A discrete geometry model of fire propagation in urban areas L-1 - - PowerPoint PPT Presentation

a discrete geometry model of fire propagation in urban
SMART_READER_LITE
LIVE PREVIEW

A discrete geometry model of fire propagation in urban areas L-1 - - PowerPoint PPT Presentation

A discrete geometry model of fire propagation in urban areas L-1 norms and fire propagation Stphane Gaubert and Daniel Jones Stephane.Gaubert@inria.fr daniel.jones@inria.fr Inria, cole Polytechnique November 14, 2017 L = Z 2 ... the


slide-1
SLIDE 1

A discrete geometry model of fire propagation in urban areas

L-1 norms and fire propagation Stéphane Gaubert and Daniel Jones Stephane.Gaubert@inria.fr daniel.jones@inria.fr

Inria, École Polytechnique

November 14, 2017

slide-2
SLIDE 2

◮ L = Z2... the integer lattice ◮ Ak... the set of points in x ∈ L to which we can propagate in

at most k time steps, k ≥ 1 (and homogeneity of the system)

◮ τk (x) =

  • propagation time to x in at most k steps, if x ∈ Ak

+∞, o.w.

◮ v (x) = τ ∗ (x) = limk→∞ τk (x)... propagation time to x ◮

fk ≡ co (τk) : dom (co (τk)) ∪ R2 → R fk

  • i∈I

λixi

  • =
  • i∈I

λiτk (xi) , where (∀i ∈ I) xi ∈ Ak and

i∈I λixi is a convex combination

slide-3
SLIDE 3

Lemma

Let x ∈ dom (fk) , then fk (x) = kf x k

  • .

Lemma

The lower boundary of k × epi (f ) is equal to fk, where k × epi (f ) denotes the kth Minkowski sum of epi (f ).

Lemma

k × epi (x → f (x)) = epi

  • x → kf

x k

slide-4
SLIDE 4

Lemma

kf x k

  • ≤ τk (x) ≤ kf

x k

  • + constant.

It follows... inf

k≥1 kf

x k

  • ≤ v ≤ inf

k≥1

  • kf

x k

  • + constant
  • Lemma

inf

k≥1 kf

x k

  • = f ′ (0; x) =

sup

p∈∂f (0)

< p, x >

slide-5
SLIDE 5

Fire propagation is a polyhedral norm...

Theorem

lim

s→∞

v (sx) s = sup

p∈δf (0)

< p, x > . The long-term geometry of the fire front depends simply on the immediate propagation directions, A1 (since Ak are Minkowski sums of A1)

slide-6
SLIDE 6

Important example, the Von-Neumann neighbourhood... A = {(1, 0) , (0, 1) , (−1, 0) , (0, −1)} , with corresponding times τ1, τ2, τ3, τ4 respectively

◮ Radiative heating between large surface areas ◮ The polyhedral norm is a deformed L1 ball

slide-7
SLIDE 7

L1 balls

Figure: L1 norm

slide-8
SLIDE 8

◮ Q... set of discrete states ◮ Q = {0, 1} (ignited or not)... purely geometric ◮ For a simulation, we use the states used in the paper of Zhao

slide-9
SLIDE 9

◮ 0 - original state (white) ◮ 1 - ignition ◮ 2 - flashover (self-developing) ◮ 3 - full development ◮ 4 - collapse ◮ 5 - extinguished

Show video 1

slide-10
SLIDE 10

Non-perfect lattices and extra factors (wind or changing urban geometries)... deformed L1 balls

Figure: L1 balls

slide-11
SLIDE 11

◮ Multiple sources... union of deformed L1 balls (show video) ◮ Changing geometry across the urban environment... Finsler

geometry

slide-12
SLIDE 12

Figure: octagonal polyhedral norm

◮ change of wind on third day ◮ octagonal geometry on extreme edge of fire

slide-13
SLIDE 13

Figure: ’dips’ after crossing low density areas

◮ small ’dips’ in the expected straight edges of the L1 ball ◮ modelled by a so-called ’Finsler-geometry’ ◮ ’cell densities’ can be incorporated into the model

slide-14
SLIDE 14

future work

◮ 3D models for non-flat cities ◮ It may be possible to reverse engineer to find the ignition point ◮ Rome 64 AD, emperor Nero ◮ Different deformation of L1 ball in different parts of the city ◮ Analytic formulas for propagation speeds to specific buildings ◮ Easier to interpret and modify than the applied model of Zhao ◮ Incorporate stochasticity (randomness) into the model (implies

rounder corners of the fire front)

slide-15
SLIDE 15

Thank you for your attention!

slide-16
SLIDE 16

Figure: deformed L1 ball with boundaries