NH3 Dispersion modelling - Chemical Engineering Facultychemeng.technion.ac.il/wp-content/uploads/2017/11/Jacques-MOUS… · 𝐹 =2 − + Type I Type II Type III. Factor of two (FAC2)

Computational Fluid Dynamics (CFD) and Stochastic Lagrangian Particle

Dispersion Models (LPDM) applied to the modeling of transport and dispersion

of accidental or malevolent releases of ammonia to the atmosphere.

Jacques MOUSSAFIR and Armand ALBERGELPresident & CEO

New and Old in the Ammonia World 2017Technion, Haifa, Israel, 15-16 November 2017

[email protected]

ARIA Technologies

ARIA Technologies SA8-10, rue de la Ferme – 92100 Boulogne Billancourt – France

Telephone: +33 (0)1 46 08 68 60 – Fax: +33 (0)1 41 41 93 17 E-mail: [email protected] – http://www.aria.fr

mailto:[email protected]

o Why use 3D models of atmospheric dispersion ?

o The COST ES1006 project: classifying ATD model Types.

o CFD example: the FLADIS experiment.

o The effect of obstacles: Jack Rabbit II example.

o Effects of terrain and buoyancy: the Haifa tank simulation

o Conclusions.

Presentation outline

Atmospheric Transport and Dispersion (ATD) models are used to simulate together the flow (micro-meteorology) and the transport/dispersion (cloud spatial distribution) of substances released to the atmosphere in the case of an accident.

ATD models are three-dimensional (3D) if they provide a description of the flow field (wind, temperature, turbulence) and of the concentration fields that are not horizontally and/or vertically homogeneous.

3D models of ATD are useful to represent the combined effects of:

• Complex terrain (inducing complex micro meteorological flow patterns)

• Obstacles (such as tanks or buildings) which can lead to:

o Enhanced initial dispersion (decreasing concentrations at a given distance)

o Increased channeling effects (in streets, between industrial buildings)

• Buoyancy and stability (detailed plume rise description, buoyancy driven flow)

• Detailed energy exchange processes in multi-phase flows, etc…

Why use 3D models for ATD?


o The COST ES1006 project: classifying ATD models.




o Conclusions.


CO

ST A

cti

on

ES

10

06

What is a COST Action?

intergovernmental framework for European

COoperation in Science and Technology

supports capacity building by connecting scientific communities

provides networking opportunities

connecting research with stakeholders

source: ABC news

source: EWTL - UHH

source: ARIANET

CO

ST A

cti

on

ES

10

06

COST ES1006

COST contributors (partial list) :B. LEITL, University of Hamburg, Germany F. HARMS, University of Hamburg, Germany

S. TRINI-CASTELLI, Consiglio Nazionale delle Ricerche, Italy K. BAUMANN-STANZER, ZAMG, Austria

S. HERRING, DSTL, UK P. ARMAND, CEA, France

G. TINARELLI, ARIANET SRL, Italy J. MOUSSAFIR, ARIA Technologies, France

M. NIBART, ARIA Technologies, France S. ANDRONOPOULOS, Demokritos Research Center, Greece

T. REISIN, SOREQ Research Center, Israël J-M. LACOME, INERIS, France

C. GARIAZZO, INAIL, Italy R. TAVARES, ECN, France

E. BERBEKAR, University of Hamburg, Hungary G. EFTHIMIOU, Demokritos Research Center, Greece

V. FUKA, Institute of Thermodynamics, Czech Republic G. GASPARAC, Gekom d.o.o. , Croatia

A. HELLSTEN, Finnish Meteorological Institute, Finland K. JURCACOVA, Institute of Thermodynamics, Czech Republic

A.PETROV, National Institute of Meteorology & Hydrology, Bulgaria A. RAKAI, Budapest University, Hungary

S. STENZEL, ZAMG, Austria

CO

ST A

cti

on

ES

10

06

COST ES1006

Happy contributors:

CO

ST A

cti

on

ES

10

06

Goals of COST ES1006

establishing consensus on the 'state-of-the-art' in local (micro) scale airborne hazards modelling

focusing on urban cases (buildings)

providing common means, tools and data for rigorously testing and evaluating models

providing guidance for reliable use of models in the context of local-scale emergency response

develop and test strategies and methodologies for new advanced modelling approaches

CO

ST A

cti

on

ES

10

06

As in many of the other COST Actions, the activity ends up with the production of documents……..

