Improved Performance of Heat Exchanger using Multi-variable

Quantitative Feedback Theory based Controller

Aniket Roy1 , Sandipan Prasad Chakravarty2 and Prasanta Roy3
Department of Electrical Engineering, NIT Silchar

Abstract—This paper deals with a pre-compensated diagonal developed by Professor Issac M. Horowitz, is a transparent
quantitative feedback theory (QFT) based controller for a multi- framework which allows the designer to reshape the open loop
variable uncertain plant. A 2×2 heat exchanger having linearised frequency response on the Nichol’s chart to simultaneously
dynamics along with interval type of parametric uncertainty is
considered as the plant. The controlled plant is required to meet achieve multiple performance specifications irrespective of
a certain stability, sensitivity and robust tracking specifications. parametric uncertainty [6]. It provides a fair trade-off between
First pre-compensators are designed to make the plant a diago- the achievable performance and the loop gain (cost of feed-
nally dominant one. Then, diagonal controllers are designed to back) in the desired frequency range. Minimisation of the cost
make the pre-compensated plant meet the stability and sensitivity of feedback is one of the key advantages of QFT since a
specifications using QFT. Lastly the pre-filters are designed to
ensure the robust tracking requirement. The effectiveness of the higher loop gain leads to sensor noise amplification, actuator
proposed controller is tested through numerical simulation. The saturation and high frequency uncertainties [7]. However, ap-
results reveal that the proposed controllers successfully control plying QFT to a MIMO plant is far challenging than its SISO
the plant with minimum coupling and interaction. The compara- counterpart due to interactions involved in a multi-variable
tive study reveals that the proposed controllers outperforms some system [8]. This paper attempts to apply the theoretical aspects
of the existing controllers in literature.
of multi-variable-QFT to a 2 × 2 heat exchanger system to
Keywords: Multi-variable control, Quantitative feedback the- improve its performance.
ory, Heat exchanger
I. I NTRODUCTION B. Prime objectives
Control of an uncertain multiple-input multiple-output The prime objectives of this work are
(MIMO) system is a far challenging task compared to its • To design a pre-compensated diagonal QFT based robust
single-input single-output (SISO) counterpart [1]. This is be- controller for a multi-variable (heat exchanger) plant
cause of some properties of MIMO system which do not have a having structured parametric uncertainties.
direct scalar analogue such as loop interactions, directionality • To compare the performance of the proposed controller
[2] etc. Designers often face the challenge of designing a with the existing controllers for heat exchanger in litera-
robust controller to meet the desired specifications due to the ture.
high loop interactions involved in a MIMO system. A heat ex-
changer is a typical MIMO system where there is an exchange C. Literature survey
of heat between two fluids flowing through separate channels
or chambers. In some cases they may be in direct contact A brief literature survey is presented here to highlight some
[3]. The control problem in a heat exchanger plant basically of the major reported works on the different control strategies
involves the task of controlling the temperature of one of the used to control a heat exchanger.
fluids at a desired value while keeping the temperature of the A stochastic approximation technique is used for enhancing
other fluid constant. Maintaining a constant temperature of stability in the heat exchanger system using a neural network
the hot fluid ensures a fair transfer of heat to the cold fluid, based PID controller in [9]. A multi-loop cascaded con-
thus maintaining its temperature at a desired level. The heat trol structure using multi-objective optimisation, an adaptive
exchanger system is one of the potential areas of interest for proportional integral controller and a H∞ state feedback
researchers because of its widespread applications in power controller with bi-linear observer are proposed for a heat
generation, chemical plants, refrigeration, air conditioning, exchanger in [10], [11] and [12] respectively. An internal
space heating [4] etc. The process inside a heat exchanger plant model control based fractional order PID controller for a heat
is dictated by the laws of thermodynamics, resulting in non- exchanger system is presented in [13]. Two robust control
linearity and parametric uncertainties. This makes the control techniques are proposed to keep the exit temperature at a
problem more challenging [5]. These call for a robust control reference value in [14]. The advantage of using 2-degree
strategy. Quantitative feedback theory (QFT) based controller of freedom PID controller over 1-degree of freedom PID
is one of the potential choices. controller is illustrated in [15], where controller parameters
are obtained through optimisation using Genetic algorithm. A
A. A brief introduction to QFT dynamic model of heat exchanger system, suitable for efficient
QFT is a classical frequency domain framework to de- control system design in frequency domain, is developed in
sign robust controller for an uncertain plant. QFT technique, [16].
 II. D ESIGN OF M ULTI - VARIABLE QFT The pre-compensated diagonal QFT based controller is de-
A. Model of the heat exchanger signed in the next few subsections for the uncertain plant
described in (1)-(2) to meet the specifications described in
A typical heat exchanger system in food sterilisation process
(3)-(5).
is considered as the plant in this paper. The system includes a
counter flow heat exchanger, a tank with a secondary fluid, C. Choice of nominal plant and control structure
an electrical heater to maintain a constant temperature of To control a MIMO system through independent SISO
the secondary fluid, an open hydraulic circuit that brings the control loops the pairing of input and the output is to be
product into the heat exchanger, and a closed hydraulic circuit determined first. To proceed with the design a nominal plant
controlled by a variable vermiculation pump that provides the Pn (s) is arbitrarily chosen from the set of uncertain plants as
hot fluid to the heat exchanger system [17]. The inputs and given below:-
outputs of the plant are as below: Inputs:  
1) u1 =Heat flux produced by the electrical heater   6.9 9
2) u2 =The volumetric flow-rate of the hot fluid line p (s) pn12 (s) 3.9s + 1 5s + 1  (6)
Pn (s) = n11
 
=
Outputs: pn21 (s) pn22 (s)  3 4.9 
1) y1 =Temperature of the hot fluid 8s + 1 2.3s + 1
2) y2 =Temperature of the product To compute the nominal interaction among the input-output
Assuming a constant volumetric flow-rate of the product line, variables, the relative gain array (RGA) of the nominal plant
the linearised dynamics of the plant is given below [17]. is found as follows [18].
      
y1 (s) p (s) p12 (s) u1 (s) 4.96 −3.96
= 11 (1) RGA{Pn } = Pn (0) ⊗ [Pn−1 (0)]T = (7)
y2 (s) p21 (s) p22 (s) u2 (s) −3.96 4.96
Kij where ⊗ indicates element-wise product of two matrices.
where pij (s) = ∀ i, j ∈ {1, 2}
τij s + 1
If the diagonal elements of RGA are greater than 1 and
with interval type of parametric uncertainty
off-diagonal elements are negative for a 2 × 2 system, the
K11 ∈ [6.1 6.9], K12 ∈ [8.1 9], K21 ∈ [2.1 3] recommended pairing of input-output is 1-1, 2-2. At the same
K22 ∈ [4.1 4.9], τ11 ∈ [3.1 3.9], τ12 ∈ [4.1 5], time it indicates that amount of coupling is very high [18].
τ21 ∈ [7.3 8], τ22 ∈ [1.4 2.3]. (2) Considering all these a control structure is chosen as shown
in Fig. I.
All units are considered as per unit of the base value as given
in [17].
B. Performance specifications
Performance specifications are assigned for the controlled
plant for both the loops as given in (3)-(5) considering gii (s),
fii (s) to be the controller and pre-filter of the ith control loop.
1) Stability specification:
pii (jω)gii (jω)
≤ 1.01, i ∈ {1, 2} (3)
1 + pii (jω)gii (jω)
∀ ω ∈= [0.001 2000] (rad/sec)
Fig. I. Block diagram of the proposed control structure
2) Output disturbance rejection specification:
α
1 jω The objective of the pre-compensators gij (s) is to transform
≤ , i ∈ {1, 2} (4)
1 + pii (jω)gii (jω) 1 + jω the plant into a diagonally dominant one. Then the controllers
β
∀ ω ∈ [0.001 5] (rad/sec) gii (s) and pre-filters fii (s) are designed to meet the specifi-
cations.
3) Reference tracking specification:
D. Design of pre-compensator
pii (jω)gii (jω)fii (jω)
δl (ω) ≤ ≤ δu (ω) (5) The pre-compensator matrix Gα (s) is computed as below:-
1 + pii (jω)gii (jω)
 α α

where, g (s) g12 (s)
Gα (s) = 11α α
1.44 g21 (s) g22 (s)
δl (ω) =   
0.4(jω)3 + 2.12(jω)2 + 5.6(jω) + 1.5 −1 p 11 (s) 0
= µ̄ P (s) (8)
2(jω) + 1.02 0 p22 (s)
δu (ω) =
0.8(jω)2 + 1.7(jω) + 1 where, the function µ̄ adjusts the dc gain of each element
i ∈ {1, 2} ∀ ω ∈ [0.001 5] (rad/sec) such that it is the closest approximation of the mean value of
the dc gain of all the possible uncertain elements. The pre- The stability bounds for the open loop transfer function
α β
compensators gij (s) are synthesised such that the frequency (OLTF): g11 (s)p̂11 (s) of the upper loop in Fig. I (henceforth
α
response (Bode plots) for each gij (s) matches with the mean called Loop I) is found using QFT toolbox [19] considering
value of the corresponding element in the RHS of (8) to the transformed nominal plant p̂n11 (s), such that specification in
closest approximation (in both magnitude and phase) over the (3) is satisfied irrespective of uncertainties in p̂11 (s). The
frequencies of interest. The synthesised Gα (s) is given below: stability bounds at the selected frequencies are shown in Fig.
  III.
1.174s + 0.161 − 1.459s − 0.182
s + 0.030 s + 0.023
 
Gα (s) =  − 0.138s − 0.099
 (9)
1.174s + 0.161 
s + 0.030 s + 0.030
The transformed plant after pre-compensating it by Gα (s) is
henceforth called pre-compensated plant P̂ (s) as below:-

P̂ (s) = P (s)Gα (s) (10)

The nominal plant of the pre-compensated plant and its RGA

are as below:-
 
1 0
P̂n (s) = Pn (s)Gα (s), RGA{P̂n (s)} = (11)
0 1

Clearly, the RGA of P̂n (s) is a unity matrix (I2×2 ) signifying

zero coupling in P̂n (s) ideally. This is hardly possible in
reality because of parametric uncertainty. However, the diago-
nal dominance has definitely improved making it suitable for
Fig. III. Stability bounds for Loop I for the frequencies shown in the legends
designing a diagonal QFT based controller which is discussed
in the next subsection.
The output disturbance rejection or sensitivity bounds for
β
E. Design of diagonal controller the OLTF g11 (s)p̂11 (s) of Loop I is found similarly such
The diagonal controllers giiβ
(s) are designed for the trans- that the specification in (4) is satisfied irrespective of the
formed diagonally dominant pre-compensated plant P̂ (s) to uncertainties in p̂11 (s). The sensitivity bounds at the selected
meet the performance specifications specified in (3) and (4). frequencies are shown in Fig. IV.
MATLAB QFT toolbox [19] is used for the entire design
β
procedure. Only the design of g11 (s) is presented here. Then
β
g22 (s) is designed similarly but not shown to avoid repetition.
The plant templates of the uncertain pre-compensated plant
p̂11 (s) on the Nichol’s chart at some selected frequencies over
the range of interest are shown in Fig. II.

Fig. IV. Sensitivity bounds for Loop I for the frequencies shown in the legends

The stability bounds and sensitivity bounds are then grouped

together to obtain the intersection bounds, which mark a
common allowable area in the Nichol’s chart in accordance
Fig. II. Templates for p̂11 (jω) at selected frequency with desired stability and sensitivity specifications.
 Fig. VI. Tracking bounds for Loop I for the frequencies shown in the legends

β
Fig. V. Loop shaping of controller g11 (s)

The loop shaping technique of QFT toolbox [19] is used to

β
synthesise g11 (s) as shown in Fig. V. The synthesised transfer
β
function of g11 (s) is:
β 25.642(s + 0.24)(s + 0.021)
g11 (s) = (12)
s(s + 10)(s + 0.099)
Similarly the bounds are found for Loop II (not shown to
β
avoid repetition) and the controller g22 (s) obtained after loop
shaping is:
β 13.816(s + 0.027)(s + 0.370)
g22 (s) = (13)
s(s + 10)(s + 0.004)
The fully populated matrix controller G(s) is composed of
two matrices: G(s) = Gα (s)Gβ (s),
 α α
 β 
g11 (s) g12 (s) g11 (s) 0 Fig. VII. The frequency response (magnitude) of pre-filter f11 (s) compen-
G(s) = α α β (14)
g21 (s) g22 (s) 0 g22 (s) sated closed loop system

The robust tracking bounds are not included during the

β
design of gii (s) since the tracking specifications are nearly
taken care of by the stability and the sensitivity bounds. If not
then the pre-filter will meet the same as discussed in the next
section.
F. Design of pre-filter
The pre-filter elements f11 (s) and f22 (s) are designed
for the closed loop system having OLTF P (s)Gα (s)Gβ (s).
The pre-filter design demands the computation of the robust
tracking bounds for the specification given in (5). The tracking
bounds are obtained using QFT toolbox [19] as shown in Fig.
VI. The pre-filter ensures that the magnitude of the frequency
response plots for the limiting plants always remain within
the limits as specified by the robust tracking specification.The
pre-filter shaping technique of QFT toolbox [19] is used for
designing the pre-filters: f11 (s) and f22 (s), as given below:
0.029(s + 9.310) 1.282(s + 0.172)
f11 (s) = , f22 (s) = Fig. VIII. The frequency response (magnitude) of pre-filter f22 (s) compen-
(s + 0.498) (s + 0.230)
sated closed loop system
(15)
The frequency response (magnitude) of pre-filter compensated minor deviation at t= 5 secs due to the residual loop interaction
closed loop systems are shown in Fig.VII and Fig.VIII which but most of it is counteracted by the control action, which can
verify that the robust tracking requirement is met for the be well observed from Fig. X. Overall the proposed controller
limiting plants. is successful in giving a satisfactory performance in terms of
robustness.
III. R ESULTS AND D ISCUSSION
A. Result analysis B. Comparative analysis with existing controllers
The performance of the controlled plant is validated through As the dynamics of the heat exchanger is highly dependent
numerical simulation using MATLAB Simulink. The system on its size and composition, hence it is difficult to present a
is subjected to step inputs (at t=1 sec for Loop I and at t=5 conclusive comparative study among the works on control of
secs for Loop II). Response of selected plants from the set of heat exchanger. An attempt is made here to draw a comparison
uncertain plants (including the limiting ones) is shown in Fig. in terms of settling time (ts ), peak overshoot (Mp ) and integral
IX. The corresponding control signals: u1 (s) and u2 (s) are Absolute error (IAE) with existing control strategies. In [20],
shown in Fig. X. out of the several proposed control strategies, the 2 degrees of
freedom fractional order PI controller gives the minimum IAE
as 57.424 units over an interval of 50 secs. The robust PID
controller design for the shell and tube heat exchanger based
on small gain theorem in [14], gives an integral absolute error
(IAE) of about 203 units. However, the minimum value of
IAE for this paper is 2.198 units in the same time window for
the similar references. This indicates that the proposed control
scheme gives better tracking performance than [20],[14]. In
[11], an adaptive PI controller is used to maintain the air tem-
perature output keeping the air-flow rate constant at different
set point values. When the set point is changed from 40◦ C
to 60◦ C, the controlled response has a ts and Mp of 80 secs
and 16.66%, respectively. In this paper, both ts and Mp have
lower values as shown in Section III-A. In totality, very few
numbers of existing literature have considered heat exchanger
as a MIMO system. Most of the literature have considered
Fig. IX. System response: y1 (s) [Top] and y2 (s) [bottom] heat exchanger as a SISO system, keeping one of the flows
uncontrolled, resulting in non-optimal operation.

IV. C ONCLUSION AND FUTURE SCOPE

In this paper, a pre-compensated diagonal quantitative feed-
back theory (QFT) based controller is used for controlling
an uncertain MIMO plant. The pre-compensators, diagonal
controllers and pre-filters are successfully designed. The com-
plexities of a MIMO system are dealt quite well while meeting
multiple performance specifications successfully. The result
analysis justifies the effectiveness of the proposed method.
A comparative analysis with the existing control strategies
is presented, which shows improved performance and also
additional advantages of the proposed controller.
However, pre-compensating the plant to make it diagonally
dominant may not always work, specially for open loop un-
stable system or system having transmission zeros in the RHS
of s-plane. This may lead to unstable pole-zero cancellation
causing the system internally unstable. The pre-compensation
Fig. X. Control effort u1 (s) and u2 (s)
may also be difficult in case the uncertainty is unstructured.
The loop I output, when subjected to a step input at t = 1 Further research can be carried out to overcome these issues.
sec, settles at 9 seconds with 0% overshoot while for loop II,
R EFERENCES
when subjected to a step input at t= 5 secs settles at 10 seconds
with 20% overshoot (which is within acceptable limits for a [1] D. Q. Mayne, “The design of linear multivariable sys-
MIMO heat exchanger plant). The loop I output experiences a tems,” Automatica, vol. 9, no. 2, pp. 201–207, 1973.
 [2] U. Shaked, I. M. Horowitz, and S. Golde, “Synthesis January 2017, pp. 1–7.
of multivariable, basically non-interacting systems with [16] H. Bastida, C. E. Ugalde-Loo, M. Abeysekera, X. Xu,
significant plant uncertainty,” Automatica, vol. 12, no. 1, and M. Qadrdan, “Dynamic modelling and control of
pp. 61–71, 1976. counter-flow heat exchangers for heating and cool-
[3] M. Awais and A. A. Bhuiyan, “Heat and mass transfer ing systems,” in 54th International Universities Power
for compact heat exchanger (CHXs) design: A state-of- Engineering Conference (UPEC), Bucharest, Romania,
the-art review,” International Journal of Heat and Mass September 2019, pp. 1–6.
Transfer, vol. 127, no. Part C, pp. 359–380, 2018. [17] M. Garcia-Sanz, Robust Control Engineering. New
[4] R. Shah, B. Thonon, and D. Benforado, “Opportunities York, US: CRC Press; Taylor & Francis Group, 2017.
for heat exchanger applications in environmental sys- [18] E. H. Bristol, “On a new measure of interaction for
tems,” Applied Thermal Engineering, vol. 20, no. 7, pp. multi-variable process control,” IEEE Transaction on
631–650, 2000. Automatic Control, vol. 11, no. 1, pp. 133–134, 1966.
[5] G. Florides and S. Kalogirou, “Ground heat exchangers-a [19] C. Borghesani, Y. Chait, and O. Yaniv, The QFT Fre-
review of systems,models and applications,” Renewable quency Domain Control Design Toolbox For Use with
Energy, vol. 32, no. 15, pp. 2461–2478, 2007. MATLAB Users Guide, Terasoft Inc., San Diego, USA,
[6] I. M. Horowitz and M. Sidi, “Synthesis of feedback 2003.
systems with large plant ignorance for prescribed time- [20] M. Jain, A. Rani, N. Pachauri, V. Singh, and A. P. Mittal,
domain tolerances,” International Journal of Control, “Design of fractional order 2-DOF PI controller for real-
vol. 16, no. 2, pp. 287–309, 1972. time control of heat flow experiment,” Engineering Sci-
[7] M. G. Sanz, Encyclopedia of Systems and Control. ence and Technology, an International Journal, vol. 22,
Cleveland, USA: Springer, October 2014. no. 1, pp. 215–228, 2014.
[8] I. M. Horowitz, “Quantitative synthesis of uncertain
multiple input-output feedback system,” International
Journal of Control, vol. 30, no. 1, pp. 81–106, 1979.
[9] C. Renotte, A. V. Wouwer, and M. Remy, “Neural
modeling and control of a heat exchanger based on SPSA
techniques,” in Proceedings of the American Control
Conference Chicago, Illinois, vol. 5, June 2000, pp.
3299–3303.
[10] A. Wellenreuther, A. Gambier, and E. Badreddin, “Multi-
loop controller design for a heat exchanger,” in Proceed-
ings of the IEEE International Conference on Control
Applications, Munich,Germany, October 2006, pp. 2099–
2104.
[11] P. Cheingjong and M. Wongsaisuwan, “Adaptive PI con-
trol application of a heat exchanger via distributed control
system,” in IEEE International Conference on Industrial
Technology, Chengdu, China, April 2008, pp. 1–4.
[12] S. Nakayama and H. Kobayashi, “H-∞ controllers using
a bi-linear Y-K observer for heat exchangers with a
bilinear input and a bilinear disturbance,” in SICE Annual
Conference, The Grand Hotel, Taipei, Taiwan, August
2010, pp. 1577–1584.
[13] K. Rajalakshmi and V. Mangaiyarkarasi, “Design of
internal model controller for a heat exchanger system,”
in International Conference on Information Communica-
tion and Embedded Systems (ICICES), Chennai, Chen-
nai,India, February 2013, pp. 899–903.
[14] A. Vasickaninova and M. Bakosova, “Robust controller
design for a heat exchanger,” in 2015 International Con-
ference on Process Control (PC), Strbske Pleso, Slovakia,
June 2015, pp. 113–118.
[15] H. A. Suthar and D. J.Gadit, “Two degree of freedom
controller optimization using GA for shell and tube heat
exchanger,” in 11th International Conference on Intel-
ligent Systems and Control (ISCO), Coimbatore,India,