Transcript The presentation template

School of Civil Engineering
FACULTY OF ENGINEERING
Computational fluid dynamics (CFD)
based multi-objective optimization (MOP)
of hospital ward ventilation
M A I Khan C J Noakes V V Toropov
Pathogen Control Engineering Institute
Outline
•Airborne pathogens and disease spreading
•Infectious diseases and ventilation in hospital wards
•Computational Fluid Dynamics
•Why numerical Optimization
•RSM and GA optimization
•Results
•Conclusions and discussions
Airborne pathogens and
disease spreading
0.001 to 1000 mm
•There is strong evidence suggesting a link between
ventilation and the transmission/spread of infectious
diseases in indoor environments (Li et al 2007).
•The recent H1N1 pandemic is just one example of the
threat that airborne pathogens pose to human health.
•In order to reduce the risk of airborne infection it is
vital to understand the mechanisms dictating pathogen
transport through the air.
•Especially important in hospital environments where
patients with weakened immune systems are particularly
vulnerable.
•Exhalation flow from patients with airborne infectious
diseases can impose health risks to caretakers and
visitors
Health Tech
•Hence, effective ventilation in hospital wards is very
Memorandum (HTMO)
important in order to control respiratory disease
•
Spanish Influenza (1918-1920) 50 - 100
transmission through air (Tang et al. 2006).
million deaths
•
Asian flu (1957-1958) 1.5 - 2 million deaths
Ventilation of healthcare
environments
•
•
•
•
•
Ventilation in healthcare environments
has a dual purpose
Provides a comfortable indoor
environment for occupants
Controls
the
distribution
of
contaminants,
particularly
the
transmission of airborne infectious
particles.
Design of appropriate ventilation
depends on the level of risk, heat loads
in the space and occupant activity.
Understanding how to balance these
sometimes conflicting requirements is
increasingly important, particularly as
the energy demands of the system are
also now a critical factor.
•Application of computational approaches to study ventilation airflow patterns in
enclosed spaces such as hospital wards, office rooms etc. has attracted
considerable interest among engineers and scientists over the last few decades
(Yam et al., 2011)
•Most of the work to date uses Computational Fluid Dynamics (CFD) for
parametric study of the influence of the airflow on the transport of heat and
contaminants, including airborne pathogens (Noakes et al., 2006; Ho et al. 2009)
in enclosed spaces.
•While such studies indicate that certain ventilation regimes
or rates may be better than others for a particular scenario
they do not formally seek an optimum design. In addition,
running multiple CFD simulations at design stage can be
time consuming and prohibitively expensive for many
organisations.
•Numerical optimisation approaches offer the potential to both find the best
design in a particular scenario and also create design tools that allow a more
robust selection of parameters in a given case.
•Application of optimisation techniques to airflows in building environment
control is a more recent area of development
Computational Fluid Dynamics (CFD)
•CFD is one of the branches of fluid mechanics, which uses
numerical methods and algorithms to solve and analyse fluid flows.
•CFD is used in various domains, such as oil and gas reservoir
uncertainty analysis, aerodynamic body shapes optimization (e.g.
planes, cars, ships, sport helmets, skis), natural phenomena
analysis, numerical simulation for weather forecasting or realistic
visualizations.
•CFD problem is very complex and needs a lot of computational
power to obtain the results in a reasonable time.

i

  
iu j    i   Si ; j  1,2,3

t x j
x j  x j

CFD Modelling Process
Define geometry and computational mesh
Physics of Flow
Flow conditions (fluid properties, turbulence, buoyancy)
Boundary conditions (inlets,outlets, heat and contaminant sources)
Model specific parameters (eg.contaminant transport and removal)
Solve for velocity, pressure, temperature and other variables
Visualise and evaluate results
Mini course on turbulence
Big whorls have little whorls
That feed on their velocity,
And little whorls have lesser whorls
And so on to viscosity--- L F Richardson
(N~Re9/4)
ui

1 p


ui u j   


t x j
 xi x j
where i  1, 2,3
  u u j 

  i 
  Tij 
  x j xi 

Tij  ui u j  ui u j
Tij  2 T Sij where  T  C  2 S
u ( x)   u ( x)G( x, x)dx

u (t )  u (t )  u (t )
Numerical Optimization
•Numerical optimisation is increasingly popular in
many fields of engineering
•Most optimisation techniques can be broadly
classified as either deterministic (gradient based)
(Deb 2001) or stochastic (gradient free).
•The nonlinear nature of flow phenomena inside
enclosed spaces, such as rooms, leads to
discontinuous outputs being generated which in turn
causes problems for gradient-based methods
(Wetter et al., 2003).
•
•In contrast, gradient-free methods, also referred to
as global methods are based on stochastic
approaches and are better suited to building or
indoor environment applications.
Find a set of design variables X (such as
temperature, velocity etc.) which optimizes
(minimise of maximise) an objective/cost
function f(X) (Rao S S., 2009)
f ( X ); X  R n
•One of the most popular in this category and a
•
widely accepted global optimization technique is the
Genetic Algorithm (GA) method (Holland 1975).
•
Subjected to the following inequality
g j ( X )  0; j  1,2,
m
and equality constraints
hk ( X )  0; k  1,2,
r
•
Inspired from Darwin’s theory of natural selection, this method has demonstrated its
capability to handle discontinuous variables and also noisy objective functions
(Wright et al., 2002).
•
But GAs, require hundreds or sometimes thousands of evaluations of the objective
functions to search for the optimal solutions (Magnier et al., 2010).
•
In building or indoor applications, where evaluation of the objective function comes
from computationally expensive and time consuming CFD simulations.
•
The optimization process could therefore take a prohibitively long time to achieve its
goal.
•
Hence, in order to save computational time associated with GA, a surrogate based
Response Surface Approximation method is used to mimic the behavior of the
system (in our case the indoor air flow field) response with respect to the change in
design variables.
•
The surrogate models which are constructed from high-fidelity simulations provide
fast approximations of the objective and constraint functions at new design points
•
Thereby saving computational time and making optimization studies using GA
feasible (Queipo et al., 2005).
Optimization Process
PARAMETER
Number of CFD
responses used as
building points
Number of CFD
responses used as
Validation points
VALUE
15
R2 Building points
0.9932
R2 validation Points
0.9931
R2 Merged
RMS Error Build
0.9947
0.0108
RMS
Error
Validation
RMS Error Merged
0.0091
PARAMETER
Maximum
Iteration
Minimum
Iteration
Coding Type
Population size
Discrete States
Mutation Rate
Global search
Elite Population
%
Random Seed
Number of
Contenders
Penalty
Multiplier
Penalty Power
Problem formulation
Find Min of f(x), subject to g(x) ≤0
5
Design of Experiments (DOE)
Optimal Latin hypercube (OLH)
Numerical simulation (CFD)
Determination of objective function at each DOE point
0.0092
VALUE
200
25
Real
20
1024
0.01
2
10%
1
2
2.0
1.0
Construction of surrogate
Moving least squares method (MLSM)
Surrogate model validation
Evolutionary Algorithm
Invoking GA to search for the min of the surrogate of f(x)
Optimal Solution
Test Room Setup
•PACE Bio-test chamber
•Two sets of inlet and outlet
•The dimensions of the chamber
L=4.2m, H=2.26m and W=3.36m
•6 air changes per hour (ACH)
Standard ventilation rate used in
hospitals
Numerical Results from CFD
•We simulate air, temperature and pathogen concentration inside the test
chamber
•We use the finite-volume code FLUENT for all our flow simulations
•Walls and ceiling assumed as constant temperature surfaces
• We solved the flow using RANS method
•Buoyancy effect assumed negligible
•Turbulence model used RNG based k-e
•Transport equations for a scalar pathogen concentration and temperature were
solved simultaneously with the flow
• Scalar sources representing the infection source were inserted inside the
room and the corresponding monitoring regions representing healthcare
workers were used for optimisation study
• Normalised scalar or pathogen
concentration on plane normal
to the inlet for two different
outlet positions.
L
Simulation
Parameters
values
6 ACH
32m3/hrs
Re=uinD/;
D=4A/P
21,000
Tin
300K
Twall
295K
Number of
cells
0.53 million
X
Z
H
Y
W
f ( X )  wC C  wTres Tres ,
where Tres 
Tr  Ta 10 v
1  10 v
X  outlet position
.
System response parameters with source S1
and two different monitoring regions A1 (top
plots) and A2 (Bottom plots)
Optimisation results from
HyperStudy
Response curve using
moving least squares
GA convergence history for two different
monitoring regions A1 and A2 same source S1
Sensitivity of the surrogate with
respect to weights
Monitoring region A1
Monitoring region A2
Source S1
Source S2
Source S1
Source S2
Complex geometries and
boundary conditions
Conclusions & Discussions
•We have used numerical optimization techniques to address the problem of
infection control together with comfort inside a hospital room/ward.
•Our results show that, the optimum design configuration of the ventilation
system in a simple test room taking into account both infection control and
patient comfort is attainable.
•However, the sensitivity of the weights chosen by the designer and also on the
choice of monitoring regions has a substantial influence on the results.
•CFD simulation in a realistic hospital room will still be expensive
•Accelerate CFD simulations via massively parallel simulations or use
simplified or coarse flow solvers!!
•Further work is necessary to understand dynamic environments such as moving
and heat generating healthcare worker and patients, doors opening etc.
•Non spherical aerosol transport and deposition.
[1] Y. Li, G. M. Leung, J. W. Tang, X. Yang, C.Y.H Chao, J. Z. Lin, J. W. Lu, P. V.
Nielsen, J. Niu, H. Qian, A. C. Sleigh, H.-J.J Su, J. Sundell1, T. W. Wong and P. L.
Yuen, Role of ventilation in airborne transmission of infectious agents in the built
environment a multidisciplinary systematic review.Indoor Air, 17, 218 (2007).
[2] R. Yam, P. L. Yuen, R. Yung and T. Choy, Rethinking hospital general ward
ventilation design using computational uid dynamics, Journal of Hospital Infection, 77,
31-36 (2011).
[3] C. J. Noakes, P. A. Sleigh, A. R. Escombe and C. B. Beggs, Control of Infections :
Original Paper Use of CFD Analysis in Modifying a TB Ward in Lima, Indoor and Built
Environment., 15, 41-47 (2006).
[4] J. A. Wright, H. A. Loosemore, R. Farmani, Optimization of building thermal design and control by multi-criterion genetic algorithm. Energy and Buildings, 43, 959
(2002).
[5] FLUENT documentation, version 12.1, ANSYS Inc, (2009).
[6] HyperWorks v10, Altair Engineering Inc, USA, (2009).
[7] CIBSE Guide A Environmental Design, The Chartered Institution of Building Services Engineers, London, (2006).