete border.Formally, let U be the universe of discourse and u be a generic ele-ment of U, a fuzzy subset A defined in U, one set of the dual pairs(Zadeh, 1965):

A = {(u, �A(u))∣∣u ∈ U } (1)

where �A(u) is designated as membership function or membershipgrade of u in A. The membership function associates to each elementu (of U) a real number �A(u) in the interval [0,1] to express thedegree of belonging to the set.

From an encoding point of view, fuzzy sets support the repre-sentation of qualitative knowledge and its uncertainty as a uniqueentity. The resulting representation, usually called fuzzification,is notably flexible, can be easily coupled with nonfuzzy forms ofknowledge representation, and can be manipulated by a variety ofevaluation methods (Ross, 2004). However, fuzzification is usuallya difficult task; hence, we propose to use open linear triangularfunctions in this study to simplify the process.

FST deals with possible, not probable, events. Possibilitydescribes whether an outcome can happen, while probabilitydescribes whether it will happen (Ross, 2004). However, probabilitytheory is not displaced by FST; the two approaches are complemen-tary. Probabilistic risk assessment approaches have been widelyutilized (Tixier et al., 2002), but they may be difficult to use in cir-cumstances in which there is a lack of knowledge or in imprecisesituations.

Fuzzy sets seem to be quite relevant in three classes of appli-cations (Ross, 2004; Dubois and Prade, 1998): classification anddata analysis, reasoning under uncertainty, and decision-makingproblems. In our work, we use the lattermost application of deci-sion making because it will allow the combination of all risk factors(related to expected severity and possibility of work accident occur-rence) using aggregation operators to define a general level of riskassessment (Li and Yen, 1995). This topic is not within the scope ofthis paper.

In summary, in construction, vague terms are unavoidablebecause safety professionals often assess risks in qualitative lin-guistic terms. Under these circumstances, conventional OSRAapproaches may not be able to model safety for the whole con-struction process as effectively and efficiently as FST approaches.

4. Research methodology

Our model development began with a literature survey of pre-viously published work. During the research, it became evidentthat there was no methodology available to estimate occupationalaccident severity that was simultaneously accurate and practical.

We followed a four-step approach to develop the model: (1)Based on the literature, we identified a list of accident modes for

occupational accident scenarios (see next section). (2) We selectedthe energy related factors considered to be important in evaluatingthe severity of each accident mode. (3) The relationships betweenthese factors and human biomechanical limits were taken using

s and Prevention 45 (2012) 281– 290 283

pd

taismc

fsasSEpa

tethof

wif

stqa

5

assrrtcc

b(i

nodwMt

••••

•

Table 1Head tolerances to impact loads.

Cranial bones Compression force (kN)

Frontal 3.6–9.0Zygoma 0.5–2.9Temporo-parietal 5.0–12.5

A. Pinto et al. / Accident Analysi

hysics models. (4) All severity functions associated with each acci-ent mode were determined using fuzzy sets.

After wards, we conducted a brainstorming session attended bywo Ph.D. professors, one Ph.D. student and three M.Sc. students,ll with research experience related to safety. In the brainstorm-ng session, the complete set of factors that could contribute to theeverity of each accident mode was discussed. Next, it was deter-ined which factors were relevant in practice and which predictors

ould characterize them and how.As a first approach, we tried modeling work accident severity

rom statistical data that related such factors as heights and landingurfaces with physical consequences (body fractures, respiratoryrrest). In Portugal, we contacted five insurance companies, con-truction associations, and the statistical bulletin of the Office oftrategy and Planning; in Brazil, we contacted Fundacentro; and inngland, we contacted the Health and Safety Executive. It was notossible to obtain the necessary data to continue the work fromny of these organizations.

Our second approach was to model work accident severity usinghe empirical knowledge of a group of safety experts. However, thexperts disagreed about various predictors; for example, for falls,he field of results (for maximum severity (1) depending on theeight) varies between 1.2 and 4.0 m. The experts failed to agreen a precise value, and when questioned about their reasons, theyailed to provide an explanation.

Hence, the relationships between energy factors and severityere established by taking into account the biomechanical lim-

ts of human body and physics models, and severity membershipunctions associated with each accident mode were produced.

The results, i.e., the fuzzy sets, were presented to a pool of eightafety experts using a 5-point traditional Likert scale questionnaireo verify that the results agreed with their empirical knowledge. Theuestionnaires were administered individually. In general, expertsgreed with the presented results (see Annexes 1 and 2).

. Work accidents severity model

In occupational safety, risk predictors are usually imprecise,mbiguous or vague. Therefore, these variables are fuzzy andhould be represented by membership functions (Ross, 2004). Inhort, fuzzification is the process of representing concepts fromeal life (e.g., fall height) as membership functions. In this work, theisk parameter functions are fuzzified in terms of accident severityo express how serious a certain accident can be. For instance, 0orresponds to the absence of any damage (to any worker) and 1orresponds to the maximum severity.

To rate the accident severity, we follow the rationale providedy BS 8800:2004 (BSI, 2004) regarding the extreme harm levelseverity 1): premature death or permanent major disability (fatalnjuries, amputations, multiple injuries or serious fractures).

After defining the level 1 severity for occupational accidents, weeed to define the accident modes. We propose the following listf accident modes. This list is based on a preliminary list of acci-ent modes for occupational accident scenarios by Ale et al. (2008),hich has been adapted for the construction industry (Jeong, 1998;üngen and Gürcanli, 2005; Hyoung et al., 2009). The list includes

he following:

Fall.Contact with electricity.Struck by moving vehicle (including heavy equipment).

Injured by falling/dropped/collapsing object/person/wall/vehicle/crane, which falls under gravity (including building orstructure collapse and slipping handheld tools).Trapped by cave-in (during or after excavation).

Occipital 2.0–4.2Maxilla/mandible 0.8–3.4

• Hit by rolling/sliding object or person (including stuck againstobject or equipment and caught in or compressed by equipmentor objects).

• Contact with the moving parts of machinery (including injury byhandheld tools operated by oneself).