• Background and Justification Document

• Model evaluation protocol

• Model evaluation case studies: Approach and results

• Best Practice Guidelines

Link: http://www.elizas.eu/index.php/documents-of-the-action

Goals of COST ES1006

http://www.elizas.eu/index.php/documents-of-the-action

Testing hazmat dispersion models - model evaluation

model evaluation case studies:

Michelstadt - caseIdealized urban structure

CUTE - caseComplex Urban Terrain Experiment

COST ES1006 cases

I: The 'Michelstadt' case

wind tunnel setup

virtual city with typical European city structure

simplified geometry, more easy to be modelled but already more realistic than common cube arrays or building block arrays

several source positions and numerous measurement locations per source position

II: the CUTE Experiment

complex urban structure, continuous / puff releases

scale 1:350

95 continuous release scenarios

53 puff dispersion scenarios(> 200 releases each)

1 source location

45 minutes release

extensive met data

20 sampling positions

wind tunnel experiments field trial

COST ES1006 Model Types Classification

Model type Flow modelling approachDispersion modelling approach

Type I models that do not resolve the flow between buildings Gaussian

Type II

3D models for which the flow is resolved diagnostically or empirically, although not dynamically resolving the flow

between buildings (SCIPUFF, PMSS, QUIC,..)

Lagrangian

(LPDM)

Type III3D models that fully resolve the flow between buildings

(CFD, LES, LBM….)Eulerian(CFD)

Type I and Type II/III models

Cobalt-60 dispersion and dose evaluated using type I (left) and II (right) models

(wind from the North – source term due to the explosion from the ground to 20 m – 10 TBq)

SOURCE

xSOURCE

x

POOR BETTER

Input sensitivity

Comparison of concentration field with two turbulence inlet profiles, Type II models

Model evaluation

Example: CUTE case, blind test case, continuous releaseaffected area for two different time frames after the release

Type I Type II Type III

Michelstadt (upper) & CUTE case (lower), blind test, continuous release, mean concentration (ensemble)

Model evaluation


Model evaluation

Fractional Bias (FB) is a measure for mean bias and indicates systematic errors: overestimation (FB < 0) / underestimation (FB > 0)

Example: quantitative assessment of model performanceCUTE (blind test, case 3) - mean concentration

FB

common acceptance value: |FB| < 0.67Hanna S. and Chang J., 2013, Meteorology and Atmospheric Physics, 116, 133-146

𝐹𝐵 = 2𝐶𝑂 − 𝐶𝑃

𝐶𝑂 + 𝐶𝑃


Factor of two (FAC2) is a measure for the fraction of data points within the range

quantitative assessment of model performanceMichelstadt (NB: non-blind, B: blind) - mean concentration

common acceptance value: FAC2 > 0.3Hanna S. and Chang J., 2013, Meteorology and Atmospheric Physics, 116, 133-146

1

2≤𝐶𝑃𝐶𝑂

≤ 2

Model evaluation


Synopsis - continuous release scenarios:

in complex geometries, model performance measures are significantly affected by source locations and receptor points

for most of the models, metrics are within common acceptance values

performance increases with increasing complexity of the model, moving from Type I to Type III models

differences were observed between the blind and non-blind tests but no systematic dependencies were found

Urban effects: enhanced dispersion versus channeling

Model evaluation






o Conclusions.


FLADIS Ammonia Tests

CFD Simulation: MERCURE / Code_Saturne

RANS Model

Fladis experiment

High momentum jet experiment

Fladis experiment

27 trials

Fladis experiment

TRIAL 16

Release time 19h51

Tank pressure 7.9 bars

Tank temperature 17 °C

Release duration 20 minutes

Ammonia release rate 0.27 kg.s-1

Jet Momentum after flashing 18.4 N

Wind speed (z=10m) 4.4 m.s-1

Wind Direction (from the domain axis)

5 deg.

Relative humidity 65%

Ambient temperature 16 °C

Atmospheric pressure 1.025 x 105 Pa

Two-phase jet

Passive dispersion zone

Entrainmentzone

Expansion zone

Duct Breach

Fladis experiment

TRIAL 16 NX=50

NY=54

NZ =22

59400 Cells

Fladis experiment

TRIAL 16

NH3 GAS concentration NX=50

NY=54

NZ =22

59400 Cells

Animated GIF

Fladis experiment

TRIAL 16

NH3 Liquid AerosolIso 5E-5, 1E-5, 1E-6 kg/kg

NH3 Gas ConcentrationIso 5E-5, 1E-5, 1E-6 kg/g

Fladis experiment

TRIAL 16

NH3 Liquid Aerosol NH3 Gas Concentration

Fladis experiment

TRIAL 16

Fladis experiment

TRIAL 16

Jack Rabbit I ammonia release 2010.‘‘ Photos shows 5 seconds into first ammonia release on 07 April 2010.

Source : http://www.dugway.army.mil/NewsArticle.aspx?articleId=/PAO/Articles/2015/05/Saving%20Lives,%20Property%20and%20the%20Environment%20through%20Active%20Testing.htm

Jack Rabbit I tests






o Conclusions.


Jack Rabbit II tests: obstacles

PMSS simulation : JRII T5

Source term parameterization: high-speed Chlorine jet (pressurized tank)

• Total mass emitted is 8.95 tons.

• Liquid is retained into the pad and constitutes the pool source.

• Outlet velocity of flash phase 30m/s.

• Total mass emitted in each phase:• Flash phase = 25% of

the total mass (~2.25tons)

• Evaporated phase = 75% of the total mass (~6.75tons)

h : Cylinder height above the ground in m.

q : Expansion angle.

d : Tank aperture (15cm).

l : Aperture height (1m).

H / L : Tank dimensions (1.5m x 4m).

D : Cylinder diameter in m.


Concentration results NEST1 to NEST 4


Notes on JR II

• JR II tests, made with Chlorine, were shown as an example of the 3D effect of obstacles. Equivalent experiments with ammonia and obstacles are not available.

• Source term representation for JR II experiments involves high-speed jets hitting the ground, and a flash phase, because the release comes from a pressurized vessel and contains a large fraction of liquid. The initial vapor + aerosol mix (flash phase) is intrinsically denser than air because of droplets density.

• The modelling of source terms resulting from failure of pressurized tanks in general needs to jointly analyse two sources: • two-phase high-speed jets with aerosols,

• evaporation (boiling of a pool) emitting essentially vapour.

• A dense gas behaviour does occur for an evaporating pool of chlorine at ambient pressure, because Chlorine vapour is denser than air..

• Ammonia in vapour phase is lighter than air, so the releases from pool or reservoir evaporation (boiling) have positive buoyancy






o Conclusions.


Haifa Tank Simulation

3D Lagrangian Simulation: PMSS

High resolution Meteo Model

Lagrangian Particle Dispersion Model

Illustrates the combined physical effects of:

• Topography (complex micro meteorological flow pattern)

• Buoyancy against stability (plume rise limited by inversion)

(For a continuous release of vapor ammonia from a non pressurized refrigerated tank)

Haifa Tank simulation

Haifa Tank site

The ammonia tank in Haifa, with the city in the background. Source: http://www.haaretz.com/israel-news/business/1.775845

Lat : 32.818239°Long: 35.035997°

Views of the Haifa tank & site

http://www.haaretz.com/israel-news/business/1.775845

Surface wind field in Haifa

. Vectors color-coded as a function of wind intensity . Streamlines in the surface layer

Channeling and trapping effects

Wind direction driving the plume directly towards the steepest topography in

Haifa (“worst case” scenario) WD NE DD=45 degrees, WS 1.5 m/s, F stability).

Ground level Plume footprint

Plume channelingVapour cloud follows

small valleys

Scenario:

4,000 tons of refrigerated ammonia, evaporating pool formed in the tank with a diameter of 38m

3D view with topography, three iso-surfaces & ground contours of ammonia concentrations.

Buoyancy induced Plume rise

The plume from the Tank pool steeply rises up to its stabilization level of 55m ASL, due to the strong vertical stability (class F stability conditions), giving no significant impact close to the evaporating ammonia pool. The ground level impact starts being visible only several hundred meters downwind.

3D view of Plume impact on topography in stable conditions (limited plume rise)

Plume impact on topography

The 3D distribution of ammonia concentration is represented by 3 iso-surfaces (opaque reddish brown: 1.2 g/m3,

transparent orange: 0.5 g/m3, transparent yellow: 0.1 g/m3). The highest concentrations are present in the core of

the plume aloft, very different from the concentrations at ground level.

Plot of ground concentrations(Log scale) showing plume impinging terrain

Although channeling and crawling of the plume is apparent, the concentrations are low enough when the plume reaches the Carmel Mountain to avoid hazardous areas.

Complex terrain channeling

Plot of AEGL 1 and AEGL 2, along with circles of distance.

AEGL 1 & 2 (60 mn) contours

Tank ScenarioARIA MSS softwareD=38mMet: 1.5m/s – F –10°C

AEGL -1 (60mn) 22.5 mg/m3

AEGL -2 (60mn) 120 mg/m3

Animated GIF

3D animation of plume path

Animated GIF

View from the seaComplex topography

3D animation of plume path






o Conclusions.


Models of increasing complexity, ranging from Type I (Gaussian) to Type II (LPDM, Lagrangian puffs) and Type III (CFD, RANS or LES), were systematically compared to field and wind tunnel experiments in the framework of COST ES1006 action. Type II and III models, being intrinsically 3D Models, show better performance than Type I Models for urban short range simulation.

Examples were presented to show that complete 3D Models are essentially useful to represent the combined effects of:

• Complex terrain and channeling (Haifa Tank case)

• Obstacles and channeling (COST ES 1006 Cases and JR II case)

• Buoyancy and stability (Haifa Tank case)

The higher CPU requirements of CFD or LPDM models are not a serious problem anymore considering the current growth on demand for infrastructure (Big Data, AI…), as well as the stakes of accidental releases consequences.

Conclusions on 3D Models

Thank you for your attention !

Documents

NH3 Dispersion modelling - Chemical Engineering Facultychemeng.technion.ac.il/wp-content/uploads/2017/11/Jacques-MOUS… · 𝐹 =2 − + Type I Type II Type III. Factor of two (FAC2)