Energy and Buildings 155 (2017) 378–389

Contents lists available at ScienceDirect

Energy and Buildings

journal homepage: www.elsevier.com/locate/enbuild

Replication Studies

A ventilation system controller based on pressure-drop and CO2

models
Gerhard Zucker a,∗ , Andreas Sporr a , Antonio Garrido-Marijuan a , Tarik Ferhatbegovic a ,
René Hofmann a,b
a
AIT Austrian Institute of Technology, Center for Energy, Donau-City-Straße 1, 1220 Vienna, Austria
b
TU Wien, Institute for Energy Systems and Thermodynamics, Getreidemarkt 9/302, 1060 Vienna, Austria

a r t i c l e i n f o a b s t r a c t

Article history: In ventilation systems a considerable amount of energy is used for mass transport of conditioned air
Received 28 April 2017 to provide the requested volume flows. Reducing volume flow while maintaining indoor air quality has
Received in revised form 31 July 2017 leverage on energy efficiency and is commonly known as Demand Controlled Ventilation (DCV). Current
Accepted 15 September 2017
implementations require Variable Air Volume (VAV) controllers to provide a defined volume flow to
Available online 18 September 2017
each room. The controllers measure the pressure difference and adjust the motor flap accordingly. This
paper examines an approach that achieves DCV, but replaces the VAV-boxes with simple motor flaps.
Keywords:
The missing pressure-drop measurements that allow calculating the volume flow are substituted by a
Ventilation system
Pressure-drop model
model of the ventilation system. The authors develop a method for calculating the pressure drop in the
Ventilation controller ducts of a ventilation system that regards the topology and the components of the duct system. This
Volume flow optimization model is coupled with a model for the CO2 concentration in the rooms for all conditioned rooms in order
to derive the required air volume flow. Using this model, a linear controller is developed that operates
the ventilation system. It is shown that the presented approach operates the ventilation more efficiently,
while maintaining comfort constraints and saving installation costs. The modeling effort of the current
approach is expected to be reduced with the introduction of the Building Information Model (BIM) into
building operation.
© 2017 Elsevier B.V. All rights reserved.

1. Introduction reduced volume flow will in parallel also save energy for heating,
cooling, humidification and dehumidification.
Energy consumption in buildings represents 40% of the final At the same time, buildings need to balance between maximum
energy demand in Europe – and they consume more energy than energy efficiency and optimal indoor air composition and quality
any other sector of the European economy [1]. About half of the (IACQ). IACQ has gained great importance due to the significant
energy consumed in building operation is invested into heating, increment of illnesses reported by building occupants, referred as
ventilation and air conditioning (HVAC) [2], whereby in the US sick building syndrome (SBS) [4,5]. Z. Lin et al. [6] suggest that
ventilation represents about 61% of the total electric power con- multiple factors are involved in this syndrome, including indoor
sumption in office HVAC systems [3]. In many modern buildings the air quality (IAQ) such as microbiological and chemical exposures
air conditioning system does not contribute significantly to heat- not adequately characterized by current assessment approaches.
ing or cooling, but shall rather condition the indoor air to be at This study also shows that with proper design, installation, mainte-
the right level of CO2 and within the comfort zone in the Mollier- nance and operation, the ventilation mechanisms of buildings can
h-x diagram. In this work, the authors therefore focus on energy maintain satisfactory levels of IAQ and therefore reduce the SBB
optimizations based on volume flow reductions and disregard the incidence. Today there is no real integration of IACQ and energy
energy needed for heating or cooling the air as well as the energy efficiency, being mainly the domains of different communities
for adjusting the humidity – based on the valid assumption that working independently. Different studies [7,8] have shown that
some energy efficient buildings reduce the conventional energy
consumption by means of reducing volume flow with a number
of ventilations below the standards, and therefore a reduction on
∗ Corresponding author. IACQ. By observing how HVAC designers commonly address the
E-mail address: gerhard.zucker@ait.ac.at (G. Zucker). above 40% of energy use for space heating and cooling, it is found

http://dx.doi.org/10.1016/j.enbuild.2017.09.041
0378-7788/© 2017 Elsevier B.V. All rights reserved.
 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389 379

Nomenclature

Abbreviations
CSP Constant static pressure
DCV Demand controlled ventilation
HVAC Heating ventilation and air conditioning
IACQ Indoor air composition and quality
NOAA National oceanic and atmospheric administration
SPR Set-point reset
VAV Variable air volume