• Lost buoyancy in water.• Fire and explosion.

5.1. Injury criteria

To model the membership functions (fuzzification) for theseaccident modes, we need to consider the body parts most likelyto be affected (e.g., hit), particularly in the case in which workersdo not use any personal protective equipment. In terms of mechan-ical impacts, we start by establishing a threshold of values for thebiomechanical limits of the human body in terms of injury forces(kN). In the literature (Prasad, 1999; Eppinger et al., 2000; Yang,2002; Chaffin et al., 2006; Sayed et al., 2008), there are specificmodels for body segments that describe the severity of the conse-quences of an impact. In this work, we will use these definitions forthe different affected body parts as follows.

Head injury criteria: There are several models for head injurycriteria, and some models are more sophisticated than others. Someof these models use data that are hard to collect at construction sites(Sayed et al., 2008), and in practical terms, they do not add value tothe results of a safety risk assessment. In most studies, head injuryseverity is a function of the average head acceleration and its timeduration (Prasad, 1999; Eppinger et al., 2000). For head thresholdlimits, we will use the values depicted in Table 1 from Yang (2002).

Neck injury criteria: In terms of peak tensile force, the limit sug-gested (based on available biomechanical data) by McElhaney andMyers (1993) is 3.1 kN. Eppinger et al. (2000) determined the fol-lowing values: for tension, 8.216 kN (large-sized male, or LSM) and4.287 kN (small-sized female, or SSF), and for compression, 7.44 kN(LSM) and 3.88 kN (SSF).

Chest injury criteria: Viano (1989), cited by Yang (2002), deter-mined 5.5 kN for thorax lateral impact forces.

Femur injury criteria: The FMVSS208 adopted a peak compres-sive force of 10 kN as the injury criteria without any limit of timeduration. Eppinger et al. (2000) determined the femur load to be12.7 kN (LSM) and 6.8 kN (SSF).

Tibia injury criteria: The maximum allowable tibia compressiveforce (for break) could not exceed 8 kN (Mertz, 1993).

5.2. Accident modes severity fuzzification

5.2.1. FallsThe severity of an injury is determined by the age of the victim,

fall height, body impact site and impact surface (Cory and Jones,2006). For modeling, we consider the following set of factors:

• Height of fall (which determines the velocity at the impact sur-

face);

• Type of surface (given an impact velocity, or equivalently, giventhe height of the fall, this variable determines the magnitude ofthe force exerted on the fallen person).

2 s and Prevention 45 (2012) 281– 290

m

V

wf

ti(

u

I

wfimhdcii

iiesiam

�

fdda

flas

(r(

7

HetiT

bi1

rcw

sfe

0

0,2

0,4

0,6

0,8

1

0 5 10 15 20 25 30 35 40Height (in cm)

Seve

rity

SandConc ret e

84 A. Pinto et al. / Accident Analysi

To characterize the fall event, the impact velocity was deter-ined using

=√

2gh (2)

here g is the acceleration due to gravity (9.81 m/s2) and h is theall height.

The fall height was defined as the distance from the body’s cen-er of mass at the start of the fall to the ground. The mass centers equivalent to 57% of the height for males and 55% for femalesKnudson, 2007).

The change in momentum during impact (I) was determinedsing

= mV(COR + 1) (3)

here m, V, and COR are the body mass, impact velocity, and coef-cient of restitution of the impact surface, respectively. COR is theeasure of the elasticity of the surface. In falls onto a surface with a

igh COR, such as a sand pile, most of the energy from the fall will beissipated. Conversely, in falls onto surfaces with a low COR, such asoncrete, most of the fall energy will be absorbed by the body uponmpact. This energy absorption typically leads to a greater risk ofnjury (Thompson et al., 2009).

The human body limits proposed in the literature are expressedn terms of force. The change in momentum corresponds to exert-ng a force for a certain period of time, assuming that the forcexerted during the time of the impact is constant. (This expres-ion is an approximation. Typically, the interaction forces duringmpact have a thin, approximately Gaussian shape and, therefore,

peak intensity much higher than the value determined for theean force.)

t = I

Fmed(4)

According to Thompson et al. (2009), impact durations rangerom 12.1 ms to 27.8 ms (for the falls of young children), and impacturations were lower for surfaces with a lower COR. Because impacturations also depend on the object deformation, for adults, wessumed an impact duration of 28 ms.

Thompson et al. (2009) proposed a COR of 0.39 for a carpetedoor and pointed out that this COR was not found to be significantlyssociated with injury severity. Therefore, we used 0.39 for sandurfaces (or similar) and 0.8 for concrete ground (or similar).

Assuming a mean weight for a Portuguese worker of 74 kgBarroso et al., 2005) when falling on a sand surface and a crite-ion for a compression neck injury of 7440 N according to Eppingersee Section 5.1). Thus, we have

440 × 28 × 10−3 = 1.39 × 74√

2 × 9.8 × h ⇒ h = 0.21 m (5)

owever, this value of 0.21 m does not agree with empirical knowl-dge. Likewise, considering only the head weight (5.0 kg, accordingo Thorn et al., 1998, see Section 5.1), the value for the drop heights 45.3 m, which also does not agree with empirical knowledge.herefore, this is not the correct approach to the problem.

Chaffin et al. (2006) referred to the head, neck and trunk as aody segment and determined a weight value of 31.22 kg for a male

n the 5th percentile. Using these data, we obtained a value for h of.16 m.

The problem of concordance with the empirical data may beelated to the estimated time of impact. Thompson’s article is notlear about the factors he used to estimate the collision times, ande used values for an adult.

According to Knudson (2007), the center of body mass in theagittal plane is equivalent to 57% of the height for males and 55% foremales. The average Portuguese worker height is 1.69 m (Barrosot al., 2005), and the center of gravity is at a height of 0.96 m.

Fig. 1. Linguistic variable Severity falls.

Therefore, the fall height limit calculated from the surface wherethe feet stands is 0.20 m (20 cm), where the severity reaches thelimit of 1.

For a concrete surface, with h = 85 cm, the severity reaches thelimit of 1 for falls at the same level.

