RISK ANALYSIS FOR SEVERE TRAFFIC ACCIDENTS IN ROAD TUNNELS (PART I)

IF CRASC ’15 III CONVEGNO DI INGEGNERIA FORENSE

VI CONVEGNO SU CROLLI, AFFIDABILITÀ STRUTTURALE, CONSOLIDAMENTO

SAPIENZA UNIVERSITA’ DI ROMA, 14-16 MAGGIO 2015 RISK ANALYSIS FOR SEVERE TRAFFIC ACCIDENTS IN ROAD TUNNELS (PART I) C. Di Santo Università degli Studi di Roma "La Sapienza" K. Gkoumas Università degli Studi di Roma "La Sapienza" F. Bontempi Università degli Studi di Roma "La Sapienza" ABSTRACT The safety in road tunnels is a very delicate issue, since that a minor accident or a failure of a vehicle can degenerate into scenarios that can lead to a high number of victims. For example, on the 24 March 1999, 39 people died when a Belgian HGV carrying flour and margarine caught fire in the Mont Blanc Tunnel. In the first part of this study has been summarized the operation logic of a specific model for the risk analysis, the PIARC/OECD Quantitative Risk Assessment Model, and how it derives risk indicators. In the second part, a comprehensive risk analysis is performed in a long tunnel in South Italy, accounting for multifaceted aspects and parameters. The analysis is integrated with a sensitivity analysis on specific parameters that have an influence on the risk. In sections 2, 3, and 4 the concept of Risk and its assessment is dealt. In section 5, the proce-dure followed by the QRA model to derive societal and individual risk indicators is discussed, starting from a given number of possible accident scenarios. In section 6 conclusions are written regarding the application of the studied model. 1. INTRODUCTION QRAM software, created by the cooperation of OECD, PIARC and European Commission is a tool whose purpose is to calculate the risk related to road traffic of heavy good vehicles. In fact, heavy vehicles circulation in case of accident, especially for dangerous goods, implies an additional risk to road users, for facilities, for the local population and the environment. Therefore, through Quantitative Risk Analysis, the competent authorities may assess whether to allow the transition of all types of goods through a given path or, simply, through a given gallery. To help the authorities in this choice, with particular attention to the high extension galleries, the PIARC (The World Road Association) and the OECD (The Organization for

C. Di Santo, K. Gkoumas, F. Bontempi

Economic Cooperation and Development) have developed this model using risk assessment methodologies used in the past in the chemical and nuclear industries. Accidents in this field are rare, but can have major consequences. 2. THE CONCEPT OF RISK In general the risk is given by 𝑅𝑅 = 𝐹𝐹 ∙ 𝐶𝐶 (1) where F is probability of occurrence (frequency) of a given event, and C is the relative con-sequence which can be measured in number of fatalities, injuries, cost of the structure damage caused by the event and damage to the environment. In this study, consequence C refers to the number of victims N. Furthermore, the risk can be Social or Individual. The Societal Risk (SR) can be defined as the risk to which it is subjected a group of people in case occurs a scenario s and it is given by 𝑆𝑆𝑅𝑅 = 𝐹𝐹(𝑁𝑁) ∙ 𝑁𝑁 (2) where F(N) is the frequency [1/year] of a event that causes a number of victims ≥ N (i.e. cumulative frequency), and N the number of fatalities [Fat]. The SR is represented through F – N diagrams and Expected Values of victims (EVs). The F-N diagrams may be applied to illustrate the risk profile for a specific hazard such as a fire in a road tunnel. On the x-axis are shown in logarithmic scale, the number of victims N, and on the ordinate axis, still on a logarithmic scale, the corresponding annual rate F(N) with which events occur that cause a number of victims ≥ N . The expected value (EV), which is the expected amount of victims in a certain time period, can be calculated as the area under the F-N curve by using the following equation (Petelin S. 2009): 𝐸𝐸𝐸𝐸𝑠𝑠 = ∑ 𝐹𝐹(𝑁𝑁𝑖𝑖) ∙ 𝑁𝑁𝑖𝑖∞

𝑖𝑖=0 = ∫ 𝐹𝐹(𝑁𝑁)+∞1 𝑑𝑑𝑁𝑁 (3)

where EVs is the expected value of victims in one year caused by the scenario s, and F(Ni) the cumulative frequency of an accident that causes Ni fatalities [1/year]. The Individual Risk can be expressed as an annual frequency [1/year] or as the time interval between two accidents that cause the same fatalities among the local population because of an accident on the road network. Considering an accident scenario s, it indicates the proba-bility (in 1 year) that an individual, located at some distance from the road network, dies. The number of people exposed to the incident does not affect the value of the risk, thus, for this reason it is called "individual". It can be represented as the risk distribution in the space sur-rounding the road section through a 2D map.

Risk analysis for severe traffic accidents in road tunnels (Part I)

Figure 1. Example of a F-N curve. 3. GENERAL RISK ASSESSMENT PROCEDURE The risk assessment procedure consists of different steps. In a first phase data need to be collected. In case of a road route analysis, traffic data, accident frequencies, meteorological conditions, tunnel equipment (in presence of a tunnel), etc., are prepared. Next, there is the Risk Analysis phase, in which the risk indicators, through a quantitative model, such as the QRAM, are calculated. In the final step, it must be assessed whether the risk indicator value obtained with the analysis is acceptable or not. If the level of risk is not acceptable, it is necessary to provide for measures to mitigate the risk, which can be of prevention if they reduce the frequency of occurrence of accidents, or of protection if they act on the protection provided to users once a scenario has occurred.

Data Collection

Data Preparation

Risk Calculation

Using QRAM

Is Risk acceptable?

NOAdditional

risk reduction measures

START YES End

Figure 2. Typical risk assessment procedure using PIARC/OECD QRA Model. In literature, there are several methods to evaluate the risk level, some of which are based on the "absolute criterion" of risk acceptability, others on the "relative criterion". In this study, the absolute criterion is used. The absolute criterion allows to identify the minimum level safety that must be guaranteed for road users.

1.00E-06

1.00E-05

1.00E-04

1.00E-03

1.00E-02

1.00E-01

1.00E+00

1.00 10.00 100.00 1000.00

F CUM

[acc

/yea

r]

N [FAT]

Tollerable Risk LineAcceptable Risk Line

ALARP

Not Acceptable Area

Acceptable Area


About the F-N curve, a method known in the literature is the As Low As Reasonably Practi-cable (ALARP) principle, where the F-N plane is divided into three zones (Figure 1): Not Acceptable Risk Area, Acceptable Risk Area and ALARP Area (prevention and/or mitigation actions must be taken to reduce the risk, as far as reasonably practicable). In fact, if the curve is in the middle between the two lines, then measures for risk reduction should be taken, but the operations costs must not be disproportionate to the benefits obtained in risk reduction. So, a cost – benefit analysis is needed. 4. THE COMBINATION OF QRAM MODEL AND SCENARIO BASED ANALYSIS The analysis carried out by the QRAM is at the macroscopic level. In fact it is possible to set some of the parameters that characterize the various scenarios. To describe the evolution of the physical phenomenon, the authors of QRAM have used simplified models whose results (obtained for differents possible evolutions of the phenomenon) were included in contin-gency tables. Then the QRAM during the analysis uses data provided by the final user, to-gether with those that it picks up from the tables, to calculate risk indicators.

User's Inputs QRAM

AnalysisExternal Models Results

Quantitative Risk Estimation

Figure 3. QRAM analysis. So, it is difficult to evaluate the effects of risk mitigation measures on the specific scenario, because is not possible to evaluate changes that occur in the evolution of the physical phe-nomenon. For example, in case of fire in a tunnel, it is complicated to know how the flow of gasses changes if it is decided to use a transverse ventilation system characterized by dynamic dampers for the fumes extraction. For example, in order to evacuate toxic gases, it is impos-sible to understand if opening a damper, instead than another, could modify the evolution of the whole scenario and in which way. Therefore, it is convenient to combine the QRA Model with CFD simulations (Computational Fluid Dynamics) and evacuation models. An operating method to follow can be to identify the critical scenarios that give the most significant contribution to the overall risk through the QRAM, and then to simulate those scenarios in detail in order to define risk reduction measures (Petelin S. 2009). An example of application of CFD simulations is related in the paper “Computational flu-id dynamics simulations for the assessment of a road tunnel fire safety” (Baroncelli et al., Proceedings IF CRASC’ 15), while the possibility of using evacu-ation models fully coupled with fire simulations is described in the paper “Influence of panic on human behavior during emergency egress for tunnel fires” (Gai et al., Proceedings IF CRASC’ 15).


