Journal of Sound and Vibration 571 (2024) 118083

Contents lists available at ScienceDirect

Journal of Sound and Vibration

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

Control design, analysis, and optimization of fractional-order

delayed resonator for complete vibration absorption
Jiazhi Cai a , Qingbin Gao a ,∗, Yifan Liu a , Nejat Olgac b
a School of Mechanical Engineering and Automation, Harbin Institute of Technology (Shenzhen), Shenzhen, China
b Department of Mechanical Engineering, University of Connecticut, Storrs, CT, USA

ARTICLE INFO ABSTRACT

Keywords: The active vibration absorber can eliminate the vibration energy entirely from a primary system
Active vibration absorber through its anti-resonance effect, which is widely applied to vibration-sensitive scenarios. This
Delayed resonator article focuses on the control design, analysis, and optimization problems for a novel vibration-
Vibration absorption
absorption technique named fractional-order delayed resonator. Specifically, we develop a
Fractional delayed control
parameterized form for such a fractional delayed control logic and provide a detailed analysis
concerning the control parameters. By the frequency-domain integer-order approximations, we
design a control framework that applies to any of the position, velocity, and acceleration
signal sources. Then, the particle swarm optimization algorithm is utilized to minimize the
transmissibility error at the excitation frequency and determine the optimal approximation
frequency band. Additionally, we discuss the stability preservation concern brought by different
approximation configurations. Finally, we treat the order of the fraction as an additional tunable
parameter to enhance the frequency sensitivity under delay discretization. The overall concept
and analysis are validated by numerical simulations.

1. Introduction

Humans cannot downplay the detrimental effects brought by mechanical vibrations to the daily life and industrial production. The
investigations regarding vibration suppression and energy-transfer mechanism are of both theoretical significance and application
value. The vibration-absorption technique focuses on the elimination or minimization of the vibration energy by the dynamic
vibration absorber(s) mounted on the primary structure. The reaction force of the absorber can partially or completely neutralize
the harmonic oscillations from the base, leading to the energy absorption. Both passive and active schemes are widely used for
the vibration suppression task, see [1–5]. The cardinal advantages of the passive approach are the inherent stability of the overall
system and the dispensability for the energy to damp the oscillations. However, the operable frequency band, centered at the natural
frequency of the absorber, is relatively narrow. Additionally, residual oscillations certainly exist owing to the absorber’s non-zero
damping, even if its natural frequency perfectly matches the excitation frequency.
Let us introduce an outstanding passive–active hybrid vibration control strategy named delayed resonator (DR), which is invented
by Prof. Olgac in the 1990s [6]. This strategy utilizes an additional actuator installed on the passive absorber to secure an ideal
resonator by the delayed state feedback. Through the anti-resonance effect, the resonator can absorb the vibrations entirely and is
real-time tunable to track the time-varying excitation frequency. In recent years, researchers are keen to refine the DR technique
to improve control performance or accommodate diverse control needs, see [7–16]. The distributed delay-based DR models are
introduced in [7–9] to reduce the noise, circumvent the neutrality, and widen the operable frequency band. By the DR with

∗ Corresponding author.
E-mail address: [email protected] (Q. Gao).

https://doi.org/10.1016/j.jsv.2023.118083
Received 12 March 2023; Received in revised form 19 August 2023; Accepted 5 October 2023
Available online 19 October 2023
0022-460X/© 2023 Elsevier Ltd. All rights reserved.
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 1. (a) Primary structure coupled with an active absorber. (b) Absorber fixed to the ground.

