Reduced linear state model of a hollow blocks wall

Validation using hot box measurements

G. Yana,b*, J.J. Rouxb, Z. Lihuaa, R. Yezoub

a
School of Municipal and Environmental Engineering Harbin institute of technology,
92, West Da-Zhi Street, Harbin, 150001,China
b
Centre de Thermique de Lyon(CETHIL), UMR CNRS 5008, Bâtiment Freyssinet, INSA Lyon,
20, Av. Albert Einstein, 69621 Villeurbanne Cedex, France

Abstract

The hollow blocks are very often employed for the envelopes of buildings because of their
good performance from the point of view of the heat insulation. But until now the majority of
the codes of simulation of the thermal behavior of the buildings do not take into account this
type of walls. Indeed most often only the walls made up of a stacking of homogeneous layers
are treated in the great codes of simulation. In this paper, a reduced state space model, based
on linear model order reduction methods, is proposed. At best homogeneous equivalents
which respect steady state (U value) are proposed. The dynamic behavior can also be
approximate using these equivalents by supposing one-dimensionnal conduction heat transfer,
but this methodology leads to approximative dynamic behavior and is not generalizable. In
addition the use of a 3D description of the heat tranfers leads to models whose size implies
too important times of simulation. Let us remind that the codes of the thermo-aeraulic
behavior must enable us to carry out a simulation in dynamic mode over one complete year.
This reduced state model shows a very good accuracy when we compare it with the complete
state model (3D description) and the results of simulation are very close to the experimental
data we obtain using a hot box.

Keywords: Model reduction; hollow blocks; numerical analysis ; experimental data

1. Introduction

With the concerns for energy and enviroment for durable development, more and more
buildings envelope are constructed by hollow block for its better energy efficiency and its less
lands resources consumption. This kind of envelope has been widespread applied in many
countries from its appearance. And many new innovation are developed to adapt hollow block
to the new regulations by optimizing its thermal behavior and others properties [these
lacarrière]. To this type envelope, a great mumber of theory, numerical, and experimental
researchs have been represented over this decades. On the aspect of hollow block thermal
properity, thermal and mass transfer phenomena have been analysed, and numerous results
have been validated from many bibliographic study. In diversity of thermal parameters, U-
value is the most important one. Different from mutli-layers wall envelope, U-value of hollow
block envelope can not be easily got with 1- dimensional thermal transfer model owing to its
3-dimensional heat transfer between cavity and block rib, that is to say its heat bridge
disturbance. In general, this value is determined by two different procedure, the procedure
based on the hot box measurement following either EN ISO 8990 [1] or EN 1934 [2] and
second one by two- and three-dimensional mumerical calculations according to EN 1745 [3],
in Europe. As well known, tabulated U-values for different basic material and perforation
pattern are one common means adopted in many country building thermal regulations, such as
“Réglementation Thermique 2000” [4] in France. In this way, it is convenient for engineers
to determine heat loss of building and as well evaluate energy efficience of envelope. And
this method has been comprensively accepted by engineers. However, U-value is only a heat
loss indication for envelope under stastic state condition. With the further demands for
envelope heat loss evaluate at dynamic state condition, we need more accurately dynamic
results of hollow block envelope. But how to deal with the problem for envelope in hollow
block is not solved by a simple and accaptable way. And it is probably fair to say that the
most general building dynamic thermal analysis programs simplify the thermal conduction
problem by assuming one-dimensional (1D) heat flow through walls (DOE-2[5], TRNSYS[6],
CODYBA 6.0 [7]). So, it is impossible to accurately obtain heat loss of heat bridge. Indeed,
hollow block envelope does not been included in these programs. Therefore, we need a
reasonable model which satisfies for heat loss simulation of hollow block envelope. In this
paper, we propose one simple and pratical model by numerical analysis which make it is
possible to accurately and efficently model dynamic heat loss of hollow block envelope.
Various are avaiable In the literature by using transfer functions or thermal network models,
developing direct measurement procedures, sloving numberically the differential equation set
which describes the dynamic thermal behaviour of buildings.
Under stastic state, we can directly look up tabulate to determine U-value of some types
hollow block in thermal regulation. That brings much convenience for general thermal
calculation. An idea alike, if we can construct a group of simple matrixes which can be able to
describe dynamic and mutli-dimensional heat trnsfer in hollow block thermal system ? If the
answer is positive, then what we need to do is only looking up the group of low-order
matrixes corresponding for some type hollow block in the tabulate, then making a simple
dynamic cauculation under giving boundary condition. It is evidently that this cauculation is
very easy to be realize under computer if system model matrix order is enough low. To reach
this aim, we have two steps. The first step is the construction of thermal system equations and
the second one is system model’s order reduction for a low-order model.
It is not difficult to set up the differential algebraic equations for heat transfer problem
based on numerical heat transfer [6]. For a heat conduction problem example, a spatial
discretisation in n definite volumes allows to analyze the thermal system S in n definite
volumes. The center of each definite volume is the node whose temperature represents the
mean temperature of the volume. Then energy balance equation of each node i could be set up
according to “Kirckoff law”, and the equation is written in the state-space form Eq1:
(1)
where:
- T(t) is a vector (dimension: n,1) approximation of temperature field, which consists
of the temperature of Ti at different node of the spatial discretization.
- C is a square matrix (dimension: n,n), whose elements are the calorific capacity of
each control volume. It is diagonal definite positive.
-  is a square matrix (dimension: n,n)，which translates the heat exchange between
the different control volumes of thermal system S. It is a symmetric matrix because of the
reciprocity of the heat transfers.
- U(t) is a vector (dimension: p,1), which regroups the p solicitations acting on
boundary of S (flow in surface or applied temperatures in ambiance or on surface).
-  is a rectangular matrix (dimension: n,p), which translates the thermal relations
between the S domain and its environment. It represents modes of action of solicitations
among others on the system.
In general, the thermicist look forward to following the evolution of some outputs or
measures as that: the temperature in some point, the flow leaving one domain, etc. All these
valuables regroup in one vector Y, which can be expressed under the linear combinations with
constant coefficient of temperatures on nodes and inputs. We have:
(2)
Where:
- Y (t) is a vector (dimension q: number of outputs observed).
- J is the rectangular observation matrix (dimension: q,n).
-K is a matrix of direct transmission (dimension:q,p) which translates the
instantaneous actions of solicitations on outputs.
In this way, one thermal system can be clearly described by matrixes C, Λ, Π, J, K.
However, we know, to our hollow block thermal system modeling in multi-dimensional,
system matrixes order are very high because of its numberous differential equations.
Moreover, the dynamic simulatuion process for one multi-dimensional thermal system will be
often computationally intensive if we directly apply this high-order matrix model.
Thus, it is necessary to find a simple, low-order model to replace this complete model so
that it will benefit to simulation, calculation and computer implantation.