Data Collection

Data Preparation

Risk CalculationUsing QRAM

Is Risk acceptable?

NO

Additional risk reduction

measures

START

YES

End

Idintification of Critical Scenarios

Single Scenario Simulation

CFD Simulation(Fire, Ventilation)

Evacuation Model(Evacuation, Rescue)

Qualitative Risk Estimation

Measures Included

in the model?YES NO

Figure 4. Risk assessment approach that combines quantitative risk assessment and qualitative critical scenario analysis.


5. THE GENERAL FUNCTIONING OF THE QRA MODEL The procedure followed by the model to calculate SR indicators (i.e. F-N curves and EVs) is represented in Figure 5.

1 – Dangerous Goods and Accident Scenarios selection

2 – Effect j (due to the scenario “s”) and its Range from the epicentre: Ej=f(d) ↔ Rj

3 – Mortality Rate within the range Rj → %LETHj

4 – Mortality Rate corrected considering the possibility of escape → %LETHj=f(tevac)

5 – Scenario s Probability of occurrence → fs

6 – Number of victims due to the scenario s → N=∑jNj=f(Rj, Dru, Ljam, %LETHj)

7 – F-N curve for the scenario “s” and its relative Expected Value

Figure 5. QRA model functioning.

Firstly, the model considers certain Dangerous Goods (DGs): • LPG (Liquefied Petroleum Gas), flammable mixtures of hydrocarbon gasses, but in this

model reference is made to propane; • Motor Spirit: a mixture of volatile and flammable liquid hydrocarbons; • Acrolein (C3H4O): a toxic liquid; • Chlorine (Cl2) and Ammonia (NH3): toxic gases; • Liquefied CO2 under pressure. Each DG can cause different scenarios depending on the vessel type and on how the release of the substance takes place (Figure 6). For each scenario, up to three types of physical effects on road users and local population are considered by the model:


Figure 6. Accidents Scenarios in QRAM software.

• Thermal Effects (thermal radiation produced by the flames): measured through the Radi-ative Heat Flux, qr [kW/m2]. This value depends on the distance of the target from the flames, on the Surface Emissive Power (SEP) [kW /m2] (maximum amount of heat flux near the flames surface), on the orientation of the target with respect to the flames and on the atmospheric transmissivity.

• Pressure Effects (the shock wave generated by the explosion): caused by the pressure wave generated by an explosion, which propagates outward from the epicenter. When the wave front arrives at a certain point of the space, the overpressure increases instantly from zero to its peak value and immediately after the overpressure decreases. So, the overpres-sure drops to zero in a short time, and this marks the end of the positive phase. The phys-ical parameters used to represent the effects of pressure on people are the peak overpres-sure ∆Ps [bar], and the positive phase duration t+[bar].

• Toxic Effects (poisoning due to the release of toxic substances): consist in poisoning of users for exposure to a particular toxic substance. To define the level of poisoning, to which is subject an individual, are used the concentration in air of the toxic substance in question and the time of exposure to that particular concentration, C [ppmv].

With the aid of specific models, the QRAM establishes the relationship between the single-effect and the distance from the epicenter of the incident. At this point, through the probit analysis, the model derives the relationship between the mortality rate (probability of dying occuping a certain position in relation to the incident) and the distance from the center of the incident (Figure 7). The probit analysis is a type of regression used to analysing the relationsheep between a stimulus (dose) and “all or nothing” (such as death) response. This method was proposed by Finney (Finney 1971). Exposing a biological organisms population to a number of different doses of a toxic substance he plotted a Gaussian curve for each experience and the complete dose-response curve using the cumulative mean response at each dose. For convenience, he

Dangerous Good Mode of Containment Diameter release hole Mass Flow Rate Scenario n°- - - 20 MW HGV Fire 1

- - - 100 MW HGV Fire 2

Cylinder (50 kg) - - BLEVE 3

- - BLEVE 7VCE 8

Torch Fire 9Pool Fire (≥400MW Fire) 4

VCE 5

Bulk (30000 liters) 50 mm 24.8 kg/s Toxic Liquid Release 11

Cylinder (100 liters) 4 mm 0.02 kg/s Toxic Liquid Release 12

Chlorine (toxic gas) Bulk (20 t) 50 mm 45 kg/s Toxic Gas Release 6

Ammonia (toxic gas) Bulk (20 t) 50 mm 36 kg/s Toxic Gas Release 10

Liquified CO2 Bulk (20 t) - - BLEVE 13

Acrolein (toxic liquid)

Bulk (18 t)50 mm 36 kg/s

Bulk (18 t) 100 mm 20.6 kg/s

No DG

Liquefied Petroleum Gas (LPG)

Motor Spirit


Figure 7. 100MW “Fire Effect Intensity – Distance” and “Lethality percentage – Distance relationsheep”.

plotted the response versus the logarithm of the dose, because this form provides a much straighter line in the middle of the curve. In general, these curves can be drawn for different exposures, including heat, pressure, radiation, and toxic gases. However, for computational purposes, the response - dose curve is not convenient, so Finney developed a method to pro-vide a straight-line equivalent to the response-dose curve. He established a relationship be-tween the probability P that an individual dies under a given dose (i.e. lethality percentage) and the PROBIT variable Pr (PROBability unIT): P �or RATIO� = 1

√2𝜋𝜋∫ 𝑒𝑒−

12𝑢𝑢

2𝑑𝑑𝑑𝑑Pr−5

−∞ (4) Plotting the Pr – log (dose) curve using a linear probit scale, therefore, he transformed the sigmoid shape of the normal response versus dose curve into a straight line. In QRA model, the probit variable Pr is computed from: 𝑃𝑃𝑃𝑃 = 𝑎𝑎 + 𝑏𝑏 ln(𝐷𝐷) (5) Where D is the load related to the single effect (Dose), a and b are probit parameters, the values of which depend on the effect considered. In the calculation of the mortality rate for increasing distances from the point of the accident, the model also considers the possibility of escape or of find shelter, calculating the total dose of a given effect that affects a subject during the escape. This is done by calculating the ef-fective dose Dj that affects a road user: 𝐷𝐷𝑗𝑗 = ∫ �𝐸𝐸𝑗𝑗 ∙ 𝑡𝑡𝑒𝑒𝑒𝑒𝑒𝑒,𝑗𝑗�𝑑𝑑𝑡𝑡

𝑡𝑡̅0 (6)

where 𝑡𝑡̅ is the minimum time between the duration of the scenario ts and the time necessary for the user to evacuate the area tevac.


Figure 8. Response versus log dose curve (left), the relationship between percentages and probits (centre) and the transformation that converts the sigmoidal response versus log dose curve into a

straight line (right), (Crowl D. 2011). After, the model derives the frequency of occurrence of each scenario once an accident, involving HGV without dangerous goods (20MW and 100MW Fires) or DG-HGV, has oc-curred. For example, for scenarios involving DG-HGV: 𝑓𝑓𝑖𝑖𝑗𝑗𝑖𝑖 = 𝑃𝑃𝑖𝑖𝑗𝑗𝑖𝑖 ∙ 𝑓𝑓𝑎𝑎𝑎𝑎𝑎𝑎_𝐷𝐷𝐷𝐷,𝑖𝑖 ∙ (𝑇𝑇𝐷𝐷𝑖𝑖𝑖𝑖 ∙ 𝐿𝐿𝑖𝑖 ∙ 24 ∙ 365 ∙ 10−6) (7) Where fijk is the frequency of occurrence of the scenario j (involving DG of type k) on the section i in a year, Pijk is the conditional probability that occurs scenario j once the accident has taken place, facc_DG,i is the annual frequency of accidents involving DG-HGVs [acc/(MVkm*year)] and TDik is the traffic of DG-HGVs of type k passing through the section i in one hour. As regards the Pij values, they are automatically fetched, during the calculation, from precompiled tables within the software. In fact, these values were previously calculated by the analysis of the Fault Tree. At this point, the model calculates the number of victims following a scenario s, considering as the main parameter the users density per linear meter of the road section. This density has a different value depending on whether you are considering a portion occupied by a traffic jam, or a portion characterized by fluid traffic. For example, the number of victims among road users in a traffic jam is given by the following expression: 𝑁𝑁 = �𝑅𝑅 ∙ 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅� ∙ %𝐿𝐿𝐸𝐸𝑇𝑇𝐿𝐿 (8) Where Ljam is the traffic jam length [m], DRUJ is the Road Users Density in a traffic Jam (number of people in HGVs + cars on this section in direction A for each meter of a jam) [users/m], and %LETH is the Lethality Percentage [-]. Finally, the model considers that each scenario s may appear as a different event Ei, varying some parameters as the section of the path being considered, the accident location on the section, the traffic direction, the reference period of the day (peak hour traffic), etc. Thus, QRAM computes a different number of victims Ni for each one of these events Ei, and to each one of them associates a probability of occurrence fi. From these frequencies, he calcu-lates the cumulative frequencies Fi. Therefore, associating with each possible number of vic-tims (Ni) the corresponding cumulative frequency (Fi), it is possible draw the risk curve on the plane F-N (Figure 10).


Figure 9. Number of Fatalities Calculation.

Scenario "s"

Event Event Frequency Fatalities Cumulative Frequency

Ei fi Ni Fi

[-] [1/year] [Fat] [1/year]

E1 f1 N1 F1 = f1

E2 f2 N2 F2 = f1+f2

E3 f3 N3 F3 = f1+f2+f3

E4 f4 N4 F4 = f1+f2+f3+f4

... ... ... ...

En fn Nn Fn = f1+f2+f3+f4+...+fn

Figure 10. F-N construction for a generic scenario s.


6. CONCLUSIONS This model allows to make a risk assessment in a road route along which there can be one or more tunnels. It can also assess the intrinsic risk of a tunnel extrapolated from the route where it is inserted. In both cases, the risk indicators are calculated considering the main features: • of the road route (total length, etc.), • of the platform road (number of lanes, etc.) • of the tunnel (cross-sectional area, etc.), • of its safety equipment (system ventilation, drainage system, by pass, etc.) and • of the type of traffic (total traffic, heavy vehicles, dangerous goods, etc.). Therefore, the model, during the analysis, considers different aspects of road transport and up to 13 possible accident scenarios. So, the risk assessment that results is at macroscopic level, i.e. the model refers to the entire transport system as a whole. ACKNOWLEDGEMENTS The authors thank: Giordana Gai (Student of PhD in Structural Engineering from the Uni-versity of Rome "La Sapienza"), Francesco Petrini (Researcher in Structural Engineering from the University of Rome "La Sapienza"), Tiziano Baroncelli (University of Rome "La Sapienza"), Eng. Luigi Carrarini (ANAS S.p.A.), Eng. Alessandra Lo Cane (M.I.T). This work is partially supported by the spin-off company StroNGER s.r.l. which is gratefully acknowledged. REFERENCES Crowl D., Louvar J.: Chemical Process Safety, Fundamentals with Applications. III, Prentice Hall, 2011. Finney, D.J.: Probit Analysis, A statistical treatment of the sigmoid response curve. Cambridge University Press, 1971. Petelin S., Luin B., Vidmar P.: Risk Analysis Methodology for Road Tunnels and Alternative Routes. Journal of Mechanical Engineering, 2009, pp. 41-51. PIARC, OECD: Safety in Tunnels, Transport of dangerous goods through road tunnels. OECD Publications, 2001.