Professional Documents
Culture Documents
Zhang 2017
Zhang 2017
Zhang 2017
Abstract: In this paper, a quantitative robust control design is investigated for discharge air
temperature (DAT) system. For this purpose, the DAT system is described by an first-order-
plus-dead-time (FOPDT) model. With first-order Padé approximation for the time delay, both
standard and PI structured H∞ control designs are first employed for the controller synthesis. A
set of control objectives and qualitative robustness properties can be simultaneously considered
by appropriate selection of weighting functions. Then, structured singular value (µ) analysis is
used to quantitatively verify robustness properties of the closed-loop system in the presence of
parametric and/or dynamic uncertainties. The numerical simulations of a typical DAT system
demonstrate the effectiveness of the proposed control design as compared to classical SIMC PI
control design.
Keywords: Discharge air temperature control, standard and structured H∞ control designs,
quantitative robustness analysis
2613
weighting functions and determine them is critical to have z(s) =Fu (N, ∆)w(s)
an satisfactory H∞ controller. Therefore, the principle of (6)
=(N22 + N21 ∆(I − N11 ∆)−1 N12 )w(s)
phase and gain control polices is employed here for the
Based on general LFT framework, the structured singular
selection of related weighting functions such that a set
of control objectives as well as robustness properties can
be considered simultaneously. On the other hand, com-
pared to standard non-structured H∞ controllers, struc-
tured controllers such as PID, lead-lag and others are
still preferred in practical industrial control designs, for
example, PI control designs are dominant in DAT and
other HVAC control systems. The main reason is that
standard H∞ controller is usually of high order, not easy
to be implemented and often difficult to re-tune in case
of model changes (Apkarian et al., 2007). So, besides
the standard H∞ control design, in this paper the PI
structured H∞ control design is also investigated for DAT Fig. 3. General LFT framework
system. Compared to the standard H∞ control design, value µ∆ (M ) is defined (Skogestad and Postlethwaite,
the PI structured H∞ control design is to synthesize an 2005)
H∞ optimal controller with a prescribed more under-
1
standable PI structure which is ready to be employed µ∆ (M ) ,
by current DAT system. Nonsmooth optimization tech- min{km | det(I − km M ∆) = 0, ∆ ∈ B ∆ }
nique can be used to have a locally optimal PI structured (7)
H∞ controller (Gahinet and Apkarian, 2011) which is The closed-loop robust stability is then determined by the
ready to be used by practical DAT control system. following theorem (Zhou et al., 1996)
Theorem 1 Assume that the nominal system M and the
2.2 Quantitative robustness analysis perturbation ∆ are stable. Then the M − ∆ is stable for
any ∆ ∈ B ∆ if and only if
Although H∞ control designs are able to qualitatively ac- µ∆ (M (jω)) < 1, ∀ω (8)
count for parametric and dynamic uncertainties by appro- The reciprocal of upper bound of µ∆ (M ) is referred to as
priate selection of weighting functions, it is quite necessary deterministic robustness margin
to perform quantitative robustness analysis to quantita-
1
tively ensure the closed-loop robustness properties, e.g. kDRM = (9)
robust stability and worst-case performance. To achieve max µ∆ (M )
this goal, the original stochastic system with parametric It means how much the normalized parametric and dy-
and dynamic uncertainties has to be rearranged by the namic uncertainties can be enlarged simultaneously before
uncertainty block ∆ and the nominal augmented plant the closed-loop system gets instable, e.g. kDRM ≥ 1 means
N (s) as shown in Fig. 3, where w(s) consists of exogenous that the closed-loop system is ensured to be stable in the
input signals and z(s) consists of regulated variables (Zhou presence of considered uncertainties.
et al., 1996). N can always be chosen so that ∆ is block
Besides robust stability, the worst-case performances of
diagonal, that is, ∆ ∈ ∆
the closed-loop system should also be investigated. The
∆ ,{diag [δ1r It1 , . . . , δVr ItV , δVc +1 Ir1 , . . . , δVc +S IrS , ∆V +S+1 , worst-case performance λwc is defined as
. . . , ∆V +S+F ] : δkr ∈ R, δVc +i ∈ C, ∆V +S+j ∈ Cmj ×mj , λwc (ω) , sup σ̄(Fu (N, ∆)(jω)), ∀ω (10)
∆∈B ∆
1 ≤ k ≤ V, 1 ≤ i ≤ S, 1 ≤ j ≤ F }
As accurate calculation of the values of kDRM and λwc (ω)
where R and C denote the fields of real and complex
r th is NP-hard 1 , lower and upper bounds of µ∆ (M )/kDRM
numbers, δk represents the k real scale parametric uncer-
and λwc are usually computed. MatLab Robust Control
tainty with tk repetition, δVc +i represents the ith repeated Toolbox makes use of the results from Young and Dolye
complex scalar uncertainty with ri repetition and ∆V +S+j (1990) and Packard et al. (2000) to calculate their bounds.
represents the j th full dynamic uncertainty with size mj ×
mj . By incorporating suitable normalization functions in 3. NUMERICAL SIMULATIONS
N (s), we have δkr ∈ [−1, 1], |δic | ≤ 1 and σ̄(∆j ) ≤ 1 and the
notation B ∆ is introduced for the norm bounded diagonal In this paper, as shown in Fig. 1, the proposed quantitative
uncertainty block robust control design is applied to a typical single-input-
B ∆ := {∆ ∈ ∆ : σ̄(∆) ≤ 1} (4) single-output (SISO) DAT system which is identified as an
By partitioning N (s) compatibly with the dimension of FOPDT model of (11),
∆(s) we have Tnom (s) Knom e−Dnom s
Gnom (s) = = (11)
∆y N11 N12 ∆u U (s) Tnom s + 1
= ; M = N11 (5)
z N21 N22 w The nominal parameters Knom , Tnom and Dnom can be
The closed-loop transfer function from w(s) to z(s) is determined by open-loop step test for a practical building,
represented by an upper linear fractional transforma- 1 given any algorithm to compute µ, there will be problems for which
tion (LFT) Fu (N, ∆), the algorithm cannot find the answer in polynomial time.
2614
e.g. Knom = −0.1, Tnom = 400 and Dnom = 100. Since the
time delay is difficult to be directly handled by standard
H∞ control design or PI structured H∞ control design,
its first-order Padé approximation is used and we have
the Padé approximated model of (12) (Carr and O’Dwyer,
1999)
−Knom (Dnom s − 2)
Gpade (s) = (12)
(Tnom s + 1)(Dnom s + 2)
As discussed in Section 2.1, with the general H∞ control
design framework of Fig. 2, we have Gp (s) = Gpade (s). The
disturbance dynamic model Gd (s) the weighting function
Wd and the exogenous input w1 are not necessary here.
According to phase and gain control policies (Zhang et al.,
2013b), the following weighting functions are selected for
the synthesis of standard H∞ controller K∞ (s) PI
Fig. 4. DAT responses to step change with K∞ (s), K∞ (s)
PI
and KSIM C (s)
(s + 2.99) × 10−1
Wv (s) = , Wn = Wy = Wu = 1 Table 1. Control performances with
s + 2.99 × 10−6 PI PI
K∞ (s), K∞ (s) and KSIM C (s)
PI K∞ (s) P I (s)
K∞ PI
KSIM
For the synthesis of PI structured H∞ controller K∞ (s), C (s)
the weighting functions are selected as Rise time/s 130 180 190
Setting time/s 281 497 605
(1.2s + 12.0) × 10−3 Undershoot 0% 2.5% 4.0%
Wv (s) = , Wn = Wy = Wu = 1 Steady-state offset 0% 0% 0%
s + 1.0 × 10−6 ISE 560 663 676
ITAE 33270 56972 57095
With the nominal model Gnom (s), as comparisons the
PI
SIMC PI controller KSIM C (s) is also well tuned here to
have fast responses with nice robustness properties (Sko-
gestad, 2003). Below are the designed controllers
2615
PI information of the parametric uncertainties should be
ties. For instance, with K∞ (s) the smallest kDRM = 1.56
which means that even in the presence of 46.8% variations investigated. The application of the quantitative robust
on nominal model parameters, the closed-loop system can control design to multiple-input-multiple-output (MIMO)
still be stable. The worst-case performance of the uncertain DAT system is also promising. It is expected that model
sensitivity functions and corresponding DAT responses to identification, uncertainty quantification and comparisons
step change are shown in Fig. 6. between numerical simulation and real-time experimental
result can be achieved for practical DAT systems.
REFERENCES
Apkarian, P., Bompart, V., and Noll, D. (2007). Non-
smooth structured control design with application to
PID loop-shaping of a process. International Journal
of Robust and Nonlinear Control, 17(14), 1320–1342.
Apkarian, P. and Noll, D. (2006). Nonsmooth H∞ synthe-
sis. IEEE Transactions on Automatic Control, 51(1),
71–86.
Bi, Q., Cai, W.J., Wang, Q.G., Hang, C.C., Lee, E.L.,
Sun, Y., Liu, K.D., Zhang, Y., and Zou, B. (2000).
Advanced controller auto-tuning and its application in
HVAC systems. Control Engineering Practice, 8(6),
Fig. 6. Worst-case performances of sensitivity function- 633–644.
PI Carr, S. and O’Dwyer, A. (1999). Robust controller
s and DAT responses with K∞ (s), K∞ (s) and
PI design for time delay systems using H∞ techniques. In
KSIM C (s)
Proceedings of the Irish Signals and Systems Conference,
45–52. Galway, Ireland.
From above analysis it is demonstrated that full unstruc- Doyle, J. (1985). Structured uncertainty in control system
tured K∞ (s) has some advantages over PI structured design. In 24th IEEE Conference on Decision and
PI PI
K∞ (s) and classical KSIM C (s) in terms of nominal time- Control, 260–265.
domain control performances but at cost of a certain Doyle, J., Glover, K., Khargonekar, P., and Francis, B.
level of degradations of robust stability and worst-case (1989). State-space solutions to standard H2 and H∞
PI
performance. The K∞ (s) is able to achieve a good bal- control problems. IEEE Transactions on Automatic
ance among control performances, robustness properties Control, 34(8), 831–847.
and practical implementations. The analysis shows that Doyle, J., Packard, A., and Zhou, K. (1991). Review
the proposed quantitative robust control design allows of LFTs, LMI’s, and µ. In 30th IEEE Conference on
us to make a trade-off between control performance and Decision and Control, 1227–1260.
quantitative robustness properties in the presence of un- Fan, M., Tits, A., and Doyle, J. (1991). Robustness
certainties. Note that although classical PI control design in the presence of mixed parametric uncertainty and
is used here for comparisons, it is not to conclude that unmodeled dynamics. IEEE Transactions on Automatic
H∞ control design can replace PI control design. Actually, Control, 36(1), 25–38.
considering controlled plants, hardware limitations and Gahinet, P. and Apkarian, P. (2011). Structured H∞
required control objectives, H∞ and PI control designs synthesis in MATLAB. IFAC Proceedings Volumes,
should be used complementarily. For example, PI struc- 44(1), 1435–1440.
tured H∞ controller can be used initially for further PI Kasahara, M., Matsuba, T., Kuzuu, Y., Yamazaki, T.,
tuning and PI controller deign can also provide valuable et al. (1999). Design and tuning of robust PID controller
information for appropriate selection of weighting func- for HVAC systems. ASHRAE Transactions, 105(2),
tions used by H∞ control designs. 154–166.
Liu, T., Wang, Q.G., and Huang, H.P. (2013). A tuto-
4. CONCLUSIONS rial review on process identification from step or relay
feedback test. Journal of Process Control, 23(10), 1597–
This paper focuses on developing quantitative robust con- 1623.
trol design for discharge air temperature (DAT) system. Nesler, C.G. and Stoecker, W.F. (1984). Selecting the
To achieve this goal, first standard and PI structured proportional and integral constants in the direct digital
H∞ control designs are employed for the controller syn- control of discharge air temperature. ASHRAE Trans-
thesis. The trade-off between different control objectives actions, 90, 834–845.
and qualitative robustness properties can be achieved by Packard, A., Balas, G., Liu, R., and Shin, J.Y. (2000).
appropriate weighting functions. Then, µ analysis is used Results on worst-case performance assessment. In 2000
to quantitatively verify the closed-loop robust stability American Control Conference, volume 4, 2425–2427.
and worst-case performance in the presence of parametric IEEE.
and dynamic uncertainties. The numerical simulations of Qu, G. and Zaheeruddin, M. (2010). Online H∞ adaptive
a typical SISO DAT system demonstrate the effectiveness tuning of PI controllers for discharge air temperature
of proposed control design. In the future, the neglected system. Energy Conversion and Management, 51(6),
high-frequency dynamics of DAT model and statistical 1179–1195.
2616
Seborg, D.E., Edgar, T.F., Mellichamp, D.A., and Doyle I-
II, F.J. (2010). Process Dynamics and Control. John
Wiley & Sons, 3rd edition.
Skogestad, S. and Postlethwaite, I. (2005). Multivariable
Feedback Control-Analysis and Design. John Wiley and
Sons.
Skogestad, S. (2003). Simple analytic rules for model
reduction and PID controller tuning. Journal of Process
Control, 13(4), 291–309.
Underwood, C. (2000). Robust control of HVAC plant
II: controller design. Building Services Engineering
Research and Technology, 21(1), 63–71.
Young, P. and Dolye, J. (1990). Computation of µ with real
and complex uncertainties. In 29th IEEE Conference on
Decirion and Control, 1230–1235.
Zhang, K., Scorletti, G., Ichchou, M., and Mieyeville, F.
(2013a). Robust active vibration control of piezoelectric
flexible structures using deterministic and probabilistic
analysis. Journal of Intelligent Material Systems and
Structures, 25(6), 665–679.
Zhang, K., Scorletti, G., Ichchou, M., and Mieyeville, F.
(2014). Quantitative robust LPV H∞ vibration control
of flexible structures for saving the control energy.
Journal of Intelligent Material Systems and Structures,
26(8), 1006–1027.
Zhang, K., Scorletti, G., Ichchou, M., and Mieyeville, F.
(2013b). Phase and gain control policies for robust
active vibration control of flexible structures. Smart
Materials and Structures, 22(7), 075025.
Zhou, K., Doyle, J., and Glover, K. (1996). Robust and
Optimal Control. Prentice, Hall Upper Saddle River,
NJ.
2617