2. Linear state space model reduction

Along with the associated additional computational burdens from high-order model, the
simulation and computer implementation may become impossible. The need for low-order
model is thus clear. The objective of model size reduction methods is to replace the original
complete state model including n ordinary differential equations(n is commonly referred to as
complete model order) by a significantly smaller one(reduced model of order m<<n ) without
sacrificing vital characteristics of the physical system. Model reduction has been the subject
of numerous research endeavours and a great mumber of methods have already been
represented over the last 30 years. On that, spectral methods are widely used in buildings
thermal simulation domains[7]. Spectral methods represent the solution of problems as a
linear combination of the eigenfunctions of a particular basis. A low-order model is obtained
by truncation of the previous state-space coordinates change. Many methods for models
reduction are based on this principle: the modal basis truncation methods(such as methods of
Marshall and aggregation linear ) and balancing realisation methods(method of Moore) [8-
13].

2.1 Modal basis methods

a. Marshall method

MARSHALL [8] presented one of the simplest method, It is built on matrix partition
expressed in the modal base and obtained through a temporal criterion. The modals of system
are classed by decreasing time constants and remain the modes that have bigger time
constants, considering that only the modes associated with the time constants such as τi=-1/i
>Δt/4 ( is eigenvalue) are contributing to the dynamic behavior of the system observed with
a temporal step Δt. The rest smaller ones are truncated.
So, the reduced model related with Eq1, 2 is:

(3)

AM, BM, HM, DM are reduced system model matrixes by Marshall method which conserve the
biggest τi values (τi=| i-1 |).
 b. Linear aggregation methods

Different from the method of Marshall, the principle criterion for model reduction
associated by linear aggregation [9] is “energy contribution”. This energy contribution
characterize the “inputs/outputs” relation through state model. The importance of each modal
is evaluated by a energy contribution matrix E. Model reduction is implemented by keeping r
modals whose contribution is enough to approximately assess this system by criteria
.

The reduction model is written in Eq4

(4)

AAL, BAL, HAL, DAL are reduced system model matrixes by aggregation method.
However, the resulting reduction model with modal basis methods remains less quality
than that coming from balancing realisation method on its less accuracy[7,14] or less
efficiency. Results from our simulation show the merit of balancing realisation method also.
At the sight of this, we trend more towards balancing realisation method to reach our low-
order model.

