Control Engineering Practice 129 (2022) 105347

Contents lists available at ScienceDirect

Control Engineering Practice

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

Robust feedback compensation and PID Tuning under actuator rate limit
effect based on Bode’s integrals
Jie Yuan a,c ,∗, Shumin Fei a , YangQuan Chen b , Yichen Ding c
a Key Laboratory of Measurement and Control of CSE, Ministry of Education, School of Automation, Southeast University, Nanjing 210096, China
b
Mechatronics, Embedded Systems and Automation Lab, University of California, Merced, CA 95343, United States of America
c
Erik Jonsson School of Engineering and Computer Science, The University of Texas at Dallas, Richardson, TX 75080, United States of America

ARTICLE INFO ABSTRACT

Keywords: The actuator rate limit is a common phenomenon in control systems while its underlying mechanism remains
Actuator rate limit elusive. Magnitude reduction and phase delay generated by the rate limiter degrade the control performance,
Flat phase but this adverse effect has always been omitted in practical engineering. This study proposes a robust controller
Bode’s integrals
design method and controller update rules for general minimum-phase systems to compensate for the rate limit
Performance improvement
effect and improve control performance. Firstly, a simplified implementation method of ‘‘flat phase’’ constraint
Fractional-order PI controller
is carried out based on the Bode’s integrals for proportional–integral–derivative (PID) and fractional-order
proportional–integral (FOPI) controllers. Secondly, the update rules of the controller design specifications are
proposed based on Bode’s integrals to mitigate the actuator rate limit effect. The proposed method maintains
the original controller structure without extra compensation filters, thereby can be easily applied to engineering
practice. The outperformance of FOPI controllers compared to PID controllers is qualitatively analyzed and
has been indicated to have lower probability of infinity loops under actuator rate limit. Illustrative examples
and experimental results have demonstrated the effectiveness, robustness under gain variation and disturbance
rejection capability of the proposed rate limit compensation strategies.

1. Introduction 2019), automotive low-dropout regulator (Răducan & Neag, 2021), and
hard disk drive systems (Zhou, Zheng, & Tomizuka, 2016), etc.
Actuator restrictions including magnitude saturation and rate limit Rate limit compensation strategies can be broadly classified into
are ubiquitous in control systems. While the magnitude saturation has three categories (Brieger et al., 2009): (A) Limiting the control com-
drawn a considerable amount of attention (Li & Lin, 2018), investiga-
mand by setting up ‘‘legislative means’’, such as adding software rate
tion on rate limit is still insufficient (Jin, Qin, Shi, & Zheng, 2017). With
limiters after the control command generator to avoid saturation of
the increasing requirements on control performance, actuator rate limit
practical actuator (Buchholz, 1993; Chalk, 1992). (B) Considering rate
has become an important issue in process control and modern manu-
facturing. Moreover, output rate of the actuator is artificially restricted limit when designing controllers, which can be done in two ways: (B1)
for safety reasons, such as to restrict rise rate of temperature within considering the rate limit as a constraint in controller online optimiza-
safe limits in distillation columns, and avoid huge current change in tion or state observers, such as saturated sliding control (Wang, Gao,
motor system to protect electrical circuits. Magnitude reduction and Liu, & Wu, 2018), nonlinear optimal control (Tran, Sakamoto, Kikuchi,
phase delay generated by rate saturated actuators degrade control & Mori, 2017), iterative leaning control (Blank, Glück, Kugi, & Kreuter,
performance and may lead to system instability (Wu, Yuan, Li, Xue, 2014), and active disturbance rejection control (Herbst, 2015; Wu,
& Chen, 2020; Yuan, Fei, & Chen, 2021). In addition to the rate Yuan, Liu, Li, & Chen, 2021); (B2) designing controllers by ignoring
limit effect on control performance, the actuator rate limit also has an rate limit first and then compensating the controller to mitigate the
significant influence on system identification (Yuan, Ding, Fei, & Chen,
rate limit effect (Yuan, Fei, & Chen, 2020). (C) Employing additional
2022; Yuan, Han, Chai, Fei, & Chen, 2020). Actuator rate limit issue has
blocks to the existing controllers which only activate when rate limit
been found crucial in boiler–turbine dynamics (Chen & Shamma, 2004),
spacecraft attitude maneuver systems (Sahu, Mahodaya, & Chokkadi, occurs, motivated by the anti-windup strategy (Brieger et al., 2009;
2020), compression systems (Eldigair, Garelli, Kunusch, & Ocampo- Flesch, Normey-Rico, & Flesch, 2017; Sofrony, Turner, & Postlethwaite,
Martinez, 2020), trajectory tracking marine vehicles (von Ellenrieder, 2009). Method A is sensitive to noise and when desaturated, the control

∗ Corresponding author at: Key Laboratory of Measurement and Control of CSE, Ministry of Education, School of Automation, Southeast University, Nanjing
210096, China.
E-mail addresses: [email protected] (J. Yuan), [email protected] (S. Fei), [email protected] (Y. Chen), [email protected] (Y. Ding).

https://doi.org/10.1016/j.conengprac.2022.105347
Received 9 May 2022; Received in revised form 16 August 2022; Accepted 13 September 2022
Available online 28 September 2022
0967-0661/© 2022 Elsevier Ltd. All rights reserved.
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

command may not be recovered due to unforeseen effects (Rundqwist

& Stahl-Gunnarsson, 1996). Method B1 is usually implemented in time-
domain and integrated with online optimization, which is more suitable
to be applied to sluggish systems. Mitigating actuator rate limit mainly
employs the anti-windup strategy (Method C) since it is effective and
able to maintain baseline control performance. This method needs to be
implemented to variable structure controllers, leading to more effort to
tune compensator parameters. Method B2 is effective in reducing the
complexity of system design and is chosen to be the compensation strat-
egy in this study. Generally, compensators in Method B2 are connected
in cascaded to the original controller; however, compensating the rate
limit effect only through changing controller parameters without extra
filters is highly practical. To the authors’ knowledge, few papers have
been published to investigate this practical strategy. In this context,
the authors are motivated to propose an actuator rate limit mitigating
approach that maintains the original controller structure.
While proportional–integral–derivative (PID) controllers remain
dominant in industry processes, fractional-order PID controllers (FOPID)
have attracted increasing attention in terms of more design free- Fig. 1. Stable responses (dashed) of rate limiter to sinusoidal inputs (Yuan et al., 2018).
dom (Caponetto, 2010; Maione & Lino, 2007; Monje, Chen, Vinagre,
Xue, & Feliu-Batlle, 2010; Petráš, 2011; Tepljakov et al., 2018; Yu et al.,
2021; Yuan, Wu, Fei, & Chen, 2019). The Bode’s ideal transfer function and a delay-dominant process; experimental results are also carried
and Bode’s ideal cut off idea are popular utilized in fractional-order out in Section 8; in Section 9, the outperformance of FOPI controller
controller design which achieves good balance between disturbance in mitigating the actuator rate limit effect is qualitatively analyzed;
rejection capability and noise rejection capability (Barbosa, Machado, finally, conclusions are drawn in Section 10.
& Ferreira, 2004; Keziz, Djouambi, & Ladaci, 2020; Yumuk, Güzelkaya,
& Eksin, 2019; Zheng, Luo, & Chen, 2020). The majority of controller 2. The describing function of the rate limiter
tuning rules are carried out in frequency domain where gain crossover
frequency and phase margin serve as important indicators of control The actuator rate limit phenomenon can be depicted with a rate lim-
performance and system robustness. ‘‘Flat phase’’ is another typical iter in convenience. Assuming the input of a rate limiter is a sinusoidal
robustness constraint originated from Bode’s works (Luo & Chen, 2012; function in the form of
Seyedtabaii, 2017). It requires the phase plot of loop transfer func-
tions to be flat around the crossover frequency maintaining the phase 𝑥(𝑡) = 𝐴 sin(𝜔𝑡), (1)
margin under certain system gain variation. This controller design where 𝐴, 𝜔, 𝑡 represent magnitude, frequency and time respectively.
approach has been successfully applied to first-order plus time-delay Changing rate of rate limiter output is restricted by
(FOPTD) systems suffering from actuator rate limit (Yuan, Chen, &
Fei, 2018). However, these traditional implementation of the flat phase |𝑦(𝑡)|
̇ ≤ 𝑅, (2)
constraint is tedious due to the parametric deduction on the derivative
where 𝑅 is called ‘‘rate limit value’’.
of loop phase functions, thereby limits its application to second-order
Rate limiters have four different behaviors, whose steady-state os-
or higher order systems. To simplify the design procedures of flat
cillations excited by sinusoidal input are shown in Fig. 1. With the
phase controllers, this study utilized the Bode’s integrals approximation
fixed frequency and magnitude of the input signal, smaller rate limit
method proposed by Karimi, Garcia, and Longchamp (2002, 2003)
values generate more severe distortion of output signals, as shown
to approximate the slope of Bode plots. This approximation method
in Fig. 1(a–c). When the rate limiter is fully activated as shown in
reshapes the slope of Nyquist curve at a certain frequency point based
Fig. 1(d), the output becomes a purely triangular wave and generates
on the Bode’s gain-phase integrals in minimum-phase systems (Bode,
dramatic magnitude reduction and phase delay.
1945) without parametric system model. Furthermore, inspired by the
Describing function of a fully activated rate limiter is
emerging rate limit issue in modern process control, update rules for
re-tuning the controller parameters are proposed based on the Bode’s 4 𝜔onset −𝑗 arccos 𝜋2 𝜔onset
𝑁(𝑗𝜔, 𝜔onset ) = e 𝜔 , (3)
integrals to mitigate the actuator rate limit effect. 𝜋 𝜔
Main contributions of this study are: (1) utilizing Bode’s integral where 𝜔onset = 𝑅∕𝐴, defined as the frequency where actuator satu-
approximation method to simplify the implementation of flat phase ration first occurs (Gilbreath, 2001). Eq. (3) is only valid when 𝜔 ≥
constraint and promoting the flat phase robust controller to be ex- 1.862𝜔onset (Duda, 1994). Bode plot of a sample rate limiter is shown
tensively applied in general minimum-phase systems; (2) proposing in Fig. 2 (cubic spline interpolations are used to draw magnitude and
the update rules of controller design specifications to mitigate the phase in partly saturated region whose analytical expressions are quite
actuator rate limit while maintaining the original control structure. complicated).
The proposed rate limit compensation method is an economical, prac-
tical and effective approach with less efforts in both computing and 3. Robust controller design specifications
controller module costs. The remainder of this paper is organized as
follows: characteristics and describing function of the rate limiter are Gain crossover frequency and phase margin of desired loop transfer
given in Section 2; controller design specifications of the flat phase functions are widely specified in robust controller design. The ‘‘flat
approach and its implementation based on Bode’s integrals are depicted phase’’ is developed to avoid extensive variation of phase margin
in Sections 3 and 4 respectively; in Section 5, parameter tuning rules under system gain perturbation, which requires the phase Bode plot of
of IOPID and FOPI controllers are deduced; the update rules of the loop transfer functions to be flat around the crossover frequency. This
design specifications to compensate the rate limit effect are given in constraint is able to generate the same overshoot of the step responses
Section 6; two illustrative examples are given in Section 7 to validate under loop gain variation for linear systems, called the iso-damping
the effectiveness of the proposed method on a fast responding system property (Chen, Hu, & Moore, 2003).

2
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

Let 𝜑 represents the slope of the Nyquist curve at the crossover fre-
quency, then
( d𝐿(𝑗𝜔) )|
𝜑 = arg |
|
d𝜔 |𝜔=𝜔𝑏
( d|𝐿(𝑗𝜔)| )|
= arg 𝑒𝑗∠𝐿(𝑗𝜔) ||
d𝜔 |𝜔=𝜔𝑏
( | d|𝐿(𝑗𝜔)| | )| (12)
= arg − || | 𝑒𝑗∠𝐿(𝑗𝜔) |
| |
| d𝜔 | |𝜔=𝜔𝑏
= 𝜋 + ∠𝐿(𝑗𝜔𝑏 )
= 𝛷𝑚 .

4. Implementation of the flat phase constraint based on Bode’s

integrals

For general minimum-phase systems, the implementation of flat

phase constraint in Eq. (7) is much more difficult to derive than
FOPTD systems. A PID tuning approach based on the Bode’s integrals
is proposed (Karimi et al., 2003), where the derivative of loop transfer
Fig. 2. The Bode plot of the describing function of a rate limiter. function is calculated without system model information.
The derivative of the loop transfer function is able to be decomposed
as
Define 𝐺(𝑠) and 𝐶(𝑠) representing the minimum-phase plant transfer ( )
d𝐿(𝑗𝜔) d𝐶(𝑗𝜔) dln |𝐺(𝑗𝜔)| d∠𝐺(𝑗𝜔)
function and controller, respectively (𝐶(𝑠) can be either integer-order = 𝐺(𝑗𝜔) + 𝐶(𝑗𝜔)𝐺(𝑗𝜔) +𝑗 .
d𝜔 d𝜔 d𝜔 d𝜔
or fractional-order). 𝜔𝑏 is the desired crossover frequency and 𝛷𝑚 is the
desired phase margin. Three design specifications are given as follows: (13)

(i) Gain crossover frequency constraint Key factors of Eq. (13) are the slope of magnitude and phase of 𝐺(𝑗𝜔)
which need to be further analyzed.
|𝐶(𝑗𝜔𝑏 )𝐺(𝑗𝜔𝑏 )| = 1; (4)
| |
4.1. Slope of magnitude Bode plot
(ii) Phase margin constraint

arg(𝐶(𝑗𝜔𝑏 )𝐺(𝑗𝜔𝑏 )) = −𝜋 + 𝛷𝑚 ; (5) The relations between magnitude and phase of a minimum-phase
system is first investigated in Bode (1945). The first relation is well
(iii) Flat phase constraint on Bode plot
known as the Bode’s integral:
( )
d arg(𝐶(𝑗𝜔)𝐺(𝑗𝜔)) | ∞
dln |𝐺(𝑗𝜔)| |𝑣|
| = 0; (6) 1
| ∠𝐺(𝑗𝜔0 ) = ln coth d𝑣, (14)
d𝜔 |𝜔=𝜔𝑏 ∫
𝜋 −∞ d𝑣 2
or equivalently on Nyquist curve, where 𝑣 = ln 𝜔𝜔 . This equation indicates that the phase of the minimum-
0
( d(𝐶(𝑗𝜔)𝐺(𝑗𝜔)) )| phase transfer function at a specific frequency is related to slope of the
|
𝜑 = arg
d𝜔 |
|𝜔=𝜔𝑏
= 𝛷𝑚 . (7) magnitude Bode plot. Since ln coth |𝑣| 2
reduces rapidly as 𝜔 deviating
from 𝜔0 , the value of Eq. (14) mainly depends on the slope of magni-
In this study, the flat phase constraint will be realized based on Nyquist tude around 𝜔0 . We assume that the slope of magnitude Bode plot is a
curve Eq. (7). Eq. (6) is the traditional implementation of the flat phase constant in the neighborhood of 𝜔0 , then
constraint based on Bode plot, which indicates that the derivative of
phase with respect to 𝜔 equals zero at 𝜔𝑏 . Eq. (7) represents the slope dln |𝐺(𝑗𝜔)| || 1 ∞
|𝑣|
∠𝐺(𝑗𝜔0 ) ≈ | ln coth d𝑣
of the Nyquist curve of the loop transfer function equaling the phase d𝑣 |𝜔0 𝜋 ∫−∞ 2
(15)
margin at 𝜔𝑏 . These two equations are equivalent and will be deduced 𝜋 dln |𝐺(𝑗𝑤)| ||
≈ 𝜔0 | .
in the following. 2 d𝜔 |𝜔0
Let 𝐿(𝑠) = 𝐶(𝑠)𝐺(𝑠) represents the loop transfer function, then it can The slope of magnitude Bode plot is approximated by
be written as
dln |𝐺(𝑗𝜔)| || 2
| ≈ 𝜋𝜔 ∠𝐺(𝑗𝜔0 ). (16)
𝐿(𝑗𝜔) = |𝐿(𝑗𝜔)|𝑒𝑗∠𝐿(𝑗𝜔) . (8) d𝜔 | 𝜔0 0

For minimum-phase system with a pure time delay 𝐺(𝑠) = 𝐺0 (𝑠)𝑒−𝜏𝑠 , it

The slope of the Nyquist curve of 𝐿(𝑗𝜔) at frequency 𝜔0 is defined as
has
the phase of the derivative of 𝐿(𝑗𝜔) at 𝜔0 . The derivative of 𝐿(𝑗𝜔) with
respect to frequency is dln |𝐺(𝑗𝜔)| || 2 ( )
| ≈ 𝜋𝜔 ∠𝐺(𝑗𝜔0 ) + 𝜏𝜔0 , (17)
d𝜔 | 𝜔0 0
d𝐿(𝑗𝜔) d|𝐿(𝑗𝜔)| 𝑗∠𝐿(𝑗𝜔) d∠𝐿(𝑗𝜔) 𝑗∠𝐿(𝑗𝜔)
= 𝑒 + 𝑗|𝐿(𝑗𝜔)| 𝑒 . (9) where 𝐺0 (𝑠) is the linear transfer function without delay.
d𝜔 d𝜔 d𝜔
According to Eq. (6), we have
4.2. Slope of phase Bode plot
d∠𝐿(𝑗𝜔) ||
= 0, (10)
d𝜔 ||𝜔=𝜔𝑏
Another gain-phase integral relation is proposed in Bode (1945)
then ( )
∠𝐺(𝑗𝑤)
d𝐿(𝑗𝜔) || d|𝐿(𝑗𝜔)| 𝑗∠𝐿(𝑗𝜔) || 𝜔 ∞ d
𝜔 |𝑣|
= 𝑒 . (11) ln ||𝐺(𝑗𝜔0 )|| = ln |𝐾| − 0
d𝜔 ||𝜔=𝜔𝑏 | ln coth d𝑣, (18)
d𝜔 |𝜔=𝜔𝑏 𝜋 ∫−∞ d𝑣 2

3
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

where 𝐾 is steady-state gain of system (without integrators). Simi-

larly, the slope of phase Bode plot is assumed to be a constant in
neighborhood of 𝜔0 , then
( )
𝜋𝜔20 1 d∠𝐺(𝑗𝜔) || ∠𝐺(𝑗𝜔0 )
| |
ln |𝐺(𝑗𝜔0 )| ≈ ln |𝐾| − − . (19)
2 𝜔0 d𝜔 ||𝜔0 𝜔20
Finally the slope of phase Bode plot is approximated as
d∠𝐺(𝑗𝜔) || ∠𝐺(𝑗𝜔0 ) 2 ( | ) Fig. 3. The closed-loop system diagram with a rate limiter.
≈ − ln |𝐺(𝑗𝜔0 )|| − ln |𝐾| . (20)
d𝜔 ||𝜔0 𝜔0 𝜋𝜔0
It should be noted that Eq. (20) is also applicable to minimum-phase
By substituting Eq. (30) into Eq. (13), the derivative of loop transfer
systems with pure time delays.
function is deduced as
[(
5. Controller tuning rules d𝐿(𝑗𝜔) dln |𝐺(𝑗𝑤)| 𝜋𝛼 𝑘𝑖 d∠𝐺(𝑗𝜔)
= 𝐺(𝑗𝜔) 𝑘𝑝 + sin
d𝜔 d𝜔 2 𝜔𝛼 d𝜔
In this study, the PID and FOPI controllers are employed both of ( 𝛼𝑘 ) ) (
𝜋𝛼 𝑖 𝑘 dln |𝐺(𝑗𝜔)| d∠𝐺(𝑗𝜔)
which have three parameters and same number of design freedom. The cos − 𝑖 +𝑗 𝑘𝑝 + (31)
PID controller is in the form of 2 𝜔𝛼+1 𝜔𝛼 d𝜔 d𝜔
) ]
( 𝑘 dln |𝐺(𝑗𝑤)| )
𝐾 𝜋𝛼 𝛼𝑘𝑖 𝜋𝛼 𝑘𝑖 d∠𝐺(𝑗𝜔)
𝐶I (𝑠) = 𝐾𝑝 + 𝑖 + 𝐾𝑑 𝑠, (21) sin − 𝛼𝑖 − cos 𝛼
.
2 𝜔 𝛼+1 𝜔 d𝜔 2 𝜔 d𝜔
𝑠
and the FOPI controller is in the form of The slope the loop Nyquist curve at 𝜔𝑏 is determined by
𝑘 ( 𝛼 )
𝐶F (𝑠) = 𝑘𝑝 + 𝛼𝑖 , (22) 𝜔𝑏 𝑘𝑝 𝐵 + sin 𝜋𝛼 (𝛼𝑘𝑖 − 𝑘𝑖 𝐴) − cos 𝜋𝛼 𝑘𝐵
𝑠 −1 2 2 𝑖
𝜑 = ∠𝐺(𝑗𝜔𝑏 ) + tan . (32)
where 0 < 𝛼 < 2. 𝜔𝛼𝑏 𝑘𝑝 𝐴 + cos 𝜋𝛼
2
(𝛼𝑘𝑖 − 𝑘𝑖 𝐴) + sin 𝜋𝛼 𝑘𝐵
2 𝑖
Three design constraints are specified: (1) gain crossover frequency
The parameters of the FOPI controller can be determined by solving
𝜔𝑏 ; (2) phase margin 𝛷𝑚 ; (3) flat phase 𝜑 = 𝛷𝑚 .
Eqs. (28), (29) and (32).
5.1. PID tuning rules
6. Rate limit compensations with new controller design specifica-
Under crossover frequency and phase margin specifications, one has tions

√
( )2 Compensation strategies of actuator rate limit effect in closed-loop
𝐾𝑖
𝐾𝑝2 + −𝐾𝑑 𝜔𝑏 ||𝐺(𝑗𝜔𝑏 )|| = 1, (23) systems is introduced in this section. Based on the proposed method
𝜔𝑏 introduced in Section 5, the uncompensated controller is first designed
( )
𝜔𝑏 𝐾 𝑑 𝐾𝑖 without considering the rate limit. Then the actuator rate limit effect
tan−1 − + ∠𝐺(𝑗𝜔𝑏 ) = −180◦ + 𝛷𝑚 . (24)
𝐾𝑝 𝜔𝑏 𝐾 𝑝 in the closed-loop system will be analyzed and compensated.
After simple derivation, one has The system diagram containing a rate limiter is shown in Fig. 3. The
concept of closed-loop onset frequency 𝜔 ̃onset is defined as the frequency
d𝐶I (𝑗𝜔) 𝐾
= 𝑗( 𝑖 + 𝐾𝑑 ). (25) point where the actuator first saturated in the closed-loop system. It can
d𝜔 𝜔2 be solved by the following equation (Gilbreath, 2001)
Substituting Eq. (25) into Eq. (13), it has
[( | 𝐶(𝑗𝜔) | 𝑅
d𝐿(𝑗𝜔) dln |𝐺(𝑗𝑤)| ( 𝐾 ) d∠𝐺(𝑗𝜔) ) 𝑟0 || |
| = , (33)
= 𝐺(𝑗𝜔) 𝐾𝑝 − 𝐾𝑑 𝜔 − 𝑖 + | 1 + 𝐶(𝑗𝜔)𝐺(𝑗𝜔) |𝜔=𝜔̃onset 𝜔
̃onset
d𝜔 d𝜔 𝜔 d𝜔
(( ) ] (26) where 𝑟0 is amplitude of reference signal 𝑟. The left side of Eq. (33)
𝐾 ) dln |𝐺(𝑗𝑤)| d∠𝐺(𝑗𝜔) 𝐾
𝑗 𝐾𝑑 𝜔 − 𝑖 + 𝐾𝑝 + 𝐾𝑑 + 𝑖 . is the amplitude of input signal of the rate limiter in the closed-
𝜔 d𝜔 d𝜔 𝜔2
loop system. In the magnitude Bode plot, the right side of Eq. (33)
Slope of the Nyquist curve at 𝜔𝑏 is derived as is a straight line with the slope −20 dB/decade crossing 0 dB at the
(( ) )
𝐾𝑑 𝜔𝑏 2 − 𝐾𝑖 𝐴 + 𝐾𝑝 𝜔𝑏 𝐵 + 𝐾𝑑 𝜔2𝑏 +𝐾𝑖 frequency point lying at 𝑅 rad/s. By looking into the intersection point
−1
𝜑 = ∠𝐺(𝑗𝜔𝑏 ) + tan ( ) , (27) of both sides of Eq. (33), the 𝜔
̃onset can be determined.
𝐾𝑝 𝜔𝑏 𝐴 − 𝐾𝑑 𝜔2𝑏 − 𝐾𝑖 𝐵
Once 𝜔 ̃onset is obtained, the rate limit effect in the closed-loop
dln|𝐺(𝑗𝑤)| | d∠𝐺(𝑗𝜔) |
where 𝐴 = 𝜔𝑏 | and 𝐵 = 𝜔𝑏 | . system can be obtained by calculating the describing function. When
d𝜔 |𝜔𝑏 d𝜔 |𝜔𝑏
By solving Eqs. (23), (24) and (27) (i.e. with the help of fsolve.m in the rate limiter is activated in the closed-loop system, control perfor-
Matlab), the parameters of PID controller can be determined. mance based on the original design specifications will deteriorate due
to the magnitude attenuation and phase delay generated by the rate
5.2. FOPI tuning rules limiter. Thus, the controller design specifications must be updated to
compensate for the rate limit effect.
The first two constraints can be easily represented as
√
√ 6.1. Update rule for gain crossover frequency
√ 2
√𝑘2 + 𝑘𝑖 − 2 𝑘𝑝 𝑘𝑖 cos 𝜋𝛼 |𝐺(𝑗𝜔 )| = 1, (28)
𝑝
𝜔2𝛼 𝜔𝛼𝑏 2 | 𝑏 |
Assume that 𝐿(𝑗𝜔) represents the original desired loop transfer
(𝑏 )
𝑘𝑖 sin 𝜋𝛼
2
function, and 𝑁(𝑗𝜔) represents the describing function of the rate
tan−1 + ∠𝐺(𝑗𝜔𝑏 ) = −180◦ + 𝛷𝑚 . (29) limiter whose onset frequency is determined as 𝜔̃onset . The new desired
𝜔𝛼𝑏 𝑘𝑝 − cos 𝜋𝛼
2
loop transfer function 𝐿∗ (𝑗𝜔) should be compensated with the rate limit
The derivative of 𝐶F (𝑗𝜔) is effect in the low and middle frequency
d𝐶F (𝑗𝜔) 𝛼𝑘𝑖
= 𝑗𝛼 . (30) 𝐿∗ (𝑗𝜔)𝑁(𝑗𝜔) = 𝐿(𝑗𝜔), (34)
d𝜔 𝜔𝛼+1

4
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

∗
( )
𝛷𝑚 ≈ 𝛷𝑚 − ∠𝑁(𝑗𝜔∗𝑏 ) − ∠𝐺(𝑗𝜔𝑏 ) − ∠𝐺(𝑗𝜔∗𝑏 ) , (44)

where ∠𝐶(𝑗𝜔∗𝑏 ) − ∠𝐶(𝑗𝜔𝑏 ) is assumed to be small and can be ignored

since ∠𝐶(𝑗𝜔∗𝑏 ) cannot be calculated currently.
When the controller is first designed based on three specifications
(𝜔𝑏 , 𝛷𝑚 and flat phase), the desired crossover frequency and phase
margin will be updated according to Eqs. (39) and (44) to compensate
the rate limit effect.

6.3. Design procedures of the compensated robust controller considering rate

limit effect

The detailed design procedures of the compensated robust controller

considering the rate limit effect are given as follows:

(1) Determine the original controller design specifications: gain

crossover frequency 𝜔𝑏 , 𝛷𝑚 and flat phase;
(2) Design the PID controller based on Eqs. (23), (24) and (27), or
the FOPI controller based on Eqs. (28), (29) and (32);
(3) Calculate the closed-loop onset frequency 𝜔
̃onset of the rate limiter
based on Eq. (33), and the describing function of the rate limiter
Fig. 4. The magnitude plot (a) and phase plot (b) of compensated loop transfer function
𝐿∗ (𝑗𝜔).
according to Eq. (3);
(4) Update the desired crossover frequency and phase margin based
on Eqs. (39)–(40) and (44);
(5) Repeat step (2) and obtain the parameters of the compensated
Thus one has
robust controller.
log10 |𝐿∗ (𝑗𝜔)| = log10 |𝐿(𝑗𝜔)| − log10 |𝑁(𝑗𝜔)|, (35)
7. Illustrative examples
and this relation is illustrated in Fig. 4(a).
Furthermore, 𝜔∗𝑏 is the new desired crossover frequency and satis- Two simulation examples, including a fast responding DC motor
| |
fied to |𝐿∗ (𝑗𝜔∗𝑏 )| = 1, then model and a delay dominant FOPTD model, will be given in this section
| |
to show the effectiveness of the proposed rate limit compensation
| ∗ | | |
|𝑁(𝑗𝜔𝑏 )| = |𝐿(𝑗𝜔∗𝑏 )| , (36) strategy and the flat phase controller design approach.
| | | |
which means that the magnitude plots of 𝑁(𝑗𝜔) and 𝐿(𝑗𝜔) intersect
7.1. DC motor
at the new desired crossover frequency. Assumed the slope of the
magnitude plot to be a constant around 𝜔𝑏 , then slope of magnitude
A first-order transfer function is used to modeling the DC motor
Bode plot is derived as
( ) 1
d 20log10 |𝐿(𝑗𝜔)| d (ln |𝐿(𝑗𝜔)|) 𝐺1 (𝑠) = , (45)
= 20𝜔 . (37) 2𝑠 + 1
d log10 𝜔 d𝜔 and the rate limit value of the actuator is 𝑅 = 1. The controller design
According to Eq. (16), the slope at 𝜔𝑏 equals specifications are 𝜔𝑏 = 0.9 rad/s, 𝛷𝑚 = 45◦ and the flat phase constraint.
( ) Based on the proposed robust controller design method, the flat phase
d 20log10 |𝐿(𝑗𝜔)| || 40
| = ∠𝐿(𝑗𝜔𝑏 ). (38) PID and FOPI controllers without considering rate limit are obtained as
d log10 𝜔 | 𝜋
|𝜔=𝜔𝑏 2.9278
𝐶I1 (𝑠) = 0.5657 + + 1.4147𝑠, (46)
Then in Fig. 4(a) one has 𝑠
( ) and
20log10 ||𝑁 ∗ (𝑗𝜔𝑏 )|| 40
= ∠𝐿(𝑗𝜔𝑏 ). (39) 1.8206
log10 𝜔∗𝑏 − log10 𝜔𝑏 𝜋 𝐶F1 (𝑠) = 0.2654 + . (47)
𝑠0.9042
If 𝐺(𝑠) contains a pure time delay 𝜏, one has The Bode plots of loop transfer functions under the two controllers
( )
20log10 ||𝑁 ∗ (𝑗𝜔𝑏 )|| 40
are plotted in Fig. 5. According to Eq. (33), the closed-loop onset
= ∠(𝐿(𝑗𝜔𝑏 ) + 𝜏𝜔𝑏 ). (40) frequency of the rate limiter will be calculated first, as listed in Table 1.
log10 𝜔∗𝑏 − log10 𝜔𝑏 𝜋
Following the update rules in Eqs. (39) and (44), the new design
By solving Eq. (39) or (40), the new desired gain crossover frequency specifications are obtained to compensate the rate limit effect. Then
𝜔∗𝑏 can be determined. the compensated PID and FOPI controllers are derived as
2.0606
6.2. Update rule for phase margin 𝐶̃I1 (𝑠) = 2.9150 + + 0.2214𝑠, (48)
𝑠
and
From Eq. (34), the phase relation between original and compensated
1.5752
loop transfer function is 𝐶̃F1 (𝑠) = 2.4074 + . (49)
𝑠0.7349
∠𝐿∗ (𝑗𝜔) = ∠𝐿(𝑗𝜔) − ∠𝑁(𝑗𝜔). (41) The closed-loop onset frequencies of the compensated controllers
are given in Table 1. It can be seen that the onset frequencies are
The relation is illustrated in Fig. 4(b). At 𝜔∗𝑏 one gets increased after compensation which means less rate limit effect, demon-
strating the effectiveness of the proposed rate limit compensation
∠𝐿∗ (𝑗𝜔∗𝑏 ) = ∠𝐿(𝑗𝜔∗𝑏 ) − ∠𝑁(𝑗𝜔∗𝑏 ), (42)
strategy. The compensated FOPI controller provides the largest 𝜔 ̃onset ,
∗
𝜋 + ∠𝐿 (𝑗𝜔∗𝑏 ) =𝜋 + ∠𝐿(𝑗𝜔𝑏 ) − ∠𝐿(𝑗𝜔𝑏 ) + ∠𝐿(𝑗𝜔∗𝑏 ) − ∠𝑁(𝑗𝜔∗𝑏 ), (43) demonstrating the outperformance of FOPI controllers. The Nichols

5
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

Fig. 5. Bode plots of loop transfer functions under uncompensated PID and FOPI controllers. 𝐺1 (𝑠) = 1∕(2𝑠 + 1).

Fig. 6. Nichols charts of the open-loop system with/without rate limiter for PID controller (a) and FOPI controller (b). 𝐺1 (𝑠) = 1∕(2𝑠 + 1).

𝑅 𝜔
̃
Table 1 describing function (NIDF) of the rate limiter 1∕𝑁(𝑘) (𝑘 = − 𝜔𝐴 = onset
𝜔
)
Design specifications and controller parameters for 𝐺1 (𝑠) (‘‘U’’ represents uncompen-
are plotted in Fig. 7. With the increase of 𝜔𝐴, NIDF intersects the loop
sated controllers and ‘‘C’’ represents compensated controllers).
Nyquist curves under uncompensated/compensated PID controllers and
Controller 𝜔𝑏 𝛷𝑚 𝑘𝑝 𝑘𝑖 𝑘𝑑 ∕𝛼 𝜔
̃onset
uncompensated FOPI controller, bringing possibility of infinity loops
UPID/𝐶I1 0.9 45 0.57 2.93 1.41 0.55
in the closed-loop systems. When applying the compensated FOPI
CPID/𝐶̃I1 1.47 88.53 2.92 2.06 0.22 0.63
controller, there is no intersection point between the Nyquist curve and
UFOPI/𝐶F1 0.9 45 0.27 1.82 0.90 0.55
NIDF, thus the closed-loop system is stable. The rate limit compensation
CFOPI/𝐶̃F1 1.46 88.22 2.41 1.58 0.73 0.68
drags the Nyquist curve to the right in the complex plane, reducing the
possibility of the interaction between Nyquist curve and NIDF, thereby
improving the stability of the system.
charts of the open-loop systems both under and without rate limit To compare the performance of the proposed rate limit compensa-
are illustrated in Fig. 6. In the Nichols chart, a phase jump appears tion strategy, the anti-windup structure oriented from Åström, Häg-
at the closed-loop onset frequency where the rate limit is activated. glund, and Astrom (2006) is integrated with the FOPI controller to
When the system suffers from actuator rate limit at 𝜔 ̃onset , the partial mitigate actuator rate limit effect. The back-calculation scheme used in
activation of rate limiter forces the system into the nonlinear mode. this study is plotted in Fig. 8, where 𝑘𝑎 is suggested to be 𝑘𝑖 ∕𝑘𝑝 (Shin,
A resonance jump appears due to a non-unique relation between the 1998).
periodic forcing input and output of the nonlinear system, thereby The step responses and control signals of the uncompensated/
resulting the phase jump in the Nichols chart. The Nyquist curves of compensated PID/FOPI and anti-windup FOPI controllers are plotted in
loop transfer functions and the frequency response of negative inverse Fig. 9. (The implementation of fraction-order controllers are realized

6
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

Fig. 7. Nyquist plots of loop transfer functions and NIDF of rate limiter. 𝐺1 (𝑠) = 1∕(2𝑠 + 1).

Fig. 8. Anti-windup control structure for rate limit compensation.

Fig. 9. Step responses (a) and control signals (b) under flat phase PID, FOPI and anti-windup FOPI controllers. 𝐺1 (𝑠) = 1∕(2𝑠 + 1).

7
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

Table 2 Table 3
Control performance of the closed-loop systems for 𝐺1 (𝑠) (‘‘AWFOPI’’ represents the Design specifications and controller parameters for 𝐺2 (𝑠).
anti-windup FOPI controller, rise time and settling time are calculated with 0%–90% Controller 𝜔𝑏 𝛷𝑚 𝑘𝑝 𝑘𝑖 𝑘𝑑 ∕𝛼 𝜔
̃onset
and 2% criterion).
UPID/𝐶I2 0.29 40 −0.005 5.22 60.24 0.21
Controller rise time /s settling time /s overshoot /% ISE
CPID/𝐶̃I2 0.45 56.50 0.37 0.32 0.70 0.20
UPID/𝐶I1 2.26 19.46 79.23 5.82 UFOPI/𝐶F2 0.29 40 −0.53 0.40 0.50 0.21
CPID/𝐶̃I1 2.26 6.57 48.86 3.14 CFOPI/𝐶̃F2 0.48 55.95 0.21 0.26 0.70 0.25
UFOPI𝐶F1 2.26 10.89 46.84 3.15
CFOPI/𝐶̃F1 2.26 5.32 17.89 2.45
AWFOPI/𝐶A1 2.46 9.37 17.43 2.52 Table 4
Control performance of the closed-loop systems for 𝐺2 (𝑠).
Controller rise time /s settling time /s overshoot /% ISE
UPID/𝐶I2 – – – –
with the Matlab toolbox ‘‘ninteger’’ written by Duarte Valério. The CPID/𝐶̃I2 – – – –
‘‘ninteger’’ toolbox can be downloaded from: https://www.mathworks. UFOPI𝐶F2 6.9 96.26 48.01 9.27
com/matlabcentral/fileexchange/8312-ninteger). A positive 20% vari- CFOPI/𝐶̃F2 5.09 70.06 35.12 4.81
ation is applied to the steady-state gain (gain has increased by 20%) AWFOPI/𝐶A2 6.41 97.00 19.58 7.51
at 50 s, and another unit step disturbance is also superimposed to the
system output at 80 s The control performance are listed in Table 2. It
can be seen that the time duration of rate limited FOPI control signal
is smaller then the PID control signal, providing smaller overshoots constraint. The PID and FOPI controller without considering rate limit
and smaller settling time than the PID controller. Compensation of rate are
limit effect obviously improves the control performance with regard to 5.2169
𝐶I2 (𝑠) = −0.0054 + + 60.2375𝑠, (51)
settling time, overshoot and the ISE index, demonstrating the effective- 𝑠
ness of the proposed rate limit compensation method. Compared with and
the proposed compensated FOPI controller, the anti-windup FOPI con- 0.3964
troller provides a slightly smaller overshoot, a slightly greater ISE and 𝐶F2 (𝑠) = −0.5259 + . (52)
𝑠0.5001
a longer settling time. At 50 s, all controllers have a good robustness
The compensated controllers are obtained as
to the gain variation, especially that the compensated FOPI controller
drags the system output to the reference value the most quickly. Both 0.3220
𝐶̃I2 (𝑠) = 0.3652 + + 0.6978𝑠, (53)
the compensated FOPI controller and anti-windup controller are mod- 𝑠
ified from the uncompensated FOPI controller based on the proposed and
compensation strategy and anti-windup method. However, anti-windup 0.2567
𝐶̃F2 (𝑠) = 0.2102 + . (54)
FOPI controller has similar performance with the uncompensated FOPI 𝑠0.6961
controller, demonstrating that the proposed rate limit compensation The closed-loop onset frequencies and updated design specifications
approach has better robustness under system gain variation compared are listed in Table 3. Nichols charts and Nyquist plot of these controllers
with the anti-windup method. At 80 s, all controllers have good distur- are depicted in Figs. 12 and 13. The Nyquist plot of the uncompen-
bance rejection capability under output disturbance, and the proposed
sated FOPI and compensated FOPI controller do not have intersection
compensated FOPI controller stabilizes the system the quickest.
with NIDF of the rate limiter, representing the system stability. Step
To test the performance robustness of the flat phase controller under
responses and control signals of these controllers and the anti-windup
system gain variation, a ±20% variation is applied to the steady-state
FOPI controller are plotted in Fig. 14, and control performance are
gain of the DC model. Fig. 10 shows the performance of different
listed in Table 4. Please note that the anti-windup filter has been
controllers under system gain variation (controllers are designed based
changed to be 𝑘𝑎 = |𝑘𝑖 ∕𝑘𝑝 | to stabilize the system. The rate limit
on the nominal system model). All controllers has robust performance
effect is serious that the uncompensated PID controller is not stable
under gain variation. Referring to the overshoot under different gain
and compensated PID controller generates a high frequency oscillation
values of the uncompensated FOPI controller, the anti-windup com-
pensation method generates more similar overshoots under gain varia- at the steady state. FOPI controllers provide stable response and the rate
tion, dedicating that the anti-windup compensation method has better limit compensation reduces the output oscillations. The anti-windup
performance robustness than the proposed rate limit compensation FOPI controller provides smaller overshoot but more oscillations than
strategy. the proposed compensated FOPI controller. This example shows that
The real motor systems usually have deadband and resolution is- the proposed flat phase controllers may not have good performance
sues. A deadband with ±0.5 bar is cascaded to the controller and a for delay-dominant processes. There are mainly two reasons for the
quantizer with resolution 0.05 is added to the system output. System unsatisfied control performance: (1) The large phase delay resulting
output under deadband and quantizer are depicted in Fig. 11. Except from time delay in delay-dominant process brings challenge to the
for the uncompensated PID controller, other controllers provides stable controller design, and the flat phase controller design method is not the
response under the deadband and the resolution, indicating that the best choice for such a system; (2) In addition to the large time delay,
proposed rate limit compensation method generates good performance the rate limit also generates large phase delay on the delay-dominant
under deadband nonlinearity and quantification errors. process. Thus the huge phase delay generated by total system makes it
challenging to control the delay-dominant systems under the rate limit
7.2. Delay-dominant process effect. In addition, this example also shows a potential limitation of the
proposed compensation method under severe rate limit effect.
In this section, the proposed compensation method will be tested To test the robustness of these three stable controllers, a positive
on a delay-dominant FOPTD process. The system model is chosen from 20% variation is employed to system gain at 100 s, as shown in Fig. 15.
the reference Muresan, Birs, and Dulf (2020) Furthermore, to test the disturbance rejection capability, a unit step dis-
2 −2𝑠 turbance is applied to system output at 100 s, as shown in Fig. 16. From
𝐺2 (𝑠) = 𝑒 , (50)
𝑠+1 Fig. 15, the proposed compensated FOPI controller has the best ro-
and the rate limit value is 𝑅 = 0.1. The controller design specifications bustness under gain variation. Anti-windup FOPI controller has similar
are 𝜔𝑏 = 0.29 rad/s, phase margin 𝛷𝑚 = 40◦ and the flat phase performance with the uncompensated FOPI controller, demonstrating

8
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

Fig. 10. Step response under system gain variation. 𝐺1 (𝑠) = 1∕(2𝑠 + 1).

Fig. 11. Step responses (a) and control signals (b) under deadband (±0.5) and output resolution (0.05). 𝐺1 (𝑠) = 1∕(2𝑠 + 1).

again that the proposed rate limit compensation method has better ro- is shown in Fig. 17. As a thermoelectric cooler, the Peltier module
bustness to gain variation than the anti-windup compensation strategy. transfers heat and generates temperature difference when a voltage
From Fig. 16, the proposed compensated FOPI controller stabilizes the is applied to two semiconductor electrodes. A metal plate is attached
system within the shortest time. to one side of the Peltier and a non-contact infrared thermometer
(MLX90614) is installed facing the metal plate to measure the real-time
8. Experimental results temperature (Li & Chen, 2014). A fan is attached to the other side of the
Peltier to remove the heat and maintains a good working condition of
In this section, the effectiveness of the proposed rate limit compen- the Peltier. A MOSFET H-bridge circuit is used to drive the Peltier and
sation strategy and the flat phase controller design approach will be thermometer. The control diagram of the Peltier system is plotted in
validated on a Peltier module-based hardware-in-the-loop temperature Fig. 18. Control algorithms are implemented via MATLAB/Simulink
controlled experimental platform. Hardware setup of the Peltier system 2018a. Communication between MATLAB and hardware experimental

9
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

Fig. 12. Nichols charts of the open-loop system with/without rate limiter for PID controller (a) and FOPI controller (b). 𝐺2 (𝑠) = 2∕(𝑠 + 1)𝑒−2𝑠 .

Fig. 13. Nyquist plots of loop transfer functions and NIDF of rate limiter 𝐺2 (𝑠) = 2∕(𝑠 + 1)𝑒−2𝑠 .

Fig. 14. Step responses (a) and control signals (b) under flat phase PID, FOPI and anti-windup FOPI controllers. 𝐺2 (𝑠) = 2∕(𝑠 + 1)𝑒−2𝑠 .

10
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

Fig. 15. Step responses (a) and control signals (b) under system gain variation (+20% variation happens at 100 s). 𝐺2 (𝑠) = 2∕(𝑠 + 1)𝑒−2𝑠 .

Fig. 16. Step responses (a) and control signals (b) under system output disturbance (unit step disturbance happens at 100 s). 𝐺2 (𝑠) = 2∕(𝑠 + 1)𝑒−2𝑠 .

sets are realized with an Arduino Uno board. The Peltier is actuated by
a voltage whose rate limit is considered to be neglected, thus a software
rate limiter will be used to imitate the real actuator.
The temperature model of the Peltier power on heating process
under room temperature 22 ◦ C is a lag-dominant process
0.1646
𝑃 (𝑠) = 𝑒−1𝑠 , (55)
51.5028𝑠 + 1
and the software rate limit value is 𝑅 = 2.
The desired crossover frequency and phase margin are selected as
0.05 rad/s and 45◦ . The flat phase PID and FOPI controllers without
considering rate limit are obtained as
0.9397
𝐶I3 (𝑠) = 7.5258 + + 75.8636𝑠, (56)
𝑠
0.8973
𝐶F3 (𝑠) = 6.1453 + . (57)
𝑠0.9416
Fig. 17. The overall hardware setup of the Peltier system.
Bode plots of loop transfer functions are shown in Fig. 19, both gain
crossover frequency and phase margin are satisfied, and the slopes of
the two loop transfer functions equal approximately zero at 𝜔𝑏 . Follow-
ing the design procedures in Section 6, the new desired crossover fre- From Table 5, it demonstrates that the onset frequency of the rate
quency and phase margin are given in Table 5. Then the compensated limiter has been increased after compensation, and FOPI outperforms
PID controller and FOPI controller are redesigned as
PID with a larger 𝜔
̃onset . The Nichols charts of the open-loop systems
0.7354
𝐶̃I3 (𝑠) = 23.6923 + + 42.6489𝑠, (58) with and without rate limiter are illustrated in Fig. 20. The Nyquist
𝑠
curves of loop transfer functions and NIDF of the rate limiter are plotted
1.8881
𝐶̃F3 (𝑠) = 17.9130 + . (59) in Fig. 21. The Nyquist plot under the compensated FOPI controller
𝑠0.5863

11
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

Fig. 18. Control diagram of the Peltier system.

0.1646
Fig. 19. Bode plots of open-loop system with PID and FOPI controllers. 𝑃 (𝑠) = 51.5028𝑠+1
𝑒−1𝑠 .

Table 5 Table 6
Design specifications and controller parameters for 𝑃 (𝑠). Control performance of the Peltier system.
Controller 𝜔𝑏 𝛷𝑚 𝑘𝑝 𝑘𝑖 𝑘𝑑 ∕𝛼 𝜔
̃onset Controller rise time /s settling time /s overshoot /% ISE
UPID/𝐶I3 0.05 45 7.53 0.94 75.86 0.0323 UPID/𝐶I3 37.64 278.10 65.02 584.88
CPID/𝐶̃I3 0.0760 84.78 23.69 0.74 42.65 0.0379 CPID/𝐶̃I 37.64 136.74 41.67 424.29
3
UFOPI/𝐶F3 0.05 45 6.15 0.90 0.94 0.0325 UFOPI𝐶F3 37.64 218.61 52.13 488.05
CFOPI/𝐶̃F3 0.0747 83.74 17.91 1.89 0.59 0.0406 CFOPI/𝐶̃F 38.27 101.46 13.26 347.60
3

AWFOPI/𝐶A3 42.07 168.18 22.14 373.95

does not intersect NIDF, representing the stability of the closed-loop

system. proposed rate limit compensation strategy maintains the good control
The simulated step responses and control signals of PID, FOPI performance.
and anti-windup FOPI controllers are depicted in Fig. 22. Control
performance of these controllers are listed in Table 6. Compared to 9. Qualitative analysis of outperformance of FOPI controller on
PID controller, FOPI controller provides better control performance. mitigating rate limit effect
For lag-dominant processes, the proposed compensation method is
also able to effectively mitigate the rate limit effect. The proposed In this section, the advantage of the FOPI controllers on mitigating
rate limit compensation strategy also outperforms anti-windup scheme the rate limit effect will be qualitatively analyzed. The analysis is
with smaller overshoot and ISE, and better robustness and disturbance mainly focused on the possible infinity loops in the closed-loop system
rejection capability. which represents system stability.
Then these controllers are successfully applied to the real Peltier The frequency responses in low and middle frequency ranges of the
system, and corresponding output and control signals are illustrated in loop transfer function play important roles in control performance for
Fig. 23. A positive 20% variation is applied to system gain at 900 s linear systems. First, the magnitude and phase of FOPI and PID con-
and a unit step disturbance is added to the output at 1600 s. Under the trollers in the frequency range (0, 𝜔𝑏 ) will be compared. In Eq. (14), as
sensor noise and temperature disturbance from the environment, the ln coth |𝑣|
2
reduces rapidly on both sides of 𝜔0 , ∠𝐺(𝑗𝜔0 ) mainly depends

12
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

0.1646
Fig. 20. Nichols charts of the open-loop system with/without rate limiter for PID controller (a) and FOPI controller (b). 𝑃 (𝑠) = 51.5028𝑠+1
𝑒−1𝑠 .

0.1646
Fig. 21. Nyquist plots of loop transfer functions and NIDF of rate limiter. 𝑃 (𝑠) = 51.5028𝑠+1
𝑒−1𝑠 .

on the slope of the magnitude Bode plot around 𝜔0 . Assuming that the −20 dB/dec to 0 dB/dec, then one has
magnitude slope is a constant in the neighborhood of 𝜔0 , the phase ∞ dln |𝐿 (𝑗𝜔)|
∠𝐿I (𝑗𝜔𝑏 ) =
1 | I | ln coth |𝑣| d𝑣
at the specific frequency can be approximated by Eq. (15). However, 𝜋 ∫−∞ d𝑣 2
the slope of the magnitude plot at frequencies below and above 𝜔0 also 𝜔𝑐 dln |𝐿 (𝑗𝜔)|
1 | I | ln coth |𝑣|
contributes to the phase value ∠𝐺(𝑗𝜔0 ). Here, we assume that the slopes = d𝑣 (60)
𝜋 ∫−∞ d𝑣 2
of the magnitude Bode plot for PID/FOPI controllers are similar in ∞ dln |𝐿 (𝑗𝜔)|
+
1 | I | ln coth |𝑣| d𝑣.
frequency range (0, 𝜔𝑏 ). Due to the derivative element in PID controller, 𝜋 ∫𝜔𝑐 d𝑣 2
the magnitude slope of loop transfer function under PID controllers in
Under PID control, the magnitude slope of the loop transfer function
high frequency is more gentle than the slope under FOPI controllers. equals 0 dB/dec at high frequencies (𝜔𝑐 , ∞), which is greater than the
For instance, for the first-order process, the magnitude slope of the loop slope at 𝜔𝑏 (generally equals to −20 dB/dec). That means
transfer function under PID controllers is equivalent to zero in high
dln ||𝐿I (𝑗𝜔𝑏 )|| dln ||𝐿I (𝑗𝜔)||
frequency while the slope under FOPI controllers is −20 dB/dec, as < , (𝜔≥ 𝜔𝑐 ), (61)
d𝑣 d𝑣
shown in Fig. 19. Thus under PID controller, the magnitude slope in
then one has
high frequencies has an effect on the phase of loop transfer functions. ∞ dln |𝐿 (𝑗𝜔)|
∠𝐿I (𝑗𝜔𝑏 ) =
1 | I | ln coth |𝑣| d𝑣
Take the FOPTD process as an example: assume that the inflection 𝜋 ∫−∞ d𝑣 2
(62)
point of the magnitude Bode plot of the loop transfer function under dln ||𝐿I (𝑗𝜔𝑏 )|| 1 ∞
|𝑣|
PID control locates at 𝜔𝑐 (𝜔𝑐 > 𝜔𝑏 ) rad/s, where the slope changes from > ln coth d𝑣.
d𝑣 𝜋 ∫−∞ 2

13
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

0.1646
Fig. 22. Step responses (a) and control signals (b) of the closed-loop systems under uncompensated/compensated PID/FOPI controllers. 𝑃 (𝑠) = 51.5028𝑠+1
𝑒−1𝑠 .

0.1646
Fig. 23. Step responses (a) and control signals (b) of the Peltier system. 𝑃 (𝑠) = 51.5028𝑠+1
𝑒−1𝑠 .

Under FOPI control, the magnitude slope of loop transfer function at According to the Eq. (14), it has
high frequencies is close to the slope at 𝜔𝑏 , thus it can be approximately
written that ∠𝐿I (𝑗𝜔) < ∠𝐿F (𝑗𝜔) < 0, (0 < 𝜔 < 𝜔𝑏 ). (67)
∞ dln |𝐿 (𝑗𝜔)|
∠𝐿F (𝑗𝜔𝑏 ) =
1 | F | ln coth |𝑣| d𝑣 The above relation of the magnitude and phase value under PID and
𝜋 ∫−∞ d𝑣 2 FOPI controllers can also be validated from Fig. 19.
(63)
dln ||𝐿F (𝑗𝜔𝑏 )|| 1 ∞
|𝑣|
= ln coth d𝑣. Based on the above analysis, the Nyquist curve of the loop transfer
d𝑣 𝜋 ∫−∞ 2
function under FOPI controllers should be closer to the negative imag-
Under the same design specifications, it has ∠𝐿I (𝑗𝜔𝑏 ) = ∠𝐿F (𝑗𝜔𝑏 ), inary axis than the Nyquist curve under PID controllers, as shown in
then
Fig. 21, resulting in a lower possibility to interact with the NIDF of
dln ||𝐿I (𝑗𝜔𝑏 )|| dln ||𝐿F (𝑗𝜔𝑏 )|| the rate limiter. Thus, FOPI controllers outperform PID controllers in
< < 0, (64)
d𝑣 d𝑣 mitigating actuator rate limit effects with a lower possibility of infinity
It should be noted that the magnitude slope of controllers at the loops in the closed-loop system.
crossover frequency is negative.
It is observed that the magnitude slope of 𝐿F (𝑗𝜔) is gentler than
The above equation represents the magnitude slope under FOPI
𝐿I (𝑗𝜔), and generally the magnitude slope under PID controllers is
controller being more gentle than the slope for PID controllers. Under
−20 dB/dec. Then, the magnitude slope under FOPI controllers is
the general assumption that the magnitude slope of the loop transfer
function maintains almost consistently in low frequencies and middle greater than −20 dB/dec, thus the order of the integrator in FOPI con-
frequencies, then it has troller should be smaller than 1, which is validated from the illustrative
example. Additionally, the gentle slope under FOPI controllers could
dln ||𝐿I (𝑗𝜔)|| dln ||𝐿F (𝑗𝜔)||
< < 0, (0 < 𝜔 < 𝜔𝑏 ), (65) lead to a wider range of middle frequencies, leading to a higher band-
d𝑣 d𝑣
width. Furthermore, the magnitude slope of the loop transfer function
and
under FOPI controllers is deeper than that of PID controllers, which
|𝐿F (𝑗𝜔)| < |𝐿I (𝑗𝜔)| , (0 < 𝜔 < 𝜔𝑏 ). (66) shows that FOPI controllers have higher capability in noise reduction.
| | | |

14
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

10. Conclusions Chen, P.-C., & Shamma, J. S. (2004). Gain-scheduled 𝑙1 -optimal control for boiler-
turbine dynamics with actuator saturation. Journal of Process Control, 14(3),
263–277.
This paper proposes a fractional-order flat phase feedback controller
Duda, H. (1994). Frequency domain analysis of rate limiting elements in flight control
design approach and an actuator rate limit compensation strategy for systems. Forschungsbericht- Deutsche Forschungsanstalt Fuer Luft- Und Raumfahrt,
general minimum-phase systems. The simplified implementation of the DLR-FB, (94), 40.
flat phase constraint is developed based on the Bode’s integrals to Eldigair, Y., Garelli, F., Kunusch, C., & Ocampo-Martinez, C. (2020). Adaptive PI
control with robust variable structure anti-windup strategy for systems with rate-
avoid tedious parametric derivation. Update rules for controller design
limited actuators: Application to compression systems. Control Engineering Practice,
specifications are established based on the Bode’s integrals to mitigate 96, Article 104282.
the actuator rate limit effect. A PID controller and an FOPI controller Flesch, R. C., Normey-Rico, J. E., & Flesch, C. A. (2017). A unified anti-windup strategy
are initially designed by neglecting rate limit effect, then redesigned for SISO discrete dead-time compensators. Control Engineering Practice, 69, 50–60.
Gilbreath, G. P. (2001). Prediction of pilot-induced oscillations (PIO) due to actuator
based on new specifications. The magnitude and phase value of the loop
rate limiting using the open-loop onset point (olop) criterion. Tech. rep., Air Force
transfer function under the two controllers are qualitatively compared, Inst of Tech Wright-Pattersonafb Oh.
demonstrating that FOPI controllers outperform PID controllers with Herbst, G. (2015). Practical active disturbance rejection control: Bumpless transfer, rate
lower possibility of infinity loops under actuator rate limit effects. The limitation, and incremental algorithm. IEEE Transactions on Industrial Electronics,
proposed rate limit compensation method is tested on a fast responding 63(3), 1754–1762.
Jin, X., Qin, J., Shi, Y., & Zheng, W. X. (2017). Auxiliary fault tolerant control with
model, a delay-dominant process and a sluggish lag-dominant process actuator amplitude saturation and limited rate. IEEE Transactions on Systems, Man,
through two illustrative examples and the hardware-in-the-loop exper- and Cybernetics: Systems, 48(10), 1816–1825.
imental platform. The results have demonstrated the effectiveness the Karimi, A., Garcia, D., & Longchamp, R. (2002). PID controller design using Bode’s
proposed compensation method on three different kinds of processes. integrals. In Proceedings of the 2002 American control conference: vol. 6, (pp.
5007–5012). IEEE.
The compensation of rate limit on delay-dominant processes is rel- Karimi, A., Garcia, D., & Longchamp, R. (2003). PID controller tuning using Bode’s
atively poorer than the lag-dominant processes and fast responding integrals. IEEE Transactions on Control Systems Technology, 11(6), 812–821.
systems, which remains challenging due to the huge phase delay re- Keziz, B., Djouambi, A., & Ladaci, S. (2020). A new fractional order controller tuning
sulting from both time delay and rate limit. The proposed rate limit method based on Bode’s ideal transfer function. International Journal of Dynamics
and Control, 8(3), 932–942.
compensation strategy are demonstrated to have better robustness to
Li, Z., & Chen, Y. Q. (2014). Identification of linear fractional order systems using the
system gain variation and better disturbance rejection capability com- relay feedback approach. In IEEE American control conference (pp. 3704–3709).
pared with the anti-windup scheme. Additionally, the results have also Li, Y., & Lin, Z. (2018). Stability and performance of control systems with actuator
validated the superiority of FOPI controllers than PID controllers when saturation. Springer.
Luo, Y., & Chen, Y. Q. (2012). Stabilizing and robust fractional order PI controller
systems suffer from actuator rate limit effect. The Bode’s integrals are
synthesis for first order plus time delay systems. Automatica, 48(9), 2159–2167.
only applicable in minimum-phase systems, thus a potential pitfall of Maione, G., & Lino, P. (2007). New tuning rules for fractional PI𝛼 controllers. Nonlinear
the proposed method is the limited application on minimum-phase lin- Dynamics, 49(1), 251–257.
ear time-invariant systems. While the proposed method is an effective Monje, C. A., Chen, Y., Vinagre, B. M., Xue, D., & Feliu-Batlle, V. (2010). Fractional-order
systems and controls: Fundamentals and applications. Springer Science & Business
approximation of flat phase, we will improve this method for a more
Media.
accurate implementation by considering the effects of Bode plot slopes Muresan, C. I., Birs, I. R., & Dulf, E. H. (2020). Event-based implementation of fractional
in low and high frequencies. order IMC controllers for simple FOPDT processes. Mathematics, 8(8), 1378.
Petráš, I. (2011). Fractional-order nonlinear systems: Modeling, analysis and simulation.
Springer Science & Business Media.
Declaration of competing interest
Răducan, C., & Neag, M. (2021). Slew-rate booster and frequency compensation circuit
for automotive LDOs. IEEE Transactions on Circuits and Systems. I. Regular Papers,
The authors declare that they have no known competing finan- 1–13. http://dx.doi.org/10.1109/TCSI.2021.3094897.
cial interests or personal relationships that could have appeared to Rundqwist, L., & Stahl-Gunnarsson, K. (1996). Phase compensation of rate limiters in
unstable aircraft. In Proceeding of the 1996 IEEE international conference on control
influence the work reported in this paper.
applications IEEE international conference on control applications held together with
IEEE international symposium on intelligent control (pp. 19–24).
Acknowledgments Sahu, A., Mahodaya, A. K., & Chokkadi, S. (2020). Slew rate and control constrained
spacecraft attitude maneuver with reaction wheel failure. In Advances in small
satellite technologies (pp. 119–131). Springer.
This work was supported by the National Natural Science Founda-
Seyedtabaii, S. (2017). New flat phase margin fractional order PID design: Perturbed
tion of China under Grant 62103098. UAV roll control study. Robotics and Autonomous Systems, 96, 58–64.
Shin, H.-B. (1998). New antiwindup PI controller for variable-speed motor drives. IEEE
References Transactions on Industrial Electronics, 45(3), 445–450.
Sofrony, J., Turner, M. C., & Postlethwaite, I. (2009). A simple anti-windup compensa-
tion scheme for systems with rate-limited actuators. In IEEE ICCAS-SICE international
Åström, K. J., Hägglund, T., & Astrom, K. J. (2006). Advanced PID control, vol. 461. joint conference (pp. 3311–3316).
ISA-The Instrumentation, Systems, and Automation Society Research Triangle Park. Tepljakov, A., Alagoz, B. B., Yeroglu, C., Gonzalez, E., HosseinNia, S. H., & Petlenkov, E.
Barbosa, R. S., Machado, J., & Ferreira, I. M. (2004). Tuning of PID controllers based (2018). FOPID controllers and their industrial applications: A survey of recent
on Bode’s ideal transfer function. Nonlinear Dynamics, 38(1), 305–321. results. IFAC-PapersOnLine, 51(4), 25–30.
Blank, M., Glück, T., Kugi, A., & Kreuter, H.-P. (2014). Slew rate control strategies for Tran, A. T., Sakamoto, N., Kikuchi, Y., & Mori, K. (2017). Pilot induced oscillation
smart power ICs based on iterative learning control. In Applied power electronics suppression controller design via nonlinear optimal output regulation method.
conference and exposition (pp. 2860–2866). IEEE. Aerospace Science and Technology, 68, 278–286.
Bode, H. W. (1945). Network analysis and feedback amplifier design. D. Van Nostrand von Ellenrieder, K. D. (2019). Dynamic surface control of trajectory tracking marine
New York. vehicles with actuator magnitude and rate limits. Automatica, 105, 433–442.
Brieger, O., Kerr, M., Leißling, D., Postlethwaite, I., Sofrony, J., & Turner, M. (2009). Wang, S., Gao, Y., Liu, J., & Wu, L. (2018). Saturated sliding mode control with limited
Flight testing of a rate saturation compensation scheme on the ATTAS aircraft. magnitude and rate. IET Control Theory & Applications, 12(8), 1075–1085.
Aerospace Science and Technology, 13(2–3), 92–104. Wu, Z., Yuan, J., Li, D., Xue, Y., & Chen, Y. (2020). The influence of rate limit
Buchholz, J. J. (1993). Time-delay induced by control surface rate saturation. Zeitschrift on proportional–integral controller for first-order plus time-delay systems. ISA
Fur Flugwissenschaften Und Weltraumforschung, 17(5), 287–293. Transactions, 105, 157–173.
Caponetto, R. (2010). Fractional order systems: Modeling and control applications, vol. 72. Wu, Z., Yuan, J., Liu, Y., Li, D., & Chen, Y. (2021). An active disturbance rejection
World Scientific. control design with actuator rate limit compensation for the ALSTOM gasifier
Chalk, C. (1992). Study of a software rate limit concept. Calspan Flight Research benchmark problem. Energy, 227, Article 120447.
Memorandum, (636). Yu, Z., Zhang, Y., Jiang, B., Su, C.-Y., Fu, J., Jin, Y., et al. (2021). Fractional
Chen, Y. Q., Hu, C., & Moore, K. L. (2003). Relay feedback tuning of robust PID order PID-based adaptive fault-tolerant cooperative control of networked unmanned
controllers with iso-damping property. In 42nd IEEE international conference on aerial vehicles against actuator faults and wind effects with hardware-in-the-loop
decision and control: vol. 3, (pp. 2180–2185). experimental validation. Control Engineering Practice, 114, Article 104861.

15
J. Yuan, S. Fei, Y. Chen et al. Control Engineering Practice 129 (2022) 105347

Yuan, J., Chen, Y. Q., & Fei, S. (2018). Analysis of actuator rate limit effects on first- Yuan, J., Wu, Z., Fei, S., & Chen, Y. Q. (2019). Hybrid model-based feed-
order plus time-delay systems under fractional-order proportional-integral control. forward and fractional-order feedback control design for the benchmark
IFAC-PapersOnLine, 51(4), 37–42. refrigeration system. Industrial and Engineering Chemistry Research, 58(38),
Yuan, J., Ding, Y., Fei, S., & Chen, Y. Q. (2022). Identification and parameter sensitivity 17885–17897.
analyses of time-delay with single-fractional-pole systems under actuator rate limit Yumuk, E., Güzelkaya, M., & Eksin, İ. (2019). Analytical fractional PID controller
effect. Mechanical Systems and Signal Processing, 163, Article 108111. design based on Bode’s ideal transfer function plus time delay. ISA Transactions,
Yuan, J., Fei, S., & Chen, Y. Q. (2020). Compensation strategies based on Bode 91, 196–206.
step concept for actuator rate limit effect on first-order plus time-delay systems. Zheng, W., Luo, Y., & Chen, Y. Q. (2020). A fractional order controller design based
Nonlinear Dynamics, 1–16. on Bode’s ideal transfer function and Bode’s ideal cut-off ideas. IFAC-PapersOnLine,
Yuan, J., Fei, S., & Chen, Y. (2021). On the actuator rate limit effect in reaction curves. 53(2), 3663–3668.
ISA Transactions, 117, 303–308. Zhou, S., Zheng, M., & Tomizuka, M. (2016). A generalized anti-windup scheme con-
Yuan, J., Han, J., Chai, L., Fei, S., & Chen, Y. Q. (2020). First-order plus time-delay sidering amplitude and rate saturations. In ASME 2016 dynamic systems and control
systems under the effects of actuator rate limit. IET Control Theory & Applications, conference: vol. 2, American Society of Mechanical Engineers Digital Collection,
14(17), 2481–2490. V002T22A007.

16