Latin Symbols
a [-] Constant
b [-] Constant
A [m2 ] Cross section of the duct
ĊCo2 [ppm/s] CO2 production per hour of the persons
Fig. 1. Cubic relation between power consumption over volume flow for a
C [ppm] CO2 level Robatherm RMC 06/12 fan.
d [m] Duct diameter
l [m] Duct length
as low as possible, as seen in Fig. 1, while maintaining the indoor
L [Pa s2 /m3 ] Hydraulic inductance
comfort.
n [1/h] Air change rate through windows and the ventila-
The second motivation is to exploit the designed systems for
tion system
reducing costs. While the operation costs are automatically reduced
p [Pa] Pressure
by the targeted optimization of volume flows (which reduces con-
R [] Electric resistance
sumption of electric energy, of thermal energy due to reduced
R [J/(kgK)] Specific gas constant
volumes that need to be conditioned and of maintenance costs due
Re [–] Reynolds number
to reduced stress on the mechanical fans), the installation costs are
V̇ [m3 /h] Volume flow
not addressed. In order to also reduce investments when construct-
ing the building, expensive components can be replaced. In the case
Greek Symbols
of ventilation systems, these are the VAV boxes. In principle, the
␣ [deg] Flap angle
VAV boxes can be replaced by motor flaps, which are also able to
 [1] Difference
adapt the volume flow for each room, but lack the additional sen-
 [m] Friction factor
sory equipment to measure the volume flow: a VAV controller can
 [kg/m3 ] Density
maintain a given volume flow set-point, so the volume flow into
ϑ [m/s] Velocity
the room is known. A motor flap can only control the flap position
 [m2 /s] Kinematic viscosity
to open and close between 0% and 100%. The volume flow, how-
ς [-] Pressure drop coefficient
ever, is unknown, since it depends not only on the flap position,
Indices but also on the current pressure in the duct system. By introduc-
a Outside ing a model for the motor flap that contains the dependencies of
i Inside flap angle, pressure and volume flow, we can use the flap model to
w Window derive the volume flow, given that the pressure in the duct system is
known. This way the construction costs for the building are reduced
in both hardware costs and installation costs by performing a shift
from hardware to control design and plant modeling.
out that the design focuses on minimizing energy use/cost, which Analysis of indoor air quality issues in national regulations are
often leads to unhealthy indoor comfort conditions. commonly addressed by measurement of CO2 levels, based on its
association with human body odor and other pollutants [9], not
2. Motivation and technical concept with any health or comfort effects of CO2 itself. This approach relies
on the fact that CO2 detection technologies are well established and
Providing indoor air quality and minimizing the energy demand it is common knowledge that this technology provides quite good
for the ventilation system are contradictory goals. An optimization indicators of IACQ, which in turn are used for operating the building
needs to be well-designed in order to provide the same, if not better, ventilation system.
services than conventional buildings and to reach the same comfort This paper is organized as follows: Section 3 discusses the exist-
parameters and indoor air quality. The main electric consumers are ing state of the art in control strategies for ventilation systems,
the fans in the duct system, namely the supply air fans and exhaust Section 4 describes the models that are needed for the design of
air fans. Therefore the first motivation is to find an appropriate con- the new control strategy, which is then described in Section 5. The
trol strategy for controlling the fans. Volume flow and pressure have results are presented in Section 6, followed by the Conclusions in
a quadratic relation, volume flow and power consumption have a Section 7.
cubic relation as shown in Fig. 1. A reduction of the volume flow by
10% results in an electric power reduction of 27%, thus giving a sig- 3. State of the art
nificant leverage for energy savings. This both applies to reduction
of the overall volume flow that is required to meet the comfort cri- Fig. 2 shows the control scheme of a CO2 room control: the linear
teria, but also to avoiding periods of high volume flow followed by PID-controller Control maintains the CO2 level in a room, it receives
periods of low volume flow, if it is possible to level out these periods the difference between the CO2 set-point and the current CO2 sen-
to an average volume flow. An energy efficient control strategy for sor value; its controlled variable is the volume flow V̇ . The VAV
a given ventilation system shall attempt to keep the volume flow box modulates an internal flap (using the flap angle ␣) to deliver
380 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389

Fig. 2. Ventilation model and room models of the controlled plant.

the requested volume flow, the internal controller VAV Control uses 4. Plant modeling
the pressure difference p to operate the flap, resulting in the vol-
ume flow reaching the room and affecting the CO2 level in the room The schematic in Fig. 3 shows a topological model of the plant
(a modified control strategy shall achieve the same, but without the that shall be controlled. The supply air fan receives a set-point for
internal VAV controller, solely operating on the flap angle). the differential pressure that it shall provide to the duct system. The
In a ventilation system the pressure set-point of the fans is often volume flow into the different rooms is controlled either by setting
fixed and does not regard the current demand situation. This con- a set-point for the volume flow in a VAV controller or by setting the
trol strategy is referred to as Constant Static Pressure (CSP) [7]. flap angle of a motor flap. Each room is equipped with a CO2 sensor
It is commonly used, due to its simplicity and its low costs: CSP that provides the current CO2 concentration (CO2 level).
does not require communication between the supply air fan and The model follows an application oriented design, since it shall
the VAV controllers, thus saving equipment and installation costs. be applicable to building energy management applications. There-
Consequently, fan control and room control cannot cooperate; this fore the model uses only available data from sensors and actuators
implies that the supply air fan provides a pressure in the ducts of that are commonly installed in a ventilation system and are accessi-
the system, which may be too high for the current demand. The ble by the supervisory control system. It does not require additional
excess pressure in the duct system is compensated by the VAV invasive sensor placement in the duct system. Therefore, the pres-
controllers at the room inlet by closing their flap and reducing sure in different segments of the duct system cannot be measured,
the aperture in order to compensate for the high pressure. In CSP, but is calculated using schematics of the duct system as shown in
the pressure of the fan is calculated once (usually during commis- Section 4.1.
sioning) for one specific occupancy scheme and is scarcely adapted The duct system (including the supply air fan) and the CO2 mod-
during operation. CSP has two significant disadvantages: firstly it els for the rooms are the foundation for the controller design that is
runs the risk of violating the comfort zone by providing low IACQ, covered in detail in Section 5. The duct system model is a pressure
if the fan pressure is too low; secondly it may suffer from excessive drop model, which derives the pressure drops for all paths between
energy demand due to an over-pressured duct system. Static Pres- supply air fan and room supply outlets as well as the volume flow
sure Set-point Reset (SPR) [8] is a control strategy that addresses in each segment. The relation between volume flow and CO2 level
these disadvantages: it allows reducing the fan pressure set-point is calculated in the CO2 room models, which assumes that supply
based on the required volume flow. SPR requires VAV controllers volume flow and exhaust volume flow are the same, thus only the
in each room, which have to communicate with the fan controller. supply side is modelled, and the exhaust fan is operated identically
The pressure is reduced until at least one VAV box is fully open (i. to the supply fan.
e. the flap that is controlled by the VAV controller, respectively). At Fig. 4 shows the dependencies of the sensors and actuators in
this point any further reduction in pressure would cause too low the system: The supply air fan receives a set-point. While it tries to
volume flow in the room. Then the pressure is slightly increased reach this set-point, it passes the current pressure to the duct sys-
again. This is necessary to detect, if a VAV box needs to open beyond tem model in the dynamic calculation. Afterwards the VAV boxes
100%. SPR needs to have a relaxed timing regarding the modifi- receive their volume flow (or their flap angle set-points, respec-
cation of the pressure in order to avoid oscillations between fan tively) and the current volume flows from the duct system model.
control and VAV controller reaction. Thus the VAV boxes cannot The dotted line indicates that all of the VAV boxes/CO2 room models
react on fast occupancy (e.g. beginning of a lecture) changes, which receive their different set-points and pass on their current volume
can cause violations of the comfort. SPR is used in different VAV flows.
controller types like the Belimo VAV compact controller [10] and The following sections elaborate the details on the models for
other controllers. the duct system (Section 4.1), the pressure characteristics of the
In [11] the authors extend the SPR approach and suggest a CO2 - VAV controllers and motor flaps (Section 4.2) and the model for the
based and occupancy-sensor-based dynamic reset of volume flow CO2 concentration in the rooms.
for multiple zone HVAC systems and indicate cost savings due to
reduced outdoor volume flow. The system still depends on con-
4.1. Modeling the duct system
trolled volume flow by VAV controllers. The impossibility of setting
an accurate volume flow by motor flaps, does not allow a replace-
The pressure drop model of the duct system calculates the cur-
ment by them. The Trim-and-Respond control strategy [8] is similar
rent volume flow into the rooms at any moment. This model is
to SPR in that it attempts to reduce the volume flow and save
based on the pressure drop of the ducts and reflects the topology of
energy; it differs in implementation by “trimming” the pressure
the duct system. According to [12] and [13] the pressure drop p
set-point until a supplied room or zone issues a request that it
in a duct segment is calculated as:
needs more pressure to fulfil its requirements for IACQ. While trim-
ming has slow dynamics, the requests for more pressure are rapidly l  2
executed in order to stay within air quality constraints. p =  ϑ (1)
d2
 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389 381