2.2 Balancing realisation method (Moore)

Moore method is constructed from the concept of balancing realisation technique [13,14].
This technique is based on the central notions of controllability and observability. The
controllability gramian allows to measure the sensitivity of the state variables to the forcing
signals, and the observability gramian brings a way for measuring the sensitivity of the model
outputs to the state variables. In addition, controllability gramian and observability gramian
can be obtained from system matrixes. Consequently, reduction of the state variables showing
either a weak degree of controllability or a small degree of observability is a possible way for
model size reduction. The controllable and observable state elements are defined by their
controllability gramian WC and observability gramian WO, obtained by the follwing Lyapunov
equations:

However, the contrillability and the observability gramians are not system invariant, they
strongly depend on the state-space coordinates selected for system behaviour representation.
Furthermore, some conflicts among state variables can appear for a state variable showing a
weak degree of controllability can show, at the same time, a strong degree of observability,
and vice versa. To avoid such difficulties, the initial system is put throgh a particular similar
transformation in order to balance the degrees of observability and controllability. Each one
of the state variables in the resulting equivalent model has the same degree of controllability
and observability. In other words, the idea for balancing technique is to find a realization to
the reduction model, which corresponding obserbability and controllability gramians are equal
and diagonal: Wc=Wo==diag(1, 2, ...., n), where 1≥ 2≥... ≥ n0 are the so-called
Hankel singular[15] values and are the square roots of the eigenvalues of the product WcWo.
In a balancing model, the state components corresponding to small Hankel singular values
require much energy to be reached, while at the same time producing little energy on the
output. Therefore, if there is a S{1, ..., n-1} for which s>>s+1, the modal elements
corresponding to m, m=s+1, ..., n can be truncated from the system description, preserving
only those features of the dynamics that are most relevant to observable and controllable
modals. Model size reduction is then performed by simple elimination of the modal parts
associated to the state variables showing weak obserbability and controllability degrees[13].
After the calculation of grammiens of obserbability and controllability and balancing
transformations, expression of system in this base is :

Conserving m modals which have the most observable and controllable, the reduction
model is expressed by Eq9 :

(5)

AMO, BMO, HMO, DMO are reduced system model matrixes, where
; ; ;

3. Calibrated hot box measurement

The experimental set-up is calibrated hot box( CETHIL- INSA de Lyon), Fig1. This
apparatus can maintain a air temperature action on ambiance of hollow block sample wall,
and mesure the variation of heat flow and air temperature through sample wall under this
temperature excitation.
In our cases, two air temperature excitations at the both sides of sample wall (one linear
step and the other constant air temperature ) are controled by thermal couple controller. To get
a linear step(un palier bas et un palier haut) temperature excitation, a condensateur with
constant current is used for(au tental?). And the duration time (rampe) is changeable from
0.25h, 0.5h, 0.75h to 1h. In our experiment, two kinds of heat flow meters was used to
measure thermal flow through sample wall. The first one is in demension 25*25cm,
49W/m2mV calibrated at 20°C. And the second is in demension 50*50cm, 10W/m 2mV
calibrated at 20°C. The thickness of heat flow material(Cu-Co) is very slight( several
Decimillimetre about 0.5mm).
 Fig1 Schematic cross-section of the hot box: (1) sample wall; (2) heat flow meters; (3) thermocouples; (4)
isolation; (5) thermocouples.

Thermal couples were used for measuring the temperature of wall surface, and ambient air
temperature in boxes.

Cas1 Cas2

Fig. 2.Plan view and cross section of hollow blocks Cas1 and Cas2

Experiments were made on four types usual hollow blocks example walls envelopes.
Here, we only use two of them to validate our reduced model, one rang and two rangs cavity
block. These two types blocks are shown in Fig 2. Dimension of blocks, experimental
boundary condition set-up and number of modal for different model in numerical simulation
are shown in Table1.

