See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/244995556

PID controller tuning and implementation aspects for building thermal

control

Article in WSEAS TRANSACTIONS ON CIRCUITS AND SYSTEMS · July 2006

CITATIONS READS

3 212

6 authors, including:

Panagiotis Patelis G. Stavrakakis

Pedagogical and Technological Education (ASPAITE) Technical University of Crete
2 PUBLICATIONS 5 CITATIONS 229 PUBLICATIONS 3,950 CITATIONS

SEE PROFILE SEE PROFILE

Denia Kolokotsa Kostas Kalaitzakis

Technical University of Crete Technical University of Crete
188 PUBLICATIONS 10,410 CITATIONS 84 PUBLICATIONS 7,240 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by G. Stavrakakis on 23 January 2018.

The user has requested enhancement of the downloaded file.

Proceedings of the 10th WSEAS International Conference on CIRCUITS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp308-313)

PID Controller tuning and implementation aspects for building

thermal control
KAFETZIS G. *, PATELIS P. *, TRIPOLITAKIS E.J. *, STAVRAKAKIS G.S. *,
KOLOKOTSA D. **, KALAITZAKIS K. *
* Department of Electronic & Computer Engineering
Technical University of Crete
Technical University of Crete, Kounoupidiana Campus, 73100, Chania, Crete
Greece
** Department of Natural Resources and Environment Laboratory of Renewable Energy
Resources Engineering
Technological Educational Institute of Crete
3, Romanou Str, 73100, Chania, Crete
Greece
http://www.elci.tuc.gr

Abstract: - The control of the indoor environmental parameters of buildings is an open problem with many
practical difficulties which stem from the non-linear, multivariable nature of the building models. This paper
presents the tuning process of a conventional PID controller for the thermal comfort control application, in
buildings utilizing fieldbus systems. The PID controller has been chosen taking into account its widespread
deployment and its attractive features like low computational cost and simplicity of implementation. On the
contrary, there is no standard methodology for the extraction of its parameters. Therefore, empirical methods
should be utilized along with trial and error techniques on the target application. The design procedure is
illustrated and there is special focus regarding the practical aspects of the implementation on a laboratory
installation. Finally, the results of the monitored and computed variables of the experimental procedure are
presented along with relevant conclusions and discussion.

Key-Words: - PID controller tuning, process reaction curve method, thermal controller

1. Introduction [1] proposed the transformation of H∞ loop-shaping

One of the key design goals for building energy tuning rules, to discrete-time tuning rules, in order to
management systems (BEMS) is the reduction and implement an adaptive PI controller for HVAC
minimization of the energy consumption. Moreover, (Heating Ventilation and Air Conditioning Systems).
along with the power saving goals, the indoor Wang, Lee, Fung, Bi and Zhang [2] suggested a PID
environmental parameters should be adjusted controller that presents high performance for a range
properly in order to ensure comfort conditions. of linear self-regulating processes, along with
Therefore, a control algorithm is necessary to laboratory experimental results. Regarding
provide a balance between comfort and low energy intelligent control, Tripolitakis, Kolokotsa et al [3]
consumption. suggested a Fuzzy PD controller for thermal comfort
Despite the scientific progress on the areas of in buildings and Moshiri and Rashidi [4] proposed
building energy management systems and building an adaptive self-tuning fuzzy PID controller for the
networking applications, the design of an optimal control of nonlinear HVAC systems. Kolokotsa et al
control algorithm presents increased difficulty. This [5] introduced a genetic algorithms optimized fuzzy
is due to the nonlinear characteristics of the building controller for the indoor environmental management
model. Moreover, it is difficult to efficiently use a set of buildings. Huang and Lam [6], illustrated a
of linear state-space equations to describe this family genetic controller approach for the tuning of a
of systems, as the existing ones present poor conventional PID controller rules.
performance and lack generalization. This paper describes the design, tuning and
Several approaches have been proposed in application of a digital PID controller for the
order to overcome the nonlinearity of the building’s adjustment of indoor conditions in the laboratory of
model. These include classic control approaches and Electric Circuits and Renewable Energy Sources at
intelligent system applications. Qu and Zaheeruddin the Technical University of Crete. The PID
Proceedings of the 10th WSEAS International Conference on CIRCUITS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp308-313)

