Practical approaches to QRA in fire Practical approaches ...

Practical approaches to QRA in

fire protection engineering

Piotr Tofilo, PhD

Practical approaches to QRA in fire

protection engineering

Piotr Tofilo PhD

The Main School of Fire Service + FirePlatform Ltd

CERN Workshop: An engineering perspective on risk assessment - November 26-27, 2018

Risk analysis in fire applications

Starting point: Fire regulations and standards

Performance Based Design (fire engineering)

Risk analysis: qualitative, semi QRA, full QRA

Probabilistic interpretation can be done in many ways

Risk categorization, risk metrics, consequences...

What if we have various types of losses ? (life, health, money,

environment, time, jobs, homes, cultural value, intellectual value,

public image…. )

How to include what we don’t know that we don’t know ?

(Grenfel, WTC… )

Timber structure apartment buildings

High rise single stair buildings

Protection of escape routes (optimization)

Sprinklered vs. non sprinklered buildings

Fire spread between buildings

Lightweight industrial buildings (cost / benefit)

Industrial problems: thermal radiation, explosion effects, toxic

releases...

Practical subjects - examples

Uncertainties to consider

Initial conditions (fire load, ditribution)

External conditions (wind, temperature, humidity etc.)

Fire (initiation, spread, heat and smoke generation)

Human effects (evacuation, intervention, errors, other)

Structural conditions (state of barriers, failures)

Systems (reliability, failures, effectivenes)

Fire event tree

Risk matrix (SFPE)

Fault tree + fire event tree

Probability of failure

Full QRA - challenges

Completeness of the problem studied

Adequacy of models (accuracy, limitations, integration)

Uncertainty of input data

Frequencies, distributions, materials, scenarios...

Multiple calculations, sampling, data processing

Meaningful results: F-N curves, risk matrix ?

Practicality: effort, time, cost, approval risk...

Monte Carlo analysis

Random variables

Sampling

Simple (crude) Monte Carlo (MCS)

Latin Hypercube (LHS) – stratification, inverse transforms

Importance sampling – rare events (black swans, tails)

Many other optimization techniques are available as well as

numerical packages (e.g. Python, C#, Java, R)

Adequate optimization necessary for models with high computation

cost (CFD)

Some promising approaches:

Response surface modeling (Qu 2003, Albrecht 2011)

ME-MDR Method (Van Coile 2017)

Confidence intervals

Fire QRA – selected software

FireCAM, FIERAsystem

(Canada)

CESARE-Risk

(Australia)

CRISP, BuildingQRA

(UK)

SAFETI

(Netherlands)

B-Risk

(New Zealand)

Probabilistic Fire Simulator (Finland)

FirePlatform – complex models / tools

FireRad FireRad QuickZone

FDS Designer Egress Designer FireFEM

FirePlatform – simple models / tools

Smoke control Cylindrical fire Detector activation

Total flooding 1D Heat Transfer Eurocode – parametric fire

EC Parametric Fire (MC mode)

Fire load density distribution (10k samples)

EC Parametric Fire (MC mode)

Temperature distribution (10k samples)

EC Parametric Fire (MC mode)

Temperature distribution (10k samples)

Discontinuity due to EC PF model split – FC / VC fires

AAMKS

Probabilistic fire and evacuation simulator

AAMKS - Fire modeling (CFAST)

AAMKS – Evacuation modeling

AAMKS - Results

Complementary cumulative density function (ccdf) – FN curves

Histograms with scenario counts for numbers of casualties

AAMKS – real life example

Change of use: 5 storey office to hotel

Length of the escape route exceeded (office 20 m, hotel 10 m).

No fire alarm system (FAS) (above 50 accommodation places)

No fire doors EI 30

Design alternatives

Alternative E. routes Ventilation Sprinklers Wall

EI 60

E. signs Alarming Training,auditing

1.1 10 m - - - - II -

1.2 20 m √ √ - - II -

2.1 20 m √ - - 5 lux II -

2.2 20 m - √ - 5 lux II -

2.3 20 m - - √ 5 lux I √

2.4 20 m - - - 5 lux II -

Event tree

FN curves (casualties vs. probability)

W 2.4

W 1.2 W 2.1

W 2.2 W 2.3

W 1.1

Risk matrix

Option Risk of fire death

1.1 1.16 * 10 -4 /year

1.2 2.57 * 10 -7 /year

2.1 1.16 * 10 -5 /year

2.2 2.37 * 10 -6 /year

2.3 1.02 * 10 -4 /year

2.4 1.21 * 10 -4 /year

Decision alternatives

Option E. routes Ventilation Sprinkler Wall

EI 60

Luminescence

Alarming Training, procedures

Economy

1.2 20 m √ √ - - II - $$$$

2.2 20 m - √ - 5 lux II - $$$

2.1 20 m √ - - 5 lux II - $$$

2.3 20 m - - √ 5 lux I √ $$

1.1 10 m - - - - II - $$$

2.4 20 m - - - 5 lux II - $

Summary

Using QRA tools is informative, educational and it can help

understand the problem in a probabilistic space

Using simple models with Monte Carlo is often sufficient in FPE or it

can be used for initial scoping analysis

For high risk applications it may be necessary to use more

advanced or interfaced models models to capture complexity

Alternatively extra conservative assumptions should be used

Next steps for fire QRA:

Fast computing of multiple scenarios

Use of complex modeling: CFD, Evacuation, FEM

More data is needed - physical, statistical

Adequate scenario sampling must be used or developed

Fire & Risk – Recommended Literature

piotr@fireplatform.eu | ptofilo@sgsp.edu.pl

THANK YOU