Table 1
Experiment and simulation data
Block dimensions Cavity dimension Linear Duration
Complete Reduced
(mm) (mm) step of linear h(W/m
1 h2
2
K) (W/m2K) modals modals
Length Width Height Length Width Height amplitude step
Cas1 50 10 20 9 4 17 15°C 1h 10.8 7 680 5
Cas2 50 17.5 20 9 4 17 30.1°C 1h 5.5 7.7 973 5
*h1(h2) is calculated from measure data heat flow, wall surface and air temperature in the two boxes

4. Numerical simulation versus experimental data

Before numerical simulation, we firstly investigate heat transfer phenomena inside

cavities of hollow block. Inside these airtight hollows, we can observe a natural convection air
flow without any external fluid source. Heat conduction, heat convection between air and
surfaces of cavities, also long wave radition heat transfer among surfaces of cavities are
coupled with each other. On account of complicated heat transfer phenomena in these closed
cavities, equivalent thermal resistance is considered as the most usual method for study of air
layer thermal properity. So, this method is also taken in our works to simplify heat transfer
analysis in cavity. Above all, to conduction regime and convection regime in cavity, a usual
means of characterizing these two heat transfer phenomenon is by calculating the Nusselt
number.. Many studies on closed cavities have proposed some correlations [] about Nusselt
number accordind to thermal and geometric parameters of cavities. Then, convective heat
transfer coefficient can be written as follows :
(6)
a, b, c : coefficients depending on the existing natural convection correlations(Table2)
We have . If Tm is supposed to be 280 K and cavities surfaces temperatures
varie between 0°C and 30°C, radiative heat transfer coefficient varies between 4.74< hrad
<5.43 Ｗ/m2K.
Consequently, according to EN ISO 6946 [], we have air layer equivalent thermal
resistance: (10)

this radiative coefficient is assumed to be constant throughout the simulation.

Table 2
Nusselt number correlations,convective coefficients and radiative coefficients

hconv hrad
t Nu Gr (10-5) (W/m2K) (W/m2K)
5°C 2.76 0.637 1.73
15°C 3.64 1.9 2.27
30°C 4.33 3.8 2.7

And we suppose that physical properties of the material and other thermal parameters
have little variation under whole dynamic process. In our hollow block cases, value is
taken as 0.264 Ｗ /m.K. Following this determination of air layer equivalent thermal
resistance, the numerical alalysis was carried out by our state-space complete model Eq1,2
and Moore reduced model Eq.5 for hollow block thermal system. Reduced model of these
two cases only have 5 modals, for their complete model 680 modals and 973 modals
respectively. On this aspect, these two reduced model have satisfiying high efficience. Then,
we campare simulation results from complete and reduced model. Heat flow outputs at the
side of Box2 are shown in Fig3, 4 to Cas1 and Cas2 respectively. We can see, the relative
errors between the complete and reduction model is always very little, less than 0.01%.
Simulation results using one commercial software Heat3 [16] are shown in this figure also.
From this figure, we can also learn the results by Heat 3 agree well with that from our reduced
model. Experimental data from measurement demonstrated in Fig3 also reveal a good
agreement with that of reduced model. Furthermore, we can see that experimental results have
a little bigger than simulation values. That can be reasonablely explained because isolation
enclosing heat boxes can not make a complete adiabatic boundary condition under our real
condition. As the result, a little part of heat flows away from this isolation position.
Fig.3. Heat flow response: simulation and experiment comparison

To further ilustrate reduced model, Bode of Cas1 reduced model and its complete model is
shown in Fig5.
From this figure, we can see relationship of magnitude and phase of output Y (heat flow
through wall at the side of Box2) with frequence of two inputs U(1) ,U(2) (air temperature at
sides of Box1 and Box2). Between original complete model and Moore reduced model, it is
clear that U(1) frenquence reponse accords with complete model in whole frenquence range.
U(2) frenquence reponse of reduced model appears deviation from complete model when
input frenquence is higher than 0.00421HZ whose time constants is about 4 minutes.
Evidently, to our building envelope thermal system, it is reasonable to considered this high
frequence range as a relative ‘quick’ frequnce for building envelope owing to its thermal

inertia. We have similar results from other cases. So far as buildings thermal simulation
domain, small deviation of reduced model within high frequence range will not deteriorate
reduit model for getting a reasonable output. So we can conclude, Moore reduced model is
accurate and efficient enough to be used for characterizing building envelope thermal transfer
performance concerned by us.
Fig.4. Bode diagram of complete and reduced model: Cas1
 In sum, we can use model reduction to deal with all kinds of hollow block, and obtain
their reduced model in the form of very low order state space matrixes. These matrixes can be
easily implanted in program as a model library. So, This library make it possible for us to
analyse energy loss performance of hollow block envelope by a simple and efficient way.

5. Conclusions

Three-dimensional heat transfer character in hollow block makes it impossible to use

traditional multi-layers 1-dimensional model to simulate heat loss through this type envelope
under dynamic state condition. However, complete 3-dimensional model which directly come
from thermal system differential equations has a very high matrix order. This high-order
model restrain its application for its complexity. In this paper, we propose the possibility to
use model size reduction techniques for constructing reduced model which is fit for
implantation in computer. For that, thermal simulation calcultion for hollow block envelope
will be simple just like looking for a tabulate value. Moreover, efficiency and accuracy of
reduit model have satisfying results.