controller is chosen due to its simplicity, low

computational overhead and exceptional behaviour to
a wide range of applications. The controller
developed in this work, is implemented on a personal
computer which communicated with the sensors and
actuators using the European Installation Bus (EIB) Fig. 1, Block diagram of plant with a PID controller
infrastructure. The system inputs are received from a An equivalent transfer function is given below:
set of sensors which monitor the indoor and outdoor
conditions. Such sensors are indoor and outdoor
U ( s) ⎡ T⎤
temperature, relative humidity, mean radiant Gc ( s) = = K c ⎢1 + sTd + i ⎥ (2)
temperature (MRT) and air speed. The output of the E ( s) ⎣ s⎦
system controls two air-conditioning units of 36,000 Kp Kd
btu each, used for both heating and cooling. The where Ti = is the integral constant, Td =
design objective is to adjust the indoor temperature to Ki Kp
a desired set value, on the presence of external is the derivative constant and K c ≈ K p . The above
disturbances. Moreover, the resulting system should
transfer function can be transformed on the
be able to compensate for the sensor inaccuracies,
continuous-time domain giving the following
present fast convergence to the set point, minimum
equation:
oscillation amplitude and minimum overshoot or
undershoot. ⎡ 1
t
de(t ) ⎤
u (t ) = K c ⎢e(t ) + ∫ e(τ )dτ + Td ⎥ (3)
Section 2 presents a brief introduction to the
⎣ Ti 0 dt ⎦
classic PID controller. Section 3 deals with the
description of the Ziegler–Nichols methodology for If the sampling time To has low value this equation
the controller’s parameters tuning. On section 4, the can be transformed to a discrete-time differential
tuning process along with the experimental results of equation. The differential term (D) is replaced by a
the laboratory application of the controller, are first order expression of difference and the integral
presented. Finally, on the fifth section there is a short term (I) by a sum giving the equation on the
discussion about the contribution of this work. discrete-time domain:
⎡ T k−1 T ⎤
u(t) = Kc ⎢e(t) + o ∑e(i) + d (e(k) − e(k −1))⎥ (4)
2. Conventional PID Controller ⎣ Ti i=o To ⎦
The conventional PID controller is the most popular
controller in industry. Its wide deployment is
justified due to its effectiveness along with simplicity 2.2 Characteristics of the PID controller
of design and implementation. Despite its
advantages, the design of such a controller is not 2.2.1 P controller
described by a general methodology and focuses on The P part gain, is proportional to the input error
empirical methods. value. It offers limited performance as it is unable to
eliminate the steady state error. It is common to use
the expression, Proportional Band (PB), to
2.1 Analysis of the controller describe the action of the proportional gain. The
The PID controller consists of three parts: the 100%
equivalent equation is given by: PB[%] =
Proportional gain (P), the integral gain (I) and the Kp
differential gain (D). Their general transfer function (5).
is given by:
Ki K d s 2 + K p s + Ki 2.2.2 I controller
Gc (s ) = K p + + sK d or (1) The output of the integral part is proportional to the
s s
calculated sum of errors and demonstrates a way of
where the gains K p , K i and K d are the gains for slow reaction control. Moreover, it manages to
proportional, integral and derivative part eliminate the steady state error by generating
respectively. The general block diagram of a plant respective actuation outputs and at the same time
with a PID controller attached to it, is given on the works against the oscillatory behaviour of the plant.
following figure:
Proceedings of the 10th WSEAS International Conference on CIRCUITS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp308-313)

2.2.3 D controller Parameters

The derivative part gain is proportional to the rate of Controller Kc Ti Td
change of the control error. It leads to fast response
stable systems at the cost of steady state error
P T
introduction and oscillatory behaviour. Kτ d
PI 0.9T τd
3. PID Parameters Extraction Kτ d 0.3
The PID controller design objective is to determine a
PID 1.2T 2τ d 0.5 τ d
set of gains (or gains along with time constants) Kτ d
( K p , K i , K d ) or ( K c , Td , Ti ), in order to fulfill the Table 1 – Controllers’ Parameters
performance requirements. These gains are to be by Ziegler-Nichols method
calculated in such a way that the transient response,
It has to be noted that a major disadvantage of the
tolerance to disturbances and steady error
Ziegler-Nichols method, is the undesired effect of
specifications are met. In practice, it is not possible to
oscillatory behaviour on a set of systems.
achieve all of these requirements. Many researchers
have tried to overcome these difficulties and obtain a
global design method of PID controllers. As a result
many methods for tuning single-loop and multi-loop 4. Experimental Procedure
PID controllers have been proposed. Some of these
are: Ziegler-Nichols, Cohen and Coon, Ho-Hang-Cao
and Internal model control. In this paper Ziegler- 4.1 General
Nichols method is reviewed and used to calculate the The PID controller used, is applied for the control of
PID controller’s parameters. laboratory’s temperature and it has the basic
characteristics of PID controllers, as they were given
3.1 Ziegler-Nichols method in the section 2. Figure 3 shows the block diagram
In 1943 Ziegler and Nichols [7] proposed a method of PID controller and its transfer function is given
known as process reaction curve method. In this by eq. (4)
method, the open loop unit step response of the plant
(the controller is disconnected from the system) is
measured and approximated by small straight lines
usually given a form as shown in figure 2.

Tangent Process
Reaction
Curve Fig. 3, PID block diagram

The inputs of the controller are three: the current

Fig. 2, Plant unit step response
temperature error, the sum of all past temperature
The tangent, in the point of the curve where the first errors and the temperature error difference, ce(k ) .
derivative receives its maximum value is drawn The error is defined as the difference of the set-
(point of inflection). The τ d , T and K parameters are point, which is the desirable achieved temperature,
from the present measured temperature, e(k ) = set
calculated below:
Delay time τ d = t1 − t0 (6) point - measured temp. The set point was selected at
28 °C as it was considered to be a suitable
response time T = t 2 − t1 (7) temperature for the summer season. The sum of
y∞ − y0 errors is defined as the sum of all past temperature
and gain K = (8) errors plus the current error. The error difference is
u ∞ − u0
defined as the difference between the current error
and the previous error, ce(k ) = e(k ) − e(k − 1) .
The rules of Ziegler- Nichols method for the
extraction of PID parameters are as shown in Table
1.
Proceedings of the 10th WSEAS International Conference on CIRCUITS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp308-313)

4.2 Extraction of the PID parameters Ziegler-Nichols method

The Ziegler-Nichols method was used as described in 32

section 2, for the extraction of the K c , Ti and 30

Td parameters. At first, while the controller was 28

Temperature oC
disconnected from the system (open loop), a unit step 26
Tangent: y = -1.55x + 33.4

input was applied by setting the two air conditions to

operate constantly at full power for an amount of 24

time sufficient for the system to reach a steady state. 22

22.2 oC

Moreover, no other sort of HVAC control was

20
performed on the installation. The period in which 7.22
the measurements were taken was, at the end of July 10 20 30 40
Measurements (every 2 minutes)
50 60

2005. At the same time, temperature measurements

Fig. 5, Temperature curve with tangent at the point of
were received every 2 minutes. The time of 2 inflection
minutes was chosen so as to yield dense
measurements and provide good approximation of According to the method, the delay time τ d
the temperature curve. The curve export procedure
was applied for a number of days in order have a calculated value was 2 min, while the response time
large set of measurements. As it can be deducted T was 12.44 min. The response time ranges from τ d
from figure 3, the laboratory temperature stabilizes up to the moment where the tangent intersects the
slightly above 22 °C after 120 minutes. This happens steady-state part of the temperature curve,
because of the constant contribution of cooling to the corresponding to 22.2 °C. The gain K is defined by
laboratory space without any significant loss due to the difference of the steady-state temperature from
external disturbances. the initial one and is calculated as:

31 K = 30.75 − 22.2 = 8.55 (9)

30

29 Subsequently, the K c , Td , Ti values are computed

28 from the equations on table 1 where the τ d , T , and
Temperature oC

27
K are substituted by their corresponding calculated
26
values. The results are depicted on the table 2.
25

24 Kc Ti Td
23 P 0.7275
22
0 20 40 60 80 100 120 140
PI 0.6547 6.6666667
Measurements (every 2 minutes) PID 0.8729 4 1
Fig. 4, Temperature curve as function of time Table 2 – Controllers’ parameters
(each sample on x axis corresponds to a two minutes
interval) After the extraction of the PID parameters, the
equation 4 takes the following form:
As suggested by the Ziegler-Nichols method, the ⎡ T k −1
1 ⎤
point in which derivative is maximum was found and u (k ) = 0.8729 ⎢e(k ) + o ∑ e(i) + T (e(k ) − e(k − 1))⎥ (10)
⎣⎢ 4 i =o o ⎦⎥
the tangent at this point was drawn as shown in
The next step was the selection of the sampling time
figure 5.
T o to be 8 min. Thus, the final transfer function of
the controller is given by:
⎡ 8 k −1
1 ⎤
u(k ) = 0.8729⎢e(k ) + ∑ e(i) + 8 (e(k ) − e(k − 1))⎥ (11)
⎣⎢ 4 i =o ⎦⎥
Proceedings of the 10th WSEAS International Conference on CIRCUITS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp308-313)

4.3 Experimental results of the PID controller 1.8

PMV (Predicted Mean Vote)

application 1.6
PMV

The PID controller is applied for the adjustment of

1.4
indoor temperature. The variables which were
monitored and calculated were: internal temperature, 1.2

external temperature, MRT (Mean Radiant

PMV
1

Temperature), PMV (Predicted Mean Vote) [8] and 0.8

A/C duty cycle (controller output), defined by the

percentage of system cycle time, the units operate . 0.6

The measurements lasted two days, starting from 0.4

21/07/2005 11:52 and ended at 22/07/2005 15:48. 0.2

11:52 14:17 16:17 18:17 20:18 22:18 0:18 2:18 4;19 6:19 8:19 10:19 12:20 14:20 15:48
The sampling time was set to 8 minutes. The period Time (hour:minutes)

of air-conditioning ’ operation could receive a value Fig. 8, Predicted Mean Vote (PMV)
ranging from 0 to 8 minutes corresponding to a 100%
duty cycle. It has to be noted that during the 36
Internal Temperature and A/C duty cycle

measurement period the values of the variables may Internal Temp

Temperature (oC)
34 Set point 28oC

fluctuate. Figures 6, 7, 8 and 9 depict the 32

measurements of the monitored variables. 30

28

26
Internal Temperature 11:52 14:17 16:17 18:17 20:18 22:18 0:18 2:18 4:19 6:19 8:19 10:19 12:20 14:2015:48
35
Time (hour:minutes)
Int Temp
Set point 28oC
34 100
A/C Duty cy cle
A/C Duty Cycle (%)

80
33

60
Temperature (oC)

32
40

31
20
11:52 14:17 16:17 18:17 20:18 22:18 0:18 2:18 4:19 6:19 8:19 10:19 12:20 14:2015:48
Time (hour:minutes)
30

29
Fig. 9, Internal Temperature vs A/C duty cycle
28

27
11:52 14:17 16:17 18:17 20:18 22:18 0:18 2:18
Time (hour:minutes)
4:19 6:19 8:19 10:19 12:20 14:2015:48 5. Discussion
The experimental results verified that the controller
Fig. 6, Internal Temperature satisfies the original design objectives. As shown on
the figure 6, the system reaches the area near the set
Internal Temperature - External Temperature - MRT
35
Internal Temp
point in 48 minutes (6 cycles). The maximum
34 MRT
External Temp
undershoot is at 27.41 °C yielding a 2.1% difference
33 from the desired set value. The oscillation amplitude
32 doesn’t exceed 2.5% of the set value which is
acceptable for the application (less than 0.5 °C).
Temperature (oC)

31

30 This behavior is due to the type of control on the

29 cooling units. As previously stated, the cooling units
28 were controlled by turning them on for a proportion
27 of a desired period, set at 8 minutes. After they were
26 turned off and until the beginning of the new system
25
11:52 14:17 16:17 18:17 20:18 22:18 0:18 2:18 4:18 6:18 8:19 10:19 12:20 14:2015:48
cycle, the temperature would rise, thus yielding the
Time (hour:minutes)
oscillations. Moreover, the oscillatory behavior is
Fig. 7, Internal Temperature vs External Temperature and dependent on external disturbances such as open
MRT doors and windows and the variation of the external
temperature of course. An engineering tradeoff was
made on the choice of the system cycle (and the
sampling time subsequently). A very small value of
the system cycle reduces the oscillation amplitude
and yields a fast response, because of the
amplification of the derivative part gain. This
subjects the mechanical parts of the units to stressful
Proceedings of the 10th WSEAS International Conference on CIRCUITS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp308-313)

conditions. Moreover, the steady-state error is not to optimize controller parameters for HVAC
eliminated as the integral gain is reduced. The systems , Energy and Buildings, Vol. 3, Issue
average computed indoor temperature was calculated 26, 1997, pp.277–282
from the measurements to be 28.073 °C. The [7] J. B. Ziegler and N. B. Nichols, Optimum
standard deviation was 0.7354 °C. The internal settings for automatic controllers, Trans. ASME,
comfort is eventually preserved taking into account vol. 64, 1942, pp. 759–768
the PMV curve from figure 8, as its largest part [8] P.O.Fanger, Thermal comfort, Mc Graw-Hill,
resides on the (0 - 0.5) range which is ideal for the New York, 1972.
summer period.
This paper presented the experimental results driven
tuning process for a conventional PID controller for
indoor thermal comfort control application. The PID
controller parameters for the plant in study were
derived using the empirical process reaction curve
method suggested by Ziegler and Nichols. The
resulting controller was tested for a long amount of
time and was proven to be stable and robust.
Moreover, it reached the desired set point presenting
low oscillation amplitude and low undershooting.
Furthermore, machine stress and low energy
consumption were taken into account for the choice
of the system cycle. The results taken from the
laboratory installation validated the proper design
and functionality of the result system and provide a
stable basis for future research.

References:
[1] Qu G., Zaheeruddin M., Real-Time tuning of PI
controllers in HVAC systems, International
Journal of Energy Research (28) 2004, pp.1313-
1327
[2] Wang Q.G., Lee T.H., Fung H.W, Bi Q., Zhang
Yu, PID tuning for improved performance, IEEE
Transactions on Control Systems Technology,
Vol.7, Issue 4, July 1999, pp. 457-465
[3] Tripolitakis E.I., Kolokotsa D., Kalaitzakis K.,
Stavrakakis G., Study and Implementation of a
Fuzzy PD Thermal Comfort Controller for
Embedded Fieldbus Systems Applications,
WSEAS Trans. on Circuits and Systems, Vol. 3,
Issue 9, Nov. 2004, pp. 2051-2057
[4] Moshiri F., Rashidi F., Self-tuning Based Fuzzy
PID Controllers: Application to Control of
Nonlinear Multivariable Systems, Lecture Notes
in Computer Science, LNCS Springer - Verlag,
August 2004
[5] Kolokotsa D., Stavrakakis G.S., Kalaitzakis K.,
Agoris D., Genetic algorithms optimized fuzzy
controller for the indoor environmental
management in buildings implemented using
PLC and local operating networks, Engineering
Applications of Artificial Intelligence, Volume
15, Issue 5, September 2002, pp.417-428
[6] Huang W., Lam H.N., Using genetic algorithms

View publication stats