polynomial-distributed delays, an optimized design is achieved in [10] to minimize the sensitivity of the resonator performance
with respect to variations of the excitation frequency. The authors in [11] demonstrate that a single resonator can damp the
multi-frequency vibrations by the multiple delayed feedback. Readers are referred to [13,14] for the planar vibration control by
a newly-designed two-dimensional DR. Our recent work indicates that the DR design with an amplifying mechanism significantly
enhances the performance and robustness of the coupled structure [15]. The tuning and stability evaluation guidelines for the
DR with virtual natural-frequency adjustment are proposed in [16]. We note that these publications are mainly based on the
refinement of the mechanical structure or the control algorithm. Obviously, the latter approach is more economical since it avoids
the direct change of the physical structure and applies to various control scenarios. However, the associated control design becomes
prohibitively intricate, which greatly increases the analysis difficulty and reduces the usability. Therefore, we are expecting a more
intuitive and concise extended control logic to further polish the DR technique.
In the common sense, only the displacement, velocity, and acceleration feedback are allowed for the generation of the vibration
control signal. The fractional calculus theory breaks this inertia by removing the integer-order restrictions on the system modeling
and feedback formation, which receives much attention from the researchers in recent years [17,18]. Readers can also refer to
the implementation of the fractional controller for systems including but not limited to the vehicle suspension system [19,20],
flight control system [21,22], electromechanical gyrostat system [23], and seismic transducer [24]. Additionally, various fractional
vibration control strategies are reported in the pioneer studies [25–28]. The authors are motivated by this popular theory and attempt
to develop a fractional delayed vibration-absorption technique based on the basic DR framework, what we call the fractional-order
DR (FODR). Since adjusting the order is an intuitive and non-fragmented operation, the FODR technique provides one additional
degree of freedom in control with the freely tunable fractional order. In our preliminary study [29], we demonstrate that it can widen
the feasible frequency band, improve the damping speed, and enhance the stability robustness. However, our knowledge for the
implementation details is still inadequate. Consequently, it is abstract for engineers to imagine an intermediate state that exists
between the position and velocity feedback (or between the velocity and acceleration feedback).
The main objective of this study is to address the design, analysis, and optimization control problems for the FODR technique
in a comprehensive manner following the recent outcome [29]. This will visualize such a fascinating but abstract control logic and
help readers understand how the order adjustment improves the flexibility and robustness for vibration control. The structure of the
paper is as follows. In Section 2, we review the mechanism of the control algorithm design for the active vibration absorber. The
parameterization and analysis of the control variables for the FODR technique are provided in Section 3. In Section 4, we explain the
effect of the polarity inversion on the control variables, and thus examine the parameter consistency. The control framework design
with multi-source signals by the integer-order approximations is presented in Section 5. We also address the associated problems
of frequency-response optimization and controller discretization in this section. In Section 6, we develop an algebraic method for
the evaluation of the stability robustness, and then discuss the stability preservation concern brought by different approximation
parameters. Finally, the order adjustment scheme is utilized to improve the frequency sensitivity under the delay discretization in
Section 7.
For notational simplicity, we use R, C, and N+ for the set of real numbers, complex numbers, and positive integers, respectively.
We use ℜ(⋅) and ℑ(⋅) for the real part and imaginary part of an equation or a variable, respectively.

2. Preliminaries on active vibration absorption

Let us take a closer look at the vibratory system displayed in Fig. 1(a) where an active absorber is attached to a single degree-of-
freedom primary structure. The feedback force 𝑢(𝑡) is exerted by the actuator and we secure a passive absorber when the feedback
force is set to zero. Let the excitation force be a harmonic sinusoidal signal:

𝑓 (𝑡) = 𝐴𝑓 sin(𝜔𝑡) (1)

where 𝐴𝑓 [N] is the force amplitude and 𝜔 > 0 [rad/s] is the excitation frequency. According to Newton’s second law, the dynamical
equations of the coupled system are formulated as

𝑚𝑎 𝑥̈ 𝑎 (𝑡) + 𝑐𝑎 𝑥̇ 𝑎 (𝑡) + 𝑘𝑎 𝑥𝑎 (𝑡) − 𝑐𝑎 𝑥̇ 𝑝 − 𝑘𝑎 𝑥𝑝 = 𝑢(𝑡), (2)