6. Nomenclature

As aspect ratio
A, B, H, D reduced model matrixes
C, , , J, K complete model matrixes
E cavity width of hollow block(m)
Ff view factor matrix
H heat transfer coefficient (W/m2K)
I unit matrix
Nu Nusselt number
Ra Rayleigh number
 thermal conductivity (W/mK)
o Stefan Bolzmann constant 5.67E-8W/m2K4
 emissivity emittance /absorptance coefficient
 heat flux W/m2

Subscripts
A air
eq equivalence
conv convection
rad radiation
MO Moore method
AL aggregation linear method
M Marshall method

References

[1] EN ISO 8990. Thermal insulation determination of steady-state thermal transmission

properties, Calibrated and guarded hot box.
[2] EN 1934, Thermal performance of building determination of thermal resistance by hot
method using heat flow meter. Masonry.
[3] En 1745, Masonry and produts: methods for determining design thermal values. April
2000.
[4] Réglementation Thermique 2000, CSTB, Paris, 2000.
[5] DoE2 Reference Manual, Parts 1 and 2. US Department of Commerce, National Technical
Information Services,1985.
[6] TRNSYS-A transient system simulation program, Madison, Solar Energy Laboratory
University of Wisconsin 1994.
[7] J.J.Roux. Proposition de modèles simplifiés pour l’étude du comportement thermique des
bâtiments. Thèse de Docteur Ingénieur, INSA de Lyon, 1984.
[] J.J.Roux. Modélisation du comportement thermique et aéraulique des bâtiments: sous
structuration, réduction et méthode de couplage. HDR, INSA de Lyon, 1993.
[] Elena Palomo Del Barrio, Gilles Lefebvre; Philippe Behar, Noël Bailly. Using model size
reduction techniques for thermal control applicatios in buildings. Energy and Building
33(2000),P1-14
[] S.A. Marshall. An approximate method for reducing the order of a linear system .Control
1966, vol 10, p642-643.
[] Aoki.M. Control of large-scale dynamic system by aggregation. IEEE 1968, vol 13, p246-
253.
[9] G.P.Michaïlesco. Approximation des systèmes complexes par des modèles de dimension
réduite. Thèse d'Etat, Centre d'Orsay, Université de Paris-sud, 1979.
[] A E.Biyaali. Réduction des systèmes complexes multi-entrées multi-sorties application aux
singularités thermiques dans les bâtiments ,Thèse de Doctorat, INSA de Lyon,1995.
[] EN ISO 6946, Building components and building elements ; thermal resistance an thermal
transmittance, calculation method.
[] Zhao Lihua, J.J.Roux ,Gao Yan. Modelization of thermal behavior of building by coupling
reduction state model. The 4th International Conference on Indoor Air Quality, Ventilation
&Energy Conservation in Buildings, Changsha, 2001.
[] B.C.Moore. Principal component analysis in linear system : controllability, observability,
and model reduction, IEEE Trans. Automatic Control, 1981,vol.26, p17-32.
[] C. Menezo. Contribution à la modélisation du comportement thermique des bâtiments par
couplage de modèles réduits, Thèse de Doctorat, INSA de Lyon,1999.
[] R.Markovinovic, E.L.Geurtsen,T.Heijin, J.D.Jansen. 8th European Conference on the
Mathematics of Oil Recovery, Freiberg, Germany, 2002.
[] Heat3,Lund-Gothenburg Group for Computional Building Physics, Lund university &
Massachusetts Institute of Technology, 2001
[] R.KmacGregor, A.F.Emery, Journal Heat Transfer,91,391,1969

C. MENEZO, J.J. ROUX, J. VIRGONE

Modelling heat transfer in building by coupling reduced order models
Building and environment, volume 37, n02,pp133-144, 2002
F.DEQUE, F.OLLIVIER, J.J.ROUX
Effect of 2D modelling of thermal bridges on the energy performance of
buidings, numerical application on the Matisse apartment
Energy and Buildings, volume 33 n°6, july 2001, p583-587.
Bruno LACARRIERE
Titre de la Thèse : « Etude numérique et expérimentale des échanges thermiques en
cavités partitionnées : application aux transferts instationnaires dans les briques à
perforations verticales.»Thèse de Doctorat de l’Université Paul Sabatier, Toulouse,
soutenue le 11/12/2003