Fig. 3. Ventilation model and room models of the controlled plant.

Fig. 4. Electrical substitute circuit diagram (left) for a ventilation system with three rooms (right).

 2 since the sensor readings in the building automation system

p = ς ϑ (2)
2 provide volume flows. The flow’s Reynolds number Re is defined
as
In Eq. (1), l is the length of the duct segment, d the diameter,  the
density of the supply air and in both equations ϑ is the air velocity.
For each straight duct segment Eq. (1) can be used, with  being the
friction factor [13–16], which describes the friction losses in pipe ϑd
Re = (4)
flow, depending on the Reynolds number (Eq. (4)) as well as on the 
roughness and on the diameter of the duct. The pressure drop of
special duct types (bifurcations, elbow ducts and other non-straight
ducts) can be calculated using Eq. (2), where ς is a coefficient that with  being the kinematic viscosity of the supply air. The dynam-
depends on the shape of the duct and can be found in well-known ics of the duct system can be modelled using the electric analogy
literature [12,17]. An important conversion is the relation between of an inductance. The hydraulic (or pneumatic) inductance accord-
volume flow and velocity as ingly stores kinetic energy of the volume flow and consequently
prohibits instant changes of the volume flow (which is the elec-
V̇ 1 dV tric current in the analogy). It reflects the dynamics in the pressure
ϑ= = (3)
A A dt drop model and allows solving it for the different volume flows. The
382 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389

hydraulic inductance L for a duct segment is used in the pressure to a small simulation error. Within the time period in which the
drop equation as: controller is tested, the amount of error is irrelevant and thus can be
neglected. During “real-time” operation the controller frequently
d2 V l
p = LV̈ = L with L = (5) obtains new volume flow values from the VAV boxes, which elim-
dt 2 A
inates aggregation of errors over time.
Using the dynamics of the ducts, a dynamic simulation can be After validating the model it is possible to replace the model of
used to calculate the controller coefficients (see Section 5). The duct the VAV box with a motor flap model. For this case the model is
topology in the building has to be compiled using a combination redefined as a static one and the dynamic parts (i. e. the change in
of duct segments, each causing a certain pressure drop. As a start
volume flow V̈ ) are deleted, looking only at a static pressure drop
the building schematics and the floorplan can be used to estimate
model. In a realistic operation environment this is a valid assump-
the number of supply inlets and exhausts. However, the schemat-
tion, since the control does not require continuous strong changes,
ics typically reflect only the as-planned state, but not the as-built
but rather remains on a static set-point. Then the total pressure
state; therefore there should be a visual inspection of the duct sys-
drop between supply fan and each room is calculated, except for
tem to verify the topology and the components that were used.
the flap pressure drop. The pressure, which needs to drop at the
Since the pressure drop is not overly sensitive on the length of seg-
flap, is the difference between this calculated pressure drop and
ments, as the pressure drop per meter ducts is low, the focus of the
the provided fan pressure. By using reference data, the controller
visual inspection is on the installation location and completeness
can calculate backwards, which opening angle is needed. Occur-
of the different pipe parts and components like dampers and bifur-
ring model inaccuracies will be compensated by the controller once
cations. Also, any possible leakages shall be recorded to improve
it is operating on the real plant (i.e. the ventilation system). The
the reliability of the model.
controller uses the difference between the CO2 -set-point and the
After the definition of the different pipe paths, the static and
current CO2 -value for calculating the required volume flow. After-
dynamic pressure calculations have to be combined. This is done in
wards it uses the abovementioned dependence between opening
the electric analogy using the electric current I as volume flow V̇ ,
angle and volume flow to calculate the needed angle, if the equa-
the resistance R as  dl 2 (or ς 2 , respectively) and the inductance
2A 2A tions are solved for RVAV1 , RVAV2 , RVAV3 .
L as l
A
. Note that in the hydraulic system the volume flow V̇ has a
squared ratio to the pressure drop instead of a linear one like the
4.2. Motor flap and VAV-box characteristics
current to the voltage.
The following example is developed for three rooms (Fig. 4). It is
One of the most common flaps are so-called Butterfly flaps. They
simplified just to show the coherence between the different volume
have a moveable part in the middle of the duct, which can be con-
flows in these rooms.
trolled to adapt the volume flow for different pressure conditions.
Using Kirchhoff’s second law we can derive the following for-
There are different ways to calculate the pressure drop of these
mulas:
flaps: preferably the manufacturer publishes a datasheet, which
2
pfan = R1 V̇fan + L1 V̈ fan + (RP1 + R2 + RVAV 3 ) V̇32 + L2 V̈ 3 (6) includes the characteristics for pressure drop depending on the vol-
ume flow. If this is not the case, literature provides measurements
and equations to calculate the required characteristics [17–19].
(RP2 + R111 + RVAV 1 ) V̇12 + L111 V̈ 1 VAV-boxes are similar in design and can therefore be modelled the
  same way. As opposed to VAV boxes, a motor flap does not mea-
= RJ2 + R12 + RVAV 2 V̇22 + L12 V̈ 2 (7) sure the pressure and thus cannot derive the current volume flow.
VAV controllers usually work with pressure drop calculations. They
    2 measure the pressure before and after the motor flap. It is possible
RJ2 + R12 + RVAV 2 V̇22 + L12 V̈ 2 + RJ1 + R11 V̇1 + V̇2 to calculate the velocity and consequently the volume flow. It is pos-
  sible to calculate the velocity, as seen in Eq. (3), and consequently
+L11 V̈ 1 + V̈ 2 = (RP1 + R2 + RVAV 3 ) V̇32 + L2 V̈ 3 (8) the volume flow depending on the Bernoulli-equation:
␳
p+ = const. (9)
The size of the set of formulas increases with the number of 2ϑ2
rooms and duct segments. In order to visualize the dynamics of
If the measurement happens at a location where the air is unim-
the system and to have a test-bed for the controller, the system
peded, we can convert the equation to
is simulated in a time-continuous simulation tool, for which the
authors chose Simulink. To do so the differential equation system 
pbefore = pafter + (10)
need to be solved for V̈ 1 , V̈ 2 , V̈ 3 and simulated using an S-function1 2ϑ2
for the system. The resistance of a VAV box is calculated using the and in the following to
characteristics taken from literature, as explained above. Its main 
parameter is the opening angle of the internal flap, which is also 2 (pbefore − pafter )
the main actuator for the control system. ϑ= (11)
␳
There are some parameters, which need to be calculated every
time-step, for example the hydraulic resistance R. This value is cal- By simplifying pbefore − pafter to p and assuming  = 1, 19kg ⁄m3
culated by using the parameter Reynolds number, which depends (for air with a temperature at T=293 K) it is possible to calculate the
on the current volume-flow V̇ (and on the roughness and duct velocity. The assumption is valid as shown in Fig. 5, since the density
diameter, which are both constants). For the calculation, the of air barely changes at constant temperature in the pressure ranges
volume-flow from the previous simulation step is used, which leads that are relevant for ventilation systems (up to 500 Pa).
A temperature change would have a bigger influence to the den-
sity  depending on the converted formula of ideal gases
1
A description of a Simulink block written in C, C++ or Fortran that is compiled
and allows interacting with the Simulink engine in a programming language other p
= , (12)
than Simulink. RT
 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389 383

Table 1
Air exchange rates for windows.

Window position Air exchange rate [1/h]

closed, class 3 0,2–0,4

closed, class 2 0,5–0,8
Tilted 3–10
opened 10–20

Table 2
Tidal volume of people depending on the age and the activity.

Age [Years] Rest [l/h] Less activity Average Intensive

[l/h] activity [l/h] activity [l/h]

>14 22 43 85 152

Fig. 5. Ratio between density and pressure. V is the air volume of the room, ĊCo2 the CO2 production per hour
of the persons, Ca the outside CO2 level, Ci the inside CO2 level and
n the air change rate through windows and the ventilation system.
with the specific gas constant R = 287, 058J/(kg K) and the tem- The air exchange rate is calculated by
perature T.
Due to additional sensory and communication equipment, a V̇
n = nW + (14)
VAV-box is more expensive than a motor flap. Therefore it is V
possible to reduce the costs of a ventilation system by replacing The air exchange rate for windows nW is estimated as shown
VAV-boxes with motor flaps. The implication is that the volume in Table 1. These values represent a summary from standards and
flow at the VAV outlet is no longer known and therefore cannot be literature that were surveyed [20–23].
used for control purposes. The approach presented in this paper fills The exchange rate depending on the volume flow V̇ is based
the gap of missing sensory equipment by using the pressure drop on the ratio between the volume flow into the room and the total
model of the duct system to derive the according volume flow into one of the room. The current CO2 level, which is needed for the
each room, which is done by calculating the pressure drop from the calculation, was taken from the National Oceanic and Atmospheric
fan to the rooms as shown in Section 4.1. Administration (NOAA), [24].
Every person produces about 4000 ppm CO2 per breath in rest
4.3. Room model for CO2 concentration [25], the amount of exhaled air depends on age and the activity and
can be estimated according to Table 2. Thus it is possible to approx-
The previous sections defined the plant under control up to the imate the concentration change rate ĊCo2 given that the number of
inlet into the room. To complete the plant, it is necessary to model people in the room is known and derive the current CO2 level. The
the CO2 level in a room depending on the incoming volume flow. number of people may be derived from mechanisms such as people
This is done under the assumption that supply and exhaust volume counters, calendars in case of meeting rooms or lectures timetables
flow are identical, since there are no significant pressure changes in in case of universities. Based on these variables, it is possible to pre-
the building. The CO2 room model needs to be initialized with the dict the indoor air quality at time step t and calculate the optimal
basic room geometry to derive the total air volume. Every room is volume flow needed to maintain the CO2 level for every room. Since
assumed to have 395 ppm CO2 as a typical outside air concentration the number of persons, the activities and the window air exchange
at the beginning of the simulation. rate are only estimates, the calculation is valid only with a cer-
The CO2 room model is used for controls and therefore is subject tain confidence. However, the ventilation system is operated in a
to the following simplifications: closed control loop, which tolerates plant-model mismatches and
uses the controller design shown in Section 5 to compensate the
model errors.
• The CO2 level of the room is stable. That means that the gas
doesn’t coalesce with other gases. 5. Controller design
• CO2 doesn’t adhere at any furnishings, walls, etc.
• The CO2 level is evenly distributed throughout the room, i. e. the
Conventional ventilation control strategies set a constant pres-
room is a well-mixed and mechanically-ventilated space sure difference at the supply air fan and control the CO2 level using
the volume flow controllers. The volume flow controller closes the
Since a controller cannot change the physical setup of the air motor flap in order to provide the required amount of volume flow
supply system, it can only attempt to control the plant using the to a room. In situations with low volume flow demand, most of
given sensor information. It is therefore assumed that the design of the volume flow controllers may be closed, thus creating a high
the ventilation system and the sensor placement ensures that with pressure drop at the motor flaps. The supply air fan then has to
the given equipment the air quality can be maintained throughout work against this pressure drop, although the pressure is not nec-
the room. essary, which results in unnecessary electricity demand for the fan.
The simplifications allow to calculate with a zero-dimensional As shown in Section 3, SPR, Trim-and-Respond and other extended
mass balance using CO2 sources and sinks. The only source is the control strategies address this problem, but require costly volume
number of persons in the room. Sinks are given by the exchange flow controllers. The control design in this work approaches the
with outside air through windows and the volume flow of the ven- problem by utilizing the plant models for duct system, flaps and
tilation system. The resulting mass balance is: CO2 -level in order to improve energy efficiency and air quality and
at the same time make way for an implementation with reduced
∂Ci component and installation costs by replacing volume flow con-
V = ĊCo2 + (Ca − Ci ) nV (13)
∂t trollers with motor flaps.
384 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389

Fig. 6. Schematic of the ventilation controller.

For a given building it is first of all necessary to design and as shown in Section 4.1. The result is the identification of the
parameterize the pressure drop model of the duct system and the highest pressure drop and therefore the needed fan pressure. The
CO2 models of the rooms. Reference values for average volume volume-flow set-point calculation is based on the CO2 room model
flows are taken from literature (see Section 4). For a given volume described in Section 4.3, using the CO2 production and air exchange
flow set-point it is then possible to calculate the pressure drop in to derive the needed volume. The calculated set-points are the
the duct system from the supply fan to the room outlet of each room inputs for the Pressure controller and the Flap controller, respec-
(see Section 4.1). This is done for each path from fan to every room tively. The pressure controller is configured using the measurement
outlet, resulting in a list of pressure drops. The highest of these data that identifies the fan characteristics i. e. the correlation
pressure drops defines the maximum pressure the fan has to pro- between pressure, volume flow and rotational speed. The flap con-
vide, and it will then provide the sum of all volume flows that the troller is actually a set of PID-controllers, one for each room that is
VAVs request. This means that one of the VAV controllers will have controlled by a VAV controller or by directly controlling the motor
its flap fully open (the one with the highest pressure drop), while flaps.
the other VAVs will react to the given pressure drop by closing their
flap so that the requested amount of volume flow is provided for 6. Validation and results
the according room. This procedure is similar to the SPR and Trim-
and-Respond strategies. However, if the characteristics of the flap The presented approach consists of models for the ventilation
are added to the plant model, the VAV boxes can be replaced with system and its components, the CO2 room model and the newly
motor flaps as argued in Section 2, without affecting the controller developed ventilation controller. The validation of the complete
design structure. approach focuses on an existing demonstration system that is used
The approach depends on the knowledge of the volume flow as a data source for operation data and as a template for the con-
set-points for each room, these are derived using the CO2 room troller design. The most critical control actuator, the motor flap
model and an estimated occupancy, which can be taken from a inside the VAV box (or as a stand-alone component), is validated
room booking schedule or from the design values for room occu- separately using the recorded operation data. Finally, the newly
pancy. The volume flow set-point for each room is calculated by designed controller is compared against an advance state-of-the-
means of a linear PI-controller that uses the current CO2 levels and art controller.
the CO2 set-points as inputs.
There is an additional caveat that the ventilation controller has 6.1. Demonstration system
to regard: the first is that the calculated volume flow will most
likely not be sufficient for supply. A real-world ventilation system The ventilation system, which is used to validate the models
is not airtight and may have suffered leaks during construction or supplies lecture halls, IT labs, seminar rooms, offices and the cor-
due to aging (e.g. installation faults causing leakages). Similarly, ridors on two floors in a passive house office building in Vienna,
the demonstration system (Section 6.1) showed to contain outlets Austria. The provisioning system consists of the supply air fan, a
without controlled airflow, which results in losses of volume flow heating and a cooling register. The differential pressure is measured
that is unknown, since it could only be measured with significant downstream of the heating register; therefore the pressure drops in
effort. This implies that the controller has to provide more vol- the provisioning components are not relevant, since the supply air
ume flow than the model calculations require. Due to the nature fan controller maintains the pressure difference between outside
of building design, these factors cannot and shall not be measured air pressure and the measurement downstream of the provisioning
(for cost reasons), therefore the controller design has to foresee a system.
safety margin to ensure proper air supply. The volume flow into the entrance hall and the offices is fixed by
Fig. 6 shows the controller schematic with the measured val- one constant volume flow controller with a volume flow set-point
ues (suffix meas) and the set-points (suffix sp). The calculations of 580 m3 /h for the entrance hall and a second controller with a
described above are located in the Pressure set-point calculation and set-point of 600 m3 /h for the volume flow for all offices. The vol-
the Volume flow set-point calculation block. Necessary (measured) ume flow into the corridors is neither controlled nor measured,
input values are the CO2 level CO2 , the volume flow dV, the pressure which is a common design decision based on lower costs for sim-
p and the flap position (or flap angle, respectively) ˛ (alpha). ple mechanical outlets in the corridors. The IT labs, lecture halls
In a first step the necessary set-points are calculated: the pres- and seminar rooms are equipped with Variable Air Volume (VAV)
sure set-point uses the pressure drop calculation of the duct system controllers that receive set-points for volume flows from the build-
 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389 385