2
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

𝑚𝑝 𝑥̈ 𝑝 (𝑡) + (𝑐𝑎 + 𝑐𝑝 )𝑥̇ 𝑝 (𝑡) + (𝑘𝑎 + 𝑘𝑝 )𝑥𝑝 (𝑡) − 𝑐𝑎 𝑥̇ 𝑎 (𝑡) − 𝑘𝑎 𝑥𝑎 (𝑡) = −𝑢(𝑡) + 𝑓 (𝑡) (3)

where 𝑥 [m] is the displacement, 𝑚 [kg], 𝑐 [kg s−1 ], and 𝑘 [N m−1 ] are the mass, damping, and stiffness parameters. The subscripts
𝑎 and 𝑝√
represent the absorber and primary base, respectively. The natural frequency and damping ratio of the absorber are given
𝑘𝑎 𝑐
as 𝛺 = [rad/s] and 𝜁 = √𝑎 , respectively.
𝑚𝑎 2 𝑚𝑎 𝑐𝑎
{ } { }
Consider the Laplace transforms of the variables: 𝑋𝑎 (𝑠) =  𝑥𝑎 (𝑡) , 𝑋𝑝 (𝑠) =  𝑥𝑝 (𝑡) , 𝑈 (𝑠) =  {𝑢(𝑡)}, and 𝐹 (𝑠) =  {𝑓 (𝑡)} where
𝑠 is the Laplace variable. Let the control logic be constructed by the state feedback of the absorber, i.e.,

𝑈 (𝑠) = 𝑄(𝑠)𝑋𝑎 (𝑠) (4)

where 𝑄(𝑠) is the feedback function. The closed-loop transfer function from 𝑓 to 𝑥𝑝 under zero initial conditions is written as
𝑋𝑝 (𝑠) 𝑅(𝑠)
𝐺𝑓 𝑝 (𝑠) = = (5)
𝐹 (𝑠) 𝑃 (𝑠)𝑅(𝑠) + 𝑉 (𝑠)
where 𝑅(𝑠) = 𝑚𝑎 𝑠2 +𝑐𝑎 𝑠+𝑘𝑎 −𝑄(𝑠), 𝑃 (𝑠) = 𝑚𝑝 𝑠2 +𝑐𝑝 𝑠+𝑘𝑝 , 𝑉 (𝑠) = 𝑚𝑎 (𝑐𝑎 𝑠+𝑘𝑎 )𝑠2 . The gain of the transfer function should be |𝑅(𝑗𝜔)| = 0
with 𝑠 = 𝑗 Ω to fully suppress the response of the primary mass. Consequently, the output (displacement) signal 𝑥𝑝 vanishes when
the input (force) signal 𝑓 is transferred in this channel. At this point, the primary mass is motionless, and the equivalent dynamical
model of the absorber is shown in Fig. 1(b). If the absorber is passive and undamped, we have

𝑅(𝑗𝜔) = −𝜔2 𝑚𝑎 + 𝑘𝑎 , (6)

which vanishes when the excitation frequency perfectly matches the natural frequency of the absorber, i.e., 𝜔 = 𝛺. For the actively
tuned absorber, we have the following equation.

𝑅(𝑗𝜔) = −𝜔2 𝑚𝑎 + 𝑗𝑐𝑎 𝜔 + 𝑘𝑎 − 𝑄(𝑗𝜔). (7)

By comparing (6) with (7), we see that the goal of the active feedback is to eliminate the damping 𝑐𝑎 and virtually change the mass
𝑚𝑎 , or the stiffness 𝑘𝑎 , or both. More specifically, any feedback satisfies

ℜ {𝑄(𝑗𝜔)} = 𝑘𝑎 − 𝜔2 𝑚𝑎 , ℑ {𝑄(𝑗𝜔)} = 𝑐𝑎 𝜔, (8)

can lead to an ideal vibration suppression. Position, velocity, and acceleration signals are the three basic feedback components.
Due to the requirement of (8), we need to combine the velocity feedback with either the position or acceleration feedback for the
delay-free control logic, e.g., the proportional–derivative (PD) type controller:

𝑢(𝑡) = 𝑔𝑝 𝑥𝑎 (𝑡) + 𝑔𝑣 𝑥̇ 𝑎 (𝑡) (9)

with 𝑔𝑝 = 𝑘𝑎 − 𝜔2 𝑚𝑎 and 𝑔𝑣 = 𝑐𝑎 . For the DR technique, the control signal can be generated by only one type of the feedback:

𝑢(𝑡) = 𝑔𝑥𝑎 (𝑡 − 𝜏), 𝑢(𝑡) = 𝑔 𝑥̇ 𝑎 (𝑡 − 𝜏), or 𝑢(𝑡) = 𝑔 𝑥̈ 𝑎 (𝑡 − 𝜏) (10)

where 𝑔 is the gain parameter and 𝜏 is the delay parameter. The delay-induced transcendental term 𝑒−𝜏𝑠 in (10) concurrently touches
the real and imaginary parts in (8), i.e., 𝑒−𝑗𝜔𝜏 = cos(𝜔𝜏) − 𝑗 sin(𝜔𝜏), which is the beauty of the DR technique.
We use the transmissibility (or amplitude magnification factor) of the primary system to measure the vibration attenuation with
respect to the variations of the excitation frequency:
| 𝑋𝑝 (𝑗𝜔) |
| | | |
𝑇 𝑟(𝜔) = | | = 𝑘𝑝 |𝐺𝑓 𝑝 (𝑗𝜔)| . (11)
| 𝐹 (𝑗𝜔)∕𝑘𝑝 | | |
| |
Aiming at practical applications, we borrow the physical parameters from an experimental platform established in [7,8] unless
otherwise stated, which are shown in Table 1. Then, we compare the vibration-suppression performance of the passive absorber
with that of the active one tuned by (9) and (10). The normalized transmissibility curves are shown in Fig. 2 where 𝜔̄ = 𝜔∕𝛺 is the
dimensionless frequency. Fig. 2 indicates that the actively tuned absorber can overcome its physical constraints to track the varying
frequency and damp vibrations entirely, which needs at least two free control parameters. However, a serious drawback of the
delay-free controllers is that they are intolerant of the computation and communication lags in the control loop. The performance
of the control law (9) would be unsatisfactory if the system experiences an inherent delay 𝜏𝑠𝑦𝑠 > 0. For the DR technique, we can
compensate a delay 𝜏𝑙𝑜𝑜𝑝 in the real-time control loop so that (10) works under

𝜏 = 𝜏𝑠𝑦𝑠 + 𝜏𝑙𝑜𝑜𝑝 , (12)

which is more cost-effective than reducing the inherent delay for the dynamics.

Remark 1. We emphasize that the utilization of active control may result in adverse effects on the vibration suppression in the
frequency bands (points) far from the target frequency. As shown in Fig. 2, both peaks of the transmissibility curve corresponding
to an active absorber exhibit an increase, illustrating a trade-off involved in achieving the ideal absorption performance.

3
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Table 1
Physical parameters of the coupled system borrowed form [7] and [8].
Mass 𝑚 [kg] Damping 𝑐 [kg s−1 ] Stiffness 𝑘 [N m−1 ]
Absorber 0.223 350 1.273
Primary structure 1.52 1960 10.11

Fig. 2. Comparison of normalized transmissibility curves. Controller parameters are specified as 𝑔𝑝 = −154 kg s−2 and 𝑔𝑣 = 1.273 kg s−1 for the PD-type feedback
in (9) and 𝑔 = 0.0732 kg and 𝜏 = 0.079 s for the DR with acceleration feedback in (10).

3. Parameterization and analysis of the FODR technique

3.1. Control logic construction

We extend the integer-order control logic in (10) to the fractional-order one and provide its parameterized form in this subsection.
Without loss of generality, the absorber displacement is exploited to formulate the feedback signal:

𝑢(𝑡) = 𝐷𝛼 𝑥𝑎 (𝑡 − 𝜏) (13)

where 𝐷𝛼 (⋅)
represents the fractional operator and 𝛼 is the feedback order. We take the order as a non-negative real number for
symbolic consistency such that 𝐷𝛼 (⋅) and 𝐷−𝛼 (⋅) denote the fractional differentiation and integration, respectively. The fractional
operator is temporarily defined in an abstract way since it is typically realized by the frequency-domain approximations [30],
which will be introduced in Section 5. The reasonable range of the feedback order in (13) is 𝛼 ∈ [0, 2] due to the causality and the
requirement of physical realization of the second-order resonator [31].
We next show how to design the control parameters, i.e., the feedback order, gain, and delay, to achieve an ideal absorber under
the feedback (13). The Laplace transform of (13) under zero initial conditions is written as

𝑈 (𝑠) = 𝑔𝑠𝛼 𝑒−𝜏𝑠 𝑋𝑎 (𝑠) (14)

so that

𝑅(𝑠) = 𝑚𝑎 𝑠2 + 𝑐𝑎 𝑠 + 𝑘𝑎 − 𝑔𝑠𝛼 𝑒−𝜏𝑠 . (15)

The biggest trouble is the treatment of such a fractional transcendental equation for the quasi-stability of the absorber. The fractional
term becomes 𝑠𝛼 = (𝑗𝜔)𝛼 when the imaginary root is assigned to (15). This obstacle can be solved by realizing that the complex
plane is the most ordinary Riemann surface [32], leading to the following theorem.

Theorem 1 ([29]). The parameterized control variables for the FODR technique are given as
√
1
𝑔(𝜔, 𝛼) = 𝛼 (𝑐𝑎 𝜔)2 + (𝑚𝑎 𝜔 − 𝑘𝑎 )2 [kg s𝛼−2 ], (16)
𝜔
( ( ) )
1 𝛼𝜋 𝑐𝑎 𝜔
𝜏(𝜔, 𝛼) = − tan−1 + 2(𝑙 − 1)𝜋 [s] (17)
𝜔 2 𝑘𝑎 − 𝑚𝑎 𝜔2

where 𝑔 > 0 and 𝑙 ∈ N+ is the branch number associated with the transcendentality.

Proof. Let us place the marginally stable root in the principal Riemann sheet, i.e.,
𝜋
𝑠 = 𝜔𝑒𝑗 2 = 𝑗𝜔 (18)

to (15). The following equation holds:

𝛼𝜋
)
𝑔𝜔𝛼 𝑒𝑗(−𝜏𝜔+ 2 = (𝑘𝑎 − 𝑚𝑎 𝜔2 ) + 𝑗𝑐𝑎 𝜔. (19)

4
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 3. Normalized transmissibility curves under ‘‘frequency tracking control’’ by the first-branch parameters in (16) and (17).

Fig. 4. Normalized transmissibility curves under ‘‘order adjustment control’’ by the first-branch parameters in (16) and (17).

According to the magnitude and phase equality, we have

√
𝑔𝜔𝛼 = (𝑐𝑎 𝜔)2 + (𝑚𝑎 𝜔 − 𝑘𝑎 )2 , (20)

𝛼𝜋
∠(𝑔𝜔𝛼 ) + ∠𝑒(−𝜏𝜔+ 2
)
= ∠(𝑘𝑎 − 𝑚𝑎 𝜔2 + 𝑗𝑐𝑎 𝜔). (21)

Consequently, the positive gain parameter in (16) and the corresponding delay parameter in (17) are obtained by (20) and (21),
respectively. □

The control logic is constructed only by the intrinsic physical parameters (mass, damping, and stiffness) of the absorber, excitation
frequency, and feedback order as per Theorem 1. Compared to the integer-order feedback in (10), the proposed control law imparts
one additional degree of control freedom since the feedback order can be arbitrarily adjustable within the range of [0,2]. Nonetheless,
this extension does not significantly increase the complexity of the control logic. Figs. 3 and 4 exhibit the transmissibility curves
under the frequency tracking control (with a fixed feedback order) and the order adjustment control (with a fixed excitation
frequency) by the first-branch parameters. As depicted, we can achieve a redundant control in both scenarios given that the feedback
order is flexibly regulated. In what follows, we present the dimensionless control variables to simplify the analysis.

3.2. Nondimensionalization

We follow the procