by - ifv · 2016-04-08 · c) 0 20 40 60 80 100 120 assimilated data data observed after...

19
1 Forecasting Enclosure Fires By Ghent University – UGent, Dept. Flow, Heat and Combustion Mechanics, Belgium Dr Tarek Beji

Upload: others

Post on 04-Aug-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

1

Forecasting Enclosure Fires

By

Ghent University – UGent, Dept. Flow, Heat and Combustion Mechanics, Belgium

Dr Tarek Beji

Page 2: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

2

Outline

1. Introduction: The Numerical Fire Forecast (NFF) concept

2. Video fire monitoring

3. Fire Modelling (two-zone modelling)

4. Data Assimilation (DA) and Inverse Fire Modelling (IFM)

5. Applications of the Fire Forecasting concept

6. Conclusions and Future work

Page 3: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

3

Inspired from the concept of Numerical Weather Predictions (NWP)

‘Forecast’ <==> Real-time predictions (as opposed to ‘design calculations’)

Objectives

•Provide guidance during fire fighting interventions

•Assist a decision support system

Methodology

Sensor readings

(e.g. thermocouples, video camera)

Fire Model

(Estimation of model invariants)

Display of future hazards

(e.g. smoke, flashover)

Data Assimilation (DA) Forecast

Previous study: The ‘FireGrid’ project at the University of Edinburgh (CFD

+ HPC resources applied to the ‘Dalmarnock’ fire tests)

1. Introduction: the concept of fire forecasting

Page 4: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

4

2. Sensor readings

Page 5: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

5

2. Sensor readings

Collaboration with the multimedia department on the video fire analysis

2.4 m

0.4 m

h XN

ρu, Tucamera

3.3 m

2.4 m

0.4 m

h XN

ρu, Tucamera

3.3 m

1) Smoke detection algorithm based on Energy Disorder (ED)

2) Flame detection algorithm

Page 6: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

6

time (s)

0 100 200 300 400 500

Smok

e he

ight

(m

)

0,0

0,5

1,0

1,5

2,0

2,5Thermocouple dataVideo data

time (s)

0 200 400 600 800 1000

Smok

e he

ight

(m

)

0,0

0,5

1,0

1,5

2,0

2,5Thermocouple dataVideo data

Smoke detection algorithm

Smoke height estimation of a sofa fire using energy disorder

Case 1

Case 2

Page 7: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

7

Flame detection algorithm

time (s)

0 100 200 300 400

Flam

e w

idth

, W

f (c

m)

0

25

50

75

100

125

150

time (s)

0 100 200 300 400

Flam

e he

ight

, L f

(cm

)

0

25

50

75

100

125

150

2/5

235.0

02.1⎟⎟⎠

⎞⎜⎜⎝

⎛ += ff WL

Q

time (s)

0 100 200 300 400

Conv

ecti

ve H

RR

, Q

c (k

W)

0

100

200

300

400

500

To be compared with oxygen calorimetry tests

Objective: Estimate the HRR from flame dimensions

Page 8: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

8

3. Fire Modelling

Page 9: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

9

3. Fire Modelling: two-zone modelling

( )[ ] ( ) ( )aupuoutfcuaupu TTcmQVTTcdtd

−−−=− ,1 λρ

( ) uoutuinpuu mmmVdtd

,, −+=ρ

Mass conservation for upper layer

Energy conservation for upper layer

ρu, Tu

pm

uoutm ,

uinm ,

Upper layer

Sub-models

• Plume entrainment model (e.g. Zukoski, McCaffrey, Heskestad)

• Model for outflow of gases through opening (derived from Bernoulli’s equation and hydrostatic principles)

Page 10: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

10

Fire Modelling: why two-zone modelling?

===> very fast calculations using 2-zone modelling

Fire Forecast = Prediction of the fire development

ahead of the event

positive lead time

Disadvantage

“loss of physics” compensated by assimilating observed data into the model.

Page 11: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

11

4. Inverse Fire Modelling

Page 12: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

12

4. Inverse Fire Modelling

Design calculations Fire Forecasting

Input:

• Geometry

•Heat Release Rate, Qf

• Heat loss factor, λc

Output (tenability conditions):

• Smoke layer height, h,

• Temperature (Tu) …etc

Input:

• Geometry

•Smoke layer height over a given period of time (i.e. assimilation window)

•Temperature

Estimation of the trend in Qf

Estimation of the heat loss factor

Forecast of h and Tu

Data assimilation Inverse modelling

Vs

Page 13: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

13

Inverse Fire Modelling

Optimization problem

Sensor readings <==>

‘Observed state’

Fire Model

Model Invariants, MI:

1. Fire growth factor

2. Heat loss factor…

Inverse Problem: What values for the MI provide the best match with the

‘observed state’ of the system provided by the sensor readings?

MATCH ?

Optimization techniques:

1. Gradient-based methods (e.g. Tangent linear technique) : Fast convergence but Initial guess need to be close enough to the solution

2. Gradient-free methods (e.g. Genetic Algorithms): Convergence is ensured but a longer calculation time is required

Page 14: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

14

5. Applications

Page 15: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

15

5. Application 1: Smoke Filling in an atrium

22.4 m 11.9 m

27 m

C

burner 2 m x 2 m

Li et al. Journal of China University of Science and Technology 29 (1999) 590-4

-Atrium dimensions = 22.4 m x 11.9 m x 27 m (Length x Width x height)

-Constant fire size 4000 kW

-Two fire locations (centre and corner)

Data Assimilation and Inverse Fire Modelling

-Model Invariants Fire size, Qc, and entrainment coeff. C

-Constants λc= 0.6

-Data Assimilated Smoke layer height, hi, and temperature, Tu.

-Number of observations N=3 (each 20 s)

<==> Assimilation window 60 s

Page 16: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

16

time (s)

0 50 100 150 200 250 300

h (m

)

0

5

10

15

20

25

30Assimilated dataData observed after assimilationForecast

time (s)

0 50 100 150 200 250 300T u

(0 C)

0

20

40

60

80

100

120

Assimilated dataData observed after assimilationForecastCASE Num. 2:

Fire at the corner

time (s)

0 50 100 150 200 250 300

h (m

)

0

5

10

15

20

25

30

Assimilated dataData observed after assimilationForecast

time (s)

0 50 100 150 200 250 300

T u (0 C

)

0

20

40

60

80

100

120

Assimilated dataData observed after assimilationForecastCASE Num. 1:

Fire at the centre

Forecast Forecast

Forecast Forecast

DA DA

DA DA

Estimated parameters

• Qc= 3303 kW

• C = 0.260

Estimated parameters

• Qc= 2656 kW

• C = 0.109

Modifying the value of λc gives different Qc and C but the same forecast.

5. Application 1: Smoke Filling in an atrium

Page 17: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

17

5. Application 2: Furniture fire in an ISO-room

Flashover Fire in an ISO-Room (Obs.: h and Tu)

0

0.5

1

1.5

2

2.5

0 100 200 300 400 500

time (s)

Hei

ght (

m)

X_N (Exp.)X_N (Pred.)h (Exp.)h (Pred.)

0

100

200

300

400

500

600

700

800

900

0 100 200 300 400 500

time (s)

Upp

er la

yer t

empe

ratu

re, T

u (C

)

exp. dataPred.

DA1 DA2

• Transient fire

• DDDAS

Forecast 1 Forecast 2

Page 18: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

18

6. Conclusions and Future Work

Page 19: By - IFV · 2016-04-08 · C) 0 20 40 60 80 100 120 Assimilated data Data observed after assimilation CASE Num. 1: Forecast Fire at the centre Forecast Forecast Forecast Forecast

19

Conclusions

• Presentation of the fire forecasting system

• Advances in video fire monitoring

• Aspects involved in Inverse Modelling and Fire forecasting

• Applications: Smoke filling in an atrium and furniture fire in an ISO-room

Future work

• More validation studies on video fire monitoring

• Application of the concept to multi-compartments