ing management system. The VAV controllers do not communicate

between each other or the supply fan. In the scope of this paper
a subset of the demonstration system with a topology shown in
Fig. 4 provides the data needed for plant modeling (see Section 6.4
for the detailed parameter specification). In the following sections
the pressure drop model of the ventilation ducts and the motor
flaps are the key components that have to be validated.

6.2. Validation of pressure drop model

The pressure drop model of the ventilation duct system that is

described in Section 4.1 has to be validated by comparing the model
Fig. 7. Validation of the pressure drop model.
data with experimental data, this is, against the behavior of the
demonstration system under real conditions of use. The recorded
operation data includes the current volume flow into each room, pressure sensors at the essential parts of the ventilation system,
the volume flow set-point for each room, the volume flow of the which could not be argued cost-wise. The model returns confident
supply fan and the provided pressure of the supply fan. In a first results about the pressure in the duct system that can be used in a
step the pressure drop model for the duct system was created, using real world environment. Possible violations of the comfort zone by
the available schematics to determine duct diameter, shape, length reason of potential plant-model mismatches can be compensated
and geometry (including 90◦ turn, bifurcation and other geome- by the closed-loop controller. This kind of controller feedbacks the
tries that affect the friction in the ducts). This model reflects the current CO2 level, compares it with the CO2 set-point and tries to
as-planned state of the duct topology and was implemented in compensate this difference.
MATLABTM .
The pressure drop model receives the measured volume flow 6.3. Flap identification
into each room as input data. As an output, the model calculates all
pressure drops for all segments in the duct topology and allows to In a ventilation system the motor flap located at the room inlet
determine the pressure drop between supply fan and every room is the main actuator for controlling the volume flow into a room.
outlet, including the total pressure drop that is needed to properly Using a given pressure in the duct system it is the task of the flap to
provide the required volume flows. Fig. 7 shows the procedure for reduce this pressure so that the required volume enters the room.
comparison and validation between the total differential pressure The remaining pressure difference is taken by the flap. As shown
from the measured system and the calculated pressures for every earlier, an efficient controller shall keep this pressure difference
room. low, in order to operate energy efficiently. Modeling of the char-
Fig. 8 shows the recorded operation data. The purple line shows acteristics is a crucial part for both VAV controllers and direct flap
the differential pressure directly at the fan i.e. the intensity how control. However, as the flap opening is usually not known, since it
much the fan pressurizes the duct system. The other lines show is not recorded by the building automation system which controls
the output of the pressure drop model including the flap model. the ventilation system. Therefore, other methods have to be inves-
The validation procedure shows relative good agreement to the tigated to fulfil the requirements for the overall modeling quality.
measured total differential pressure. While ideally measurements Therefore, the VAV box (or the flap inside the VAV box, respectively)
and simulation should match perfectly, this would require signif- characteristics are described on the basis of three major parame-
icant efforts in measuring the duct system, providing differential ters, which are the (actual) volume flow rate through the VAV box

Fig. 8. Operation data of the pressure in the duct system.

386 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389

teristics and was therefore used for further processing. Using curve
fitting algorithms led to the following linear equations:

˛ = −634.6 ∗ V̇ + 92.88 (15)

p = 1.112 ∗ ˛ + 109.9 (16)

The linear approximation also allows the transformation of the

above equation in case a flap angle is known at a certain pressure
difference and the volume flow shall be validated. For volume flow
rates that tend towards zero, logic is implemented to constrain Eq.
(15) such that the angle is limited to 90◦ . The same is done for
very high volume flow rates, meaning that the flap is fully open,
therefore making ␣ = 0◦ .
This flap characteristic completes the pressure drop model of
the duct topology and allows having a complete description of the
controlled plant from the supply fan to the room outlet.

6.4. Comparison of ventilation controllers

Out of the different air quality controllers in Section 3 the

Fig. 9. Characterization of the VAV flap with regards to the flap angle and the volume
authors have selected the Trim-and-Respond controller as a base-
flow rate.
line for comparison with the controller presented here. It is more
advanced than the Constant Static Pressure (CSP) and thus provides
a fair baseline for the work in this paper. The Trim-and-Respond
reduces air pressure until the first VAV controller has an opening
angle of 90%; this is the trigger to increase the pressure again to pre-
vent under-supply. As stated earlier, rapid changes of the IACQ are
compensated only with a delay before the requested volume flow
is provided. The comparison of the Trim-and-Respond controller
and the new controller described in Section 5 had to be done in a
simulated setup in order to eliminate the disturbances described
earlier (leakages, occupancy, window air exchange rate), which
would have provided results with high noise and potential biases.
The noisy data result from the setup in an operational building and
would require to collect a large sample of data to make statistically
significant statements on the improvements, which again requires
a tight integration into the building automation system that needs
to be run in closed-loop operation with an (experimental) innova-
tive controller. This could not be established within the scope of
this work, therefore a simulation setup had to be chosen, which
matches a section of the demonstration system described in Sec-
tion 6.1. This setup consists of three rooms, as shown in Fig. 4, two
Fig. 10. Characterization of the VAV flap with regards to the pressure drop and the
offices (both of size 6 × 4 × 2.8 m) and one conference room (size
volume flow rate. 4 × 4 × 2.8 m). The maximum volume flow for air supply is designed
to be 800 m3 /h, the occupancy profiles were selected to be constant
during office hours with two people in the offices and five people
(flap) unit, the pressure drop across the VAV (flap) unit and the in the conference room.
flap opening which is expressed in terms of an opening angle (in As shown in Fig. 11 the set-point of the CO2 level (800 ppm)
degrees, with 90◦ being fully closed). As only the volume flow rate is can be reached at about time t = 500 s (before that time there is no
recorded by the building automation system, the parameters pres- need for the controllers to act, since the CO2 level is below the set-
sure drop and flap angle need to be calculated. For that reason a point) by both the baseline Trim-and-Respond controller and the
hybrid composition of a physical pressure drop model as shown in controller developed in this work with about the same dynamics
Section 4.1 and the available monitoring data served as the basis and accuracy. The CO2 level of room 3 rises slower due to the rel-
for the identification of the VAV’s flaps. atively lower occupancy (i.e. less people per cubic meter air) and
Figs. 9 and 10 show the validation of the flap characteristics starts closing as soon as a volume flow in the duct system is recog-
using the mathematical model and overlaying it with measurement nized so that the overshoot is reduced (shown as a knee time t = 600
data. Apart from the linear approximation, we also performed curve for room 1 and 2).
fitting using exponential functions and multiple order polynomials. With regard to indoor air quality Fig. 12 shows that the newly
However, these did not show any improvements compared to the developed controller reaches the comfort zone with comparable
linear fit. It is obvious that the measured data have a high ratio of dynamics, but achieves this with lower power consumption: Fig. 12
noise, which is due to both dynamic effects of laminar air flow as shows the volume flow into the three rooms, indicating that Trim-
well as design constraints of the duct system, which also explains and-Respond needs more volume flow while having about the
e.g. the plateau in Fig. 10 between 10◦ and 40◦ . A further analysis same settling time. Additionally, the Trim-and-Respond controller
of the dynamic effects would better explain the measured data, but requires a higher fan pressure. Therefore, the built-in flaps need to
would require additional effort in modeling and instrumentation. close much more to compensate the overpressure. This happens,
Therefore, a linear model was selected to approximate the charac- because the Trim-and-Respond controller operates with a fixed
 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389 387

Fig. 11. CO2 Level in three rooms.

pressure setting that is changed only after long time periods to Fig. 13 shows the differences in supply pressure between the
avoid oscillations in the duct system. Only when the fan begins two controllers, resulting from the volume flow demands of the
to lower the pressure the power consumption reduces. The newly three rooms that need to be provided by the supply fan. The new
developed controller, on the other side, communicates between pressure controller for the supply fan is robustly designed, so that
the fan and the flaps immediately. Because of this improvement, short-term changes do not affect and destabilize the system. There-
the pressure can be kept quite low, so the power consumption fore, the controller reacts a bit slower than the Trim-and-Respond
is lower than by using the Trim-and-Respond controller and the controller, but with a much lower pressure peek. Since both con-
system operates smoother. trollers maintain a similar level of air quality (Fig. 11) it is valid
388 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389

Fig. 12. Volume flow comparison in three rooms.

Fig. 13. Comparison of pressure difference.

to compare the energy saving potential of the new approach: by controller is capable of operating either VAV boxes or motor flaps,
comparing the power consumption between the newly developed it only requires the additional flap characteristics, which have been
and the Trim-and-Respond controller an energy saving of about 5% identified using operation data.
could be identified due to the tighter link between volume flow and With regard to cost savings an estimate has been made for a
required supply fan pressure. reference office building: the building was built in 2015 according
to passive house standard, has optimized daylight usage as well
7. Conclusion and outlook as concrete core activation with a gross floor area of 3100 m2 . The
ventilation system uses a total of 100 VAV boxes of different sizes
The work presented in this paper is driven by the idea of replac- from size DN100 to the largest being 500 × 200 mm. Common mar-
ing hardware and investment costs with advanced algorithms that ket prices indicate investment cost saving from VAV controller to
can be implemented in software. A dynamic pressure drop model motor flap of 55%, which results in a total investment cost saving
for the duct ventilation system and its components was developed of about 11,000 EUR; other costs for operation and maintenance
together with a controller that increases energy efficiency by cou- remain approximately the same. This defines the margin how much
pling supply pressure and room volume flows more tightly. The the additional controller software may cost per building; addition-
 G. Zucker et al. / Energy and Buildings 155 (2017) 378–389 389

ally the presented controller improves energy efficiency by another [2] L. Pérez-Lombard, J. Ortiz, C. Pout, A review on buildings energy consumption
5%, which can be significant, seeing that the ventilation system is information, Energy Build. 40 (3) (2008) 394–398.
[3] Commercial Buildings Energy Consumption Survey (CBECS) – U.S. Energy
one of the main energy consumers in a building. Information Administration (EIA). [Online]. Available: https://www.eia.gov/
The approach requires significant modeling effort, which is not consumption/commercial/estimation-enduse-consumption.php (Accessed:
feasible for practical implementation. Also, the validation of the 14-Jul-2017).
[4] M.F. Silva, S. Maas, H.A. de Souza, A.P. Gomes, Post-occupancy evaluation of
controlled plant is difficult due to its multiple disturbances (includ- residential buildings in Luxembourg with centralized and decentralized
ing deviations from as-planned versus as-built system, mismatches ventilation systems, focusing on indoor air quality (IAQ). Assessment by
in the flap model and the CO2 room model). Although the controller questionnaires and physical measurements, Energy Build. 148 (2017)
119–127.
is capable of compensating for these plant-model mismatches,
[5] C.-A. Roulet, Santé et qualité de l’environnement intérieur dans les bâtiments,
since it operates in closed-loop, improvements need to be made. Presses Polytechniques et Univ. Romandes, Lausanne, 2004.
The main modeling effort affects the pressure drop model. Although [6] Z. Lin, T.T. Chow, K.F. Fong, C.F. Tsang, Q. Wang, Comparison of performances
of displacement and mixing ventilations. Part II: indoor air quality, Int. J.
there is some commercial software, it may be difficult to combine
Refrig. 28 (2) (2005) 288–305.
with some controller software, since the tools are mainly used for [7] C.S. Koulani, C.A. Hviid, S. Terkildsen, Optimized damper control of pressure
planning and not for operation. and airflow in ventilation systems, 10th Nordic Symposium on Building
A promising approach is grey box modeling, where the model Physics (2014) 822–829.
[8] S.T. Taylor, Increasing efficiency with VAV system static pressure setpoint
is created by its topology and provided with the start values taken reset, ASHRAE J. 49 (6) (2007) 24.
from literature as shown in this paper, but is calibrated with oper- [9] Standing Standard Project Committee (SSPC), ASHRAE Standard 62–2001,
ation data in order to achieve an optimal plant-model fit. These Ventilation for Acceptable Indoor Air Quality, 2001.
[10] Belimo, Belimo VAV compact, 05-Jul-2016. [Online]. Available: http://www.
obtained models can be used as local pressure drop models for each belimo.ch/pdf/d/LE EU-CH DE-CH VAV-Compact.pdf [Accessed: 05-Jul-2016].
room/section in an office building. Thus, the determination of the [11] X. Lin, J. Lau, Demand controlled ventilation for multiple zone HVAC systems:
highest pressure drop is possible and the ventilators pressure set- CO2-based dynamic reset (RP 1547), HVAC&R Res. 20 (8) (2014) 875–888.
[12] R.J. Garde, Fluid Mechanics Through Problems, New Age International, 1997.
point can be kept minimal. Thereby, the high modeling effort is [13] M.C. Potter, D.C. Wiggert, B.H. Ramadan, Mechanics of Fluids, si edition,
reduced and different errors, like leakages, are compensated. This Cengage Learning, 2016.
procedure can then be executed regularly in order to address aging, [14] W.-P. Sung, J.C.M. Kao, R. Chen, Frontiers of Energy and Environmental
Engineering, CRC Press, 2012.
leaks or refurbishments.
[15] D.I.H. Barr, H.W., (Firm), Tables for the Hydraulic Design of Pipes, Sewers and
Since the way of planning buildings is currently redefined using Channels Thomas Telford (2006).
digital integrated planning by means of the Building Information [16] W.S. Janna, Design of Fluid Thermal Systems – SI Version, Cengage Learning,
2010.
Model (BIM), the modeling effort for duct system and components
[17] I.E. Idelchik, E. Fried, Handbook of Hydraulic Resistance, 1986.
will potentially be reduced significantly, once this data is available [18] A.W.W. Association, B.E. Bosserman, A. Ali, I.M. Schuraytz, Butterfly Valves:
from the planning phase and can directly be used in operation. This Torque, Head Loss, and Cavitation Analysis, American Water Works
may go as far as extracting the ventilation system from a BIM and Association, 2012.
[19] B.G. Liptak, Instrument Engineers’ Handbook: Process Control, Vol. 2, Third
automatically deriving the optimal control strategy based on the Edition, CRC Press, 1995.
topology and the indoor comfort parameters. [20] Fenster und Türen -Luftdurchlässigkeit -Klassifizierung. DIN EN 12207,
The BIM process has already penetrated the planning phase, (2000).
[21] EnEV Anforderungen an die Dichtheit und den Mindestluftwechsel,
but has not yet arrived in the operation phase; however, stan- 07-Jul-2015. [Online]. Available: http://www.enev-online.org/enev 2009
dardization activities are taking place, for example, in the Austrian volltext/enev 2009 anlage 04 anforderungen dichtheit mindestluftwechsel.
Standards Committee 011.09 and it is expected that in the near htm.
[22] A. Maas, Experimentelle Quantifizierung des Luftwechsels bei Fensterlüftung,
future the necessary building data will be available for equipment Kassel.
operation. [23] J. Burkhardt, Deutschland, Daten zur Umwelt: der Zustand der Umwelt in
Deutschland. 7. 2000, E. Schmidt, Berlin, 2001.
[24] N. US Department of Commerce, ESRL Global Monitoring Division − Global
Acknowledgement Greenhouse Gas Reference Network. [Online] Available: https://www.esrl.
noaa.gov/gmd/ccgg/trends/global.html [(Accessed: 07-Feb-2017).
This work is partly funded by the Austrian Research Promo- [25] Ausschuss für Umwelthygiene, Standards zur Expositionsabschätzung
Hamburg, 1995.
tion Agency (FFG) within project ARIS (grant agreement number
845143).

References

[1] D. Bosseboeuf, Energy Efficiency Trends and Policies in the Household and
Tertiary Sectors – An Analysis Based on the ODYSSEE and MURE Databases,
2015.