Although there are records of accidents with much higher dropheights and with less severe consequences, there are others thatconfirm this result. Based on conservative criteria and consideringthat these calculations were made for one of the weakest parts ofthe body, the experts agreed with these results.

Finally, we propose a linguistic variable (see Fig. 1) with twolabels: Severity falls = (Sand-surface, concrete surface) where eachlabel is represented by a membership function:

Sand surface ={

0.05x, 0 ≤ x ≤ 211, x > 21

Concrete surface ={

1, x ≥ 0

There are some limitations to our model. The human center ofbody mass can move around because joints allow the body seg-ments to move. The change in momentum observed during impactwas measured at the center of mass, so the change in momentumis difficult to determine accurately. Safety experts agree based ontheir experiences that the position where the contact occurs is cru-cial to the accident severity; for example, when the worker hits hisor her back, the severity is always greater (compared with othersimilar falls with the same height and surface) because it is a help-less fall. The contact surface can be important, especially in smallfalls. The severity may be different if the fall is on a sand surface, aconcrete surface, corners, sharp edges or iron points.

There are several other factors that could contribute to this typeof injury; namely, initial body position, fall dynamics, body massand age. However, in this work, these factors were not taken intoaccount for the reasons given in Section 6.

5.2.2. Contact with electricityThe severity of electrical injuries can vary widely from an

unpleasant tingling sensation caused by a low-intensity current tothermal burns, cardiopulmonary arrest, and death. Thermal burnsmay result from burning clothing that is in contact with the skinor from electric current traversing a part of the body. When cur-rent transverses the body, thermal burns may be present at thepoints where the current enters and exits the body and internallyalong its pathway. Cardiopulmonary arrest is the primary cause ofimmediate death for electrocution. Cardiac arrhythmias, includingventricular fibrillation, ventricular asystole, and ventricular tachy-

cardia that progresses to ventricular fibrillation, may result fromexposure to low or high-voltage current. Respiratory arrest mayresult from electrical injury to the respiratory center in the brainor from tetanic contractions or paralysis of respiratory muscles.

A. Pinto et al. / Accident Analysis and

0

0,5

1Se

verit

y

Voltage (in Volt)

Wet environment

Dry environment

iupm

alnpahtc

atsfco

twv

1ia0

wUlbw

tef

D

D

fipt

chest injury criteria of 5.5 kN, the maximum vehicle safe speed is

Fig. 2. Linguistic variable Severity contact-electricity.

Because ventricular fibrillation has the most drastic physiolog-cal effect because of the lack of resilience of the human body, wese it as a criterion to model the severity of electrical injuries. As aredictor, we use the tension value in volts. As in the other accidentodes, we assume that there are no protective devices.Standard IEC 479-1:1984 refers the value of 40 mA (AC current)

s the threshold of ventricular fibrillation (the lower limit of currentikely to cause the heart to cease functioning correctly, which isot a single value of current but rather varies with duration). Therobability exceeds 50% for currents of 90 mA and for passage timesbove 5 s. Standard IEC 479-1:1984 states that the impedance of theuman body is voltage dependent (varies inversely as a function ofhe applied voltage) and path dependent (varies according to theurrent path through the human body).

However, IEEE Std 80-1986 uses other important variables, suchs human body mass and soil resistivity. The IEC 479-1 safety cri-eria are rather complex, while the safety criteria of IEEE Std 80 areimplified and address more relevant practical factors. Given theact that the safety criteria include comfortable safety margins, onean conclude that the simplicity of IEEE Std 80 does not compromiseccupational safety.

According to information provided by REN (2011), in Portugal,he value of soil resistivity varies with geographical location andith the seasons; however, 100 � m can be treated as the average

alue.Using a conservative criterion for shock duration, we chose

89.6 V as the voltage limit to reach the maximum severity accord-ng to permissible touch voltages per IEEE Std 80-1986, assuming

50 kg human body and a probability of ventricular fibrillation of.5%.

In accordance with the IEC 479-1standard, moisture (normalater) lowers the impedance values of the human body by 10–25%.sing conservative criteria, we assumed that the impedance was

owered by 25%. Considering that other factors remain constantased on Ohm’s law, in wet environments, the maximum voltageas decreased 25% from 189.6 to 142.2 V.

Finally, we propose a linguistic variable (see Fig. 2) withwo labels, Severity contact-electricity = (dry enviroment, wet-nvironment), which are represented by the following membershipunctions:

ry environment ={

0.00714x, 0 ≤ x ≤ 1401, x > 140

ry environment ={

0.00526x, 0 ≤ x ≤ 1901, x > 190

Limitations: In this accident mode and when using ventricular

brillation as the severity criterion, rescue time can be an importantredictor. However, we did not consider this aspect because rescueime is difficult to estimate proactively.

Prevention 45 (2012) 281– 290 285

5.2.3. Struck by a moving vehicleThe following set of factors was considered for the modeling

(Mizuno and Kajzer, 1999):

• Vehicle speed and mass (which determines the energy impact ona human body);

• Vehicle design (vehicle geometry).

In our bibliographic research, no studies were found related toconstruction sites. Therefore, we extrapolate data from road acci-dent studies.

The vehicle morphology is important to determine the accidentseverity. The study by Lefler and Gabler (2004) pointed out that,given an impact speed, the probability of serious head and thoracicinjury is substantially greater when the striking vehicle is an LTVrather than a conventional car.

In terms of the vehicle morphology, it is important to considerthe probable zone of contact: lower limbs, main body and head(including neck). Another study developed by Ballesteros et al.(2004) noted that pedestrians hit by SUVs and pickup trucks weremore likely to have more severe injuries compared with conven-tional cars, but the increase in danger may be explained primarilyby larger vehicle weights and faster vehicle speeds. Regardless ofvehicle weight, pedestrians struck at slower speeds by SUVs, pick-ups, and vans incurred a rate of brain, thoracic, and abdominalinjuries twice that of those struck by conventional cars, which indi-cates that vehicle design contributes to severity.

If the worker is stopped at the moment of impact and the col-lision is perfectly inelastic, i.e., after the collision, the vehicle andworker move together:

(mW + mV )V(w+v) = mV VV (6)

where mW is the worker mass, mV is the vehicle mass, VV isthe vehicle speed and V(w+v) is the (scalar) velocity of the wholeworker + vehicle.

If the worker is moving, mWVW is added on the right side ofexpression (6):

V(w+v) = mV

mW + mVVV (7)

The change in momentum during impact (on the worker body)was determined by

M = mW �V(w+v) (8)

However, the human body biomechanic limits established inthe literature are given in terms of force. The change in momentumcorrespond to a force exerted for a certain period of time:

Fmed = M

�t(9)

Fmed is the average force exerted during the time interval consid-ered.

We again assume an impact duration of 28 ms. For a skid steerloader with 3348 kg of mass, we assume that this machine’s mor-phology is similar to that of a conventional car; the impact mainlyaffects the lower limbs. Assuming a tibia injury criteria (see Sec-tion 5.1) of 8 kN and a mean weight for a Portuguese worker of74 kg (Barroso et al., 2005), the maximum safe speed of the vehicleis 3.08 m/s (11.09 km/h).

For a hydraulic excavator with 21,860 kg of mass, we assumethat this machine morphology is similar to an SUV. Considering the

2.08 m/s (7.49 km/h).Finally, we propose a linguistic variable (see Fig. 3) with

two labels, Severity struck by moving vehicle = (Skid Steer Loader,

286 A. Pinto et al. / Accident Analysis and Prevention 45 (2012) 281– 290

0

0,2

0,4

0,6

0,8

1

0 10 20

Seve

rity

Skid Steer Loade r

Hydraulic exc avator

Hb

S

H

cb

5o

•

•

bisi

V

wf

ac

(

wmt

sc

V

d

M

M

0

0,2

0,4

0,6

0,8

1

0 5 10 15Height (in m)

Seve

rity Brick 11

Claw hammer

Vel ocity (in km/h)

Fig. 3. Linguistic variable Severity struck by moving vehicle.

ydraulic Excavator), where each label is represented by a mem-ership function:

kid Steer Loader ={

0.09x, 0 ≤ x ≤ 11.081, x > 11.08

ydraulic Excavator ={

0.13x, 0 ≤ x ≤ 7.491, x > 7.49

Limitations: The vehicle stiffness is not considered because atonstruction sites all of the vehicles have hard surfaces (surfacesarely deform).

.2.4. Injured by a falling/dropped/collapsingbject/person/wall/vehicle/crane that is falling under gravity

The following set of factors was considered for the modeling:

Height of the fall (which determines the velocity at the impactsurface);Mass of the object (together with the velocity, determines theenergy impact);

In the case of falling objects, the most exposed body part wille the head. The most commonly used object at construction sites

n Portugal is brick 11 (made of clay and 30 cm × 20 cm × 11 cm inize), which has a mass of 45 kg. To characterize a fall event, thempact velocity was determined using

B =√

2gh (10)

here g is the acceleration due to gravity (9.81 m/s2) and h is theall height.

Because of the principle of conservation of momentum andssuming that the collision is perfectly inelastic, i.e., during theollision, the brick and the worker’s head move together, let

mB + mH)V(h+b) = mBVB + mHVH (11)

here mB is the brick mass, VB is the brick velocity, mH is the headass, VH is the head velocity and V(h+b) is the (scalar) velocity of

he whole head + brick.Thorn et al. (1998) present a head form table listing weights,

hapes and sizes of skulls. We chose the DOT medium size with aircumference of 56 cm and a mass of 5.00 kg. From (10) and (11),

(h+b) (m/s) = 2.09√

h (12)

The change in momentum during impact (on the head) wasetermined by

= mH�V(h+b) (13)

(N m/s) = 10.45√

h

Fig. 4. Linguistic variable Severity injured by a falling/dropped/collapsingobject/person/wall/vehicle/crane that is falling under gravity.

The average force applied to the head was determined by

Fmed = M

�t(14)

where �t is the contact time between the brick and the head.

�t = �x

V(h+b)(15)

where �x is the elastic absorption of the skull bones (we consider5 mm to be a representative value, adapted from Jaslow, 1990).

Fmed = 0.00436h (16)

In a vertical object fall, the bones most likely to be hit are thefrontal and parietal bones. Assuming the average values indicatethe tolerances to impact loads, we obtain 6.3 kN for frontal bonesand 8.75 kN for parietal bones. Choosing the smallest value, we con-clude that the severity reaches the maximum for a fall of 1.44 m inheight (for an object with a 4.5 kg mass).

The most widely used tools at construction sites are hammersand flat chisels. Considering a claw hammer with a mass of 800 g,for the frontal bone biomechanic limit (6.3 kN – see Section 5.1),the severity reach the maximum for a fall of 16.9 m.

Finally, we propose a linguistic variable (see Fig. 4) withtwo labels: Severity injured by a falling/dropped/collapsingobject/person/wall/vehicle/crane that is falling under grav-ity = (Brick 11, Claw hammer) where each label is represented by amembership function:

Brick 11 ={

0.69x, 0 ≤ x ≤ 1.441, x > 1.44

Claw hammer ={

0.06x, 0 ≤ x ≤ 16.91, x > 16.9

Limitations: The object’s stiffness was not considered becauseat construction sites almost all of the objects (that may fall) havehard surfaces (surfaces barely deform). The object morphology (e.g.,corners or sharp edges) was not considered because, during the fall,it is not possible to estimate accurately which object surface willhit the worker.

5.2.5. Cave-insThe rescue time factor was considered for modeling cave-ins.

In this accident mode, the rescue time is the determining factorbecause the life of an injured person who is deprived of air dependson being rescued quickly.

According to Guidelines 2000 for Cardiopulmonary

Resuscitation and Emergency Cardiovascular Care (2000), thepercentage of survival decreases approximately 7–10% with everyminute that defibrillation is delayed until, at 10 min, the proba-bility of survival is almost 0. We propose a fuzzy set (see Fig. 5),

A. Pinto et al. / Accident Analysis and Prevention 45 (2012) 281– 290 287

0

0,2

0,4

0,6

0,8

1

0 2 4 6 8 10 12

Seve

rity

Rescue tim e

Sf

R

hessbeav

5

••

b

(tt

F

wm

vts8

T

P

wo

tmbe

I

w

0

0,20,4

0,60,8

1

0 0,2 0,4 0,6 0,8 1 1,2

Linear aceleration (in m/s)

Seve

rity Hydraulic

excavator

Backhoeloader

Council of 17 May 2006 determined a set of particularly dangerousmachine categories so labeled because of the potential severity ofaccident consequences:

00,20,40,60,8

1

020

040

060

080

0

Seve

rity

Sliding ston e

Time (in min)

Fig. 5. Fuzzy set Severity cave-ins.

everity cave-ins = (Rescue time), represented by a membershipunction:

escue time ={

0.1x, 0 ≤ x ≤ 101, x > 10

Limitations: The soil type (e.g., soft or with rock blocks) mayave a serious influence on the accident severity, but it is difficult tostablish a relation between soil type and the occupational accidenteverity. In the literature search, we did not find any mention of thisubject. The safety experts did not agree that the soil type shoulde used as a determining factor of severity because, for example,ven though a rocky soil may cause more serious injuries, it canlso create channels or air pockets that can extend the life of theictim.

.2.6. Hit by rolling/sliding object or personThe following set of factors was considered for the modeling:

Object mass and velocity (which determine the energy impact);Object morphology (e.g., corners or sharp edges).

At construction sites, this accident mode could be observedecause of the release of stones from slopes.

A stone sliding from a slope with an angle of � and no frictionwhich is difficult to estimate because the friction depends on soilype, soil moisture and the shape and surface of the stone) strikeshe victim with a force of

= m · g · sen � (17)

here g is the acceleration due to gravity (9.81 m/s2), m is the objectass and � is the slope angle.A small/medium stone (up to 1000 kg) does not have a large

olume, so it will probably hit the victim’s legs; the tibia injury cri-eria of 8 kN (see Section 5.1) was assumed. For example, a spheroidtone sliding from a slope with an angle of 70◦ with a mass of68.7 kg (or more) can crack the tibia of a worker.

Another example is the rolling boom of hydraulic excavators.he boom of a hydraulic excavator hits the victim with a force of

= I · ̨ (18)

here I is the moment of inertia (which depends on the geometryf the rolling object) and ̨ is the angular acceleration.

For the hydraulic excavator, the boom mass is 5820 kg, andhe boom range is 9.0 m (for a 1.2 m height). Therefore, the boom

oment of inertia could be calculated approximately (treating theoom as a thin bar with a length that is much larger than the diam-ter whose axis of rotation is at one end) by

= 13

M · L2 (19)

here M is the boom mass and L the boom range.

Fig. 6. Linguistic variable Severity hit by rolling/sliding object or person (Rollingobject Hydraulic excavator, Rolling object Backhoe loader).

At a height of 1.2 m, the body part that it is hits is the chest.The thorax lateral impact injury tolerance is 5.5 kN (Viano (1989),cited by Yang (2002)) such that the maximum ̨ is 0.035 rad/s2,corresponding to a linear acceleration at the boom tip of 0.315 m/s2.

For a backhoe loader, the boom mass is 2547 kg, the boom rangeis 4.5 m (at a height of 1.2 m), and the maximum ̨ is 0.21 rad/s2,corresponding to a linear acceleration at the boom tip of 0.96 m/s2.

Finally, we propose a linguistic variable (see Figs. 6 and 7)with three labels: Severity hit by rolling/sliding object or per-son = (Sliding stone, Rolling object hydraulic excavator, Rollingobject backhoe loader). Each label is represented by a membershipfunction:

Sliding stone ={

0.00115x, 0 ≤ x ≤ 868.71, x > 868.7

Rolling object Hydraulic excavator ={

3.17x, 0 ≤ x ≤ 0.3151, x > 0.315

Rolling object Backhoe loader ={

1.04x, 0 ≤ x ≤ 0.961, x > 0.96

5.2.7. Contact with machinery moving parts (including injuriesfrom handheld tools operated by oneself)

The following set of factors was considered for the modeling:

• Machine category.• Machine power.

Directive, 2006/42/EC of the European Parliament and of the

Mass (in kg)

Fig. 7. Linguistic variable Severity hit by rolling/sliding object or person (Slidingstone).

2 s and Prevention 45 (2012) 281– 290

•

••

•

•

•

••

•

•

•

••

••

5

tsi(td

ii

ts

i

0

0,2

0,4

0,6

0,8

1

0 5 10 15

Seve

rity

Brick 11Claw hamm er

88 A. Pinto et al. / Accident Analysi

Circular saws (single- or multiblade) for working with wood andmaterial with similar physical characteristics or for working withmeat and material with similar physical characteristics of thefollowing types: (a) sawing machinery with fixed blade(s) dur-ing cutting, having a fixed bed or support with manual feed ofthe workpiece or with a demountable power feed; (b) sawingmachinery with fixed blade(s) during cutting, having a manu-ally operated reciprocating saw-bench or carriage; (c) sawingmachinery with fixed blade(s) during cutting, having a built-inmechanical feed device for the workpieces with manual loadingand/or unloading and; or (d) sawing machinery with movableblade(s) during cutting, having mechanical movement of theblade with manual loading and/or unloading.Hand-fed surface planning machinery for woodworking.Thicknessers for one-side dressing having a built-in mechanicalfeed device, with manual loading and/or unloading for wood-working.Band-saws with manual loading and/or unloading for workingwith wood and material with similar physical characteristics orfor working with meat and material with similar physical char-acteristics of the following types: (a) sawing machinery withfixed blade(s) during cutting, having a fixed or reciprocating-movement bed or support for the workpiece or (b) sawingmachinery with blade(s) assembled on a carriage with recipro-cating motion.Hand-fed tenoning machinery with several tool holders forwoodworking.Hand-fed vertical spindle moulding machinery for working withwood and material with similar physical characteristics.Portable chainsaws for woodworking.Presses, including press-brakes, for the cold working of metals,with manual loading and/or unloading, whose movable workingparts may have a travel exceeding 6 mm and a speed exceeding30 mm/s.Machinery for underground working of the following types: (a)locomotives and brake-vans or (b) hydraulic-powered roof sup-ports.Manually loaded trucks for the collection of household refuseincorporating a compression mechanism.Removable mechanical transmission devices, including theirguards.Vehicle servicing lifts.Devices for the lifting of persons or of persons and goods involvinga hazard of falling from a vertical height of more than 3 m.Portable cartridge-operated fixing and other impact machinery.Severity should be considered the maximum (1) for all this cate-gories of machinery.

.2.8. Lost buoyancy in waterThe outcome following drowning depends on the duration of

he submersion, the water temperature, and how promptly CPR istarted. Case reports have documented intact neurological survivaln small children following prolonged submersion in icy watersSuominen et al., 2002). For the modeling, the rescue time fac-or was considered because the life of an injured person who iseprived of air depends on being rescued quickly.

Based on survival rates according to the rescue time (see figuren Section 5.2.5), we propose a fuzzy set, as seen in Fig. 5 (Sever-ty cave-ins = (Rescue time)).

Limitations: The water temperature is not considered because

here are no icy temperatures in countries with a moderate climate,uch as Portugal.

In water with a current, the flow can make it difficult andncrease the rescue time because it hampers locating the victim,

Height (in m)

Fig. 8. Examples of the expected work accident severity.

but no one from a rescue team could give us any data about thisphenomenon.

5.2.9. Fire and explosionA literature survey about contact with extreme temperature

severity models proved impractical for the scope and objectives ofthis study for two reasons: (1) the models are theoretical [Pennes’equation and thermal wave model of bioheat transfer – TWMBT(Liu et al., 1999)] and (2) factors such as the temperature of theskin surface, the activation energy, and the thickness of the tissueare impossible to estimate because of the geographical and climaticvariations.

As for explosions, we did not find any consequence models, andthe variability of the factors is complex. Directive 99/92/EC of the ofthe European Parliament and of the Council of 16 December 1999determines a set of hazardous places classified in terms of zoneson the basis of the frequency and duration of the occurrence of anexplosive atmosphere:

Zone 0—A place in which an explosive atmosphere consisting of amixture of air and dangerous substances in the form of gas, vaporor mist is present continuously or for long periods or frequently.Zone 1—A place in which an explosive atmosphere consisting of amixture of air and dangerous substances in the form of gas, vaporor mist is likely to occur in normal operation occasionally.Zone 2—A place in which an explosive atmosphere consisting of amixture of air and dangerous substances in the form of gas, vaporor mist is not likely to occur in normal operation, but if it doesoccur, it will persist for a short period only.Zone 20—A place in which an explosive atmosphere in the form ofa cloud of combustible dust in the air is present continuously orfor long periods or frequently.Zone 21—A place in which an explosive atmosphere in the form ofa cloud of combustible dust in the air is likely to occur in normaloperation occasionally.Zone 22—A place in which an explosive atmosphere in the formof a cloud of combustible dust in the air is not likely to occur innormal operation, but if it does occur, it will persist for a shortperiod only.

The severity should be considered the maximum (1) for all ofthe zones previously pointed out.

The safety experts agreed that the maximum severity (1) shouldbe expected in any explosion.

5.3. The use of the “Severity” membership functions

The use of the membership functions presented before is illus-

trated in this section.

Let us consider first the accident severity estimation due tofalling objects. In Fig. 8 (from Fig. 4), we can see that the expectedwork accident severity due to a brick 11 or similar object (in mass)

A. Pinto et al. / Accident Analysis and

0

0,2

0,4

0,6

0,8

1

0 2,5 4,5 6,5 8,5 10,5

12,5

Velocity (in km/h)

Seve

rity Skid S teer

Loa derHydrauli cexc avat or

Fm

ffh

tw1ev

6

wiblmctwt

apc

cFft

stdtw

ati

sfiooncefi

ig. 9. Examples of the expected work accident severity due to being struck by aoving vehicle traveling 10 km/h.

alling from a height of 10 m is 1 (corresponding to a fracture of arontal head bone). The severity is 0.58 if the falling object is a clawammer or similar.

Another example is shown in Fig. 9 (resulting from Fig. 3). Ifhe speed limit imposed at the site is 10 km/h, then the expectedork accident severity due to being struck by a moving vehicle is

(corresponding to a tibia fracture) if the vehicle is a hydraulicxcavator or a similar vehicle (in mass and design) and 0.8 if theehicle is a skid steer loader or a similar vehicle.

. Limitations and discussion

Empirical quantities represent measurable properties of the realorld. Any model is a simplified approximation of some underly-

ng causal structure. Adequate models are reasonable compromisesetween the credibility of results and the effort to create and popu-

ate the structure with adequate parameters (which in OSRA classicodels should be statistically measurable). However, despite the

onstruction industry ranking as one of the more hazardous inerms of frequency of deaths and fatality rates, the major causes oforkplace morbidity and mortality across the construction indus-

ry are fairly well-established (Lipscomb et al., 2010).Because of the lack of statistical data about accident severity

t construction sites and ill-defined models on biomechanics, theresent severity approach has some limitations that will be dis-ussed in this section.

In this work, we considered the maximum severity level (1) inlinically distinct situations, e.g., skull fracture and tibia fracture.rom empirical expert knowledge, there are cases in which a skullracture heals more rapidly than a tibia fracture (perhaps becausehe medical staff pay much more attention to critical cases).

From our literature survey, we found little data referring to theeverity of occupational accidents at construction sites, which washe focus of this work. Therefore, it was necessary to extrapolateata from other areas, namely, traffic accidents and child accidentso serve as surrogates for the adult male human body response,hich introduced model limitations that are difficult to determine.

Certain types of accidents (fires and explosions, for example)re hard to model because of the diversity of factors that may con-ribute to the severity and the virtual impossibility of knowing themn advance or even estimating them accurately.

Rescue time is an important severity predictor. All constructionites should have rescue plans for accident responses, includingrst aid–trained personnel and rescue equipment on-site, and allperatives should know and understand what to do. The qualityf medical intervention (preparation of medical staff, hospital fit-ess facilities) is also an important factor for accident severity but

annot be controlled by the site staff and was therefore not consid-red. Rescue time could be important in almost all accident modes,or example, bleeding control is one of the actions that can crit-cally influence the outcome severity. However, only in cave-ins

Prevention 45 (2012) 281– 290 289

and lost buoyancy in water was rescue time seen as of paramountimportance.

There are some other influencing predictors that are not consid-ered because, in practice, it is difficult to manage them in order toimprove safety, and some of them can even raise ethical questions.However, this constitutes a limitation of the proposed model. Thefollowing predictors were not considered in the modeling:

• The BMI score because, in practice, its management is difficultand raises ethical questions (it is not acceptable for social andethical reasons to choose workers by their BMI score).

• The impact angle because it is difficult to estimate it accurately (insome cases, it will determine if the applied force is compressiveor tensile; in other cases, it will determine the bending momentor the shear force in long bones).

• The movement kinematics because it is difficult to estimate itaccurately (e.g., in a high fall, it is difficult to estimate in advancethe area of the body that will be affected).

• The worker gender, because in Portugal, the working populationon construction sites is overwhelmingly male.

• The worker age because its management is difficult and raisesethical questions (for social and ethical reasons, it is not accept-able to prohibit people above a certain age from working).

• The worker physical fitness because for social and ethical reasonsit is not acceptable to choose workers based on their physicalfitness.

• The hour the accident happens. According to López et al. (2011),there is an incomprehensibly high rate of severe and fatal occu-pational accidents suffered by construction workers in Spainbetween 13:00 and 17:00. However, in practice, it is impossibleto accurately estimate the accident time.

In addition, the lack of statistical data relating to accidents in theconstruction industry, including information on predictors relatedto the severity, such as height of fall, velocity and mass of engineparts that cause entrapment, and soil density, made it impossibleto validate the proposed model by comparing the results estimatedby this model and information taken from actual accidents.

Regarding the choice of topology for the severity curves, i.e.,membership functions, we used linear open triangular functionsbecause of the lack of knowledge about the severity evolutionrelated to the increase of energy involved in work accidents. Wecould have used sigmoids for our fuzzification process, but withthese linear functions, it is easier to establish the bounds for ourphysical limits, which we believe is an advantage.

7. Conclusions

The goal of this work was to propose several severity functionsfor the construction industry (based on the safety risk assessment)to represent biomechanical knowledge with the aim of determin-ing the severity level of occupational accidents and consequentlyimproving the occupational risk assessment quality. We followed afuzzy approach because it allowed to capture and represents impre-cise knowledge in a simple and understandable way for users andspecialists.

This study is part of a larger study that involves the developmentand validation of a methodology for OSRA in the construction indus-try. For instance, protection measures are an important predictorof accident severity, and they are a dimension of our methodology.However, this analysis dimension is outside the scope of this article.

The work started from the existing knowledge about the factorsthat contribute to the severity of work accidents and the desireto model the possible severity evolution from a set of practicalpredictors that can be easily collected at construction sites based

2 s and

oeioswma

wvt

m(act

se

eowct

A

a

A

A

t

R

A

A

B

B

B

C

C

C

15, 291–303.

90 A. Pinto et al. / Accident Analysi

n objective (physical) criteria. The main problem was the lack ofxisting knowledge about this subject, namely, models of how bod-es fall and respective related biomechanical responses and modelsf how the quality of medical care can improve or worsen theeverity and historic data bases on accident severity. The presentork appears to have the potential to improve the risk assess-ent methodologies and will contribute to estimate the expected

ccident severity with a more (although coarse) objective criteria.In spite of the notable limitations (see Section 6), we believe

e have presented a more accurate model to estimate the possibleariation of work accident severity in terms of physical predictorshat can be easily “measured” at construction sites.

The main advantage of using the proposed fuzzification in ourodeling compared with other mathematical modeling techniques

Bayesian belief networks, for example) is the easy representationnd manipulation of empirical knowledge about ill-defined con-epts, such as is the case for the “severity” concept for the differentypes of accidents (in terms of occupational safety).

In future developments, it will be desirable to improve the fuzzyeverity functions by incorporating further empirical expert knowl-dge to reflect better the reality of work accident severity.

It will also be important to associate a confidence level toxpress the analyst’s uncertainty in the estimated results. More-ver, we also plan to define a global risk assessment level, whichill be included in a complete risk assessment methodology for the

onstruction industry. All of these subjects warrant future work inhis area.

cknowledgements

This work was funded by the Portuguese Foundation for Sciencend Technology, Scholarship No. SFRH/BD/39610/2007.

The authors are grateful to comments and suggestions bylexandre Escalhão Gomes as well as two reviewers.

ppendix A. Supplementary data

Supplementary data associated with this article can be found, inhe online version, at doi:10.1016/j.aap.2011.07.015.

eferences

le, B.M., Baksteen, H., Bellamy, L.J., Bloemhof, A., Goossens, L., Hale, A., Mud, M.L., Oh,J.H., Papazoglou, I.A., Post, J., Whiston, J.Y., 2008. Quantifying occupational risk:the development of an occupational risk model. Safety Science 46 (2), 176–185.

neziris, O.N., Papazoglou, I.A., Baksteen, H., Mud, M., Ale, B.J., Bellamy, L.J., Hale,A.R., Bloemhoff, A., Post, J., Oh, J., 2008. Quantified risk assessment for fall fromheight. Safety Science 46, 198–220.

allesteros, M., Dischinger, P., Langenberg, P., 2004. Pedestrian injuries and vehicletype in Maryland, 1995–1999. Accident Analysis and Prevention 36, 73–81.

SI – BS 8800, 2004. Guide to Occupational Health and Safety Management Systems.British Standard Institutions, UK.

arroso, M., Arezes, P., Costa, L., Miguel, A., 2005. Anthropometric study of Por-tuguese workers. International Journal of Industrial Ergonomics 35, 401–410.

haffin, B., Gunnar, A., Martin, B., 2006. Occupational Biomechanics, fourth edition.John Wiley & Sons, Hoboken, NJ.

ory, Z., Jones, D., 2006. Development of a simulation system for performing in situsurface tests to assess the potential severity of head impacts from alleged child-hood short falls. Forensic Science International 163, 102–114.

uny, X., Lejeune, M., 1999. Occupational risks and the value and modeling of ameasurement of severity. Safety Science 31, 213–229.

Prevention 45 (2012) 281– 290

Directive 99/92/EC of the of the European Parliament and of the Council of 16December 1999 on minimum requirements for improving the safety and healthprotection of workers potentially at risk from explosive atmospheres. OfficialJournal L 23, 28/01/2002 P. 0057–0064.

Directive, 2006/42/EC of the European Parliament and of the Council of 17 May 2006on machinery, and amending Directive 95/16/EC (recast). Official Journal L 157,09/06/2006 P. 0024–0086.

Dubois, D., Prade, H., 1998. An introduction to fuzzy systems. Clinica Chimica Acta270, 3–29.

Eppinger, R., Sun, E., Kuppa, S., Saul, 2000. Supplement: Development of ImprovedInjury Criteria for the Assessment of Advanced Automotive Restraint Systems –II. National Highway Traffic Safety Administration, http://www.nhtsa.gov/VRTC.

Faber, M.H., Stewart, M.G., 2003. Risk assessment for civil engineering facilities:critical overview and discussion. Reliability Engineering and System Safety 80,173–184.

Gennarelli, A., Wodzin, E., 2006. AIS 2005. A contemporary injury scale. Injury. Inter-national Journal of the Care of the Injured 37, 1083–1091.

Guidelines 2000 for Cardiopulmonary Resuscitation and Emergency CardiovascularCare, 2000. The American Heart Association in collaboration with the Interna-tional Liaison Committee on Resuscitation.

Hyoung, I., Kwon, Y., Kim, S., Kim, Y., Ju, T., Lee, H., 2009. The characteristics offatal occupational injuries in Korea’s construction industry, 1997–2004. SafetyScience 47, 1159–1162.

Jaslow, C.R., 1990. Mechanical properties of cranial sutures. Journal of Biomechanics23 (4), 313–321.

Jeong, B.Y., 1998. Occupational deaths and injuries in the construction industry.Applied Ergonomics 29 (5), 355–360.

Knudson, D.V., 2007. Fundamentals of Biomechanics, 2nd edition. Springer, SpringStreet, New York.

Lefler, D., Gabler, H., 2004. The fatality and injury risk of light truck impacts withpedestrians in the United States. Accident Analysis and Prevention 36, 295–304.

Lipscomb, H., Schoenfisch, A., Shishlov, K., 2010. Non-fatal contact injuries amongworkers in the construction industry treated in U.S. emergency departments,1998–2005. Journal of Safety Research 41, 191–195.

Li, H.X., Yen, V.C., 1995. Fuzzy Sets and Fuzzy Decision Making. CRC Press.Liu, J., Chen, X., Xu, L., 1999. New Thermal Wave Aspects on Burn Evaluation of Skin

Subjected to Instantaneous Heating. IEEE Transactions on Biomedical Engineer-ing 46 (4), 420–428.

López, M., Fontaneda, I., Alcántara, O., Ritzel, D., 2011. The special severity ofoccupational accidents in the afternoon: the lunch effect. Accident Analysis &Prevention 43 (3), 1104–1116.

McElhaney, H., Myers, S., 1993. Biomechanical aspects of thoracic trauma. In:Nahum, Melvin (Eds.), Accidental Injury. Springer-Verlag.

Mertz, H.J., 1993. Anthropomorphic test devices. In: Nahum, Melvin (Eds.), Acciden-tal Injury. Springer-Verlag.

Mizuno, K., Kajzer, J., 1999. Compatibility problems in frontal, side, single car colli-sions and car-to-pedestrian accidents in Japan. Accident Analysis and Prevention31, 381–391.

Müngen, U., Gürcanli, G.E., 2005. Fatal traffic accidents in the Turkish constructionindustry. Safety Science 43, 299–322.

Prasad, P., 1999. Biomechanical basis for injury criteria used in crashworthinessregulations. In: IRCUBI Conference, Sitges (Spain), September.

REN – National Energy Networks, 2011. Personal communication.Ross, T., 2004. Fuzzy Logic with Engineering Applications, 2nd edition. John Wiley

and Sons, Chichester, England.Sayed, T., Mota, A., Fraternali, F., Ortiz, M., 2008. Biomechanics of traumatic

brain injury. Computer Methods in Applied Mechanics and Engineering 197,4692–4701.

Suominen, P., Baillie, C., Korpela, R., Rautanen, S., Ranta, S., Olkkola, K.T., 2002. Impactof age, submersion time and water temperature on outcome in near-drowning.Resuscitation 52, 247–254.

Thorn, D.R., Hurt, H.J., Smith, T., 1998. Motorcycle helmet test headform and testapparatus comparison. In: 16th ESV Conference, Windsor, Canada, NHTSA.

Thompson, A., Bertocci, G., Pierce, C., 2009. Assessment of head injury risk associatedwith feet-first free falls in 12-month-old children using an anthropomorphic testdevice. Journal of Trauma-Injury Infection & Critical Care 66 (4), 1019–1029.

Tixier, J., Dusserre, G., Salvi, O., Gaston, D., 2002. Review of 62 risk analysis method-ologies of industrial plants. Journal of Loss Prevention in the Process Industries

Yang, J., 2002. Review of Injury Biomechanics in Car–Pedestrian Collisions (Report toEuropean Passive Safety Network). Crash Safety Division, Machine and VehicleSystems, Chalmers University of Technology, SE-412 96 Göteborg, Sweden.

Zadeh, L.A., 1965. Fuzzy sets. Information and Control 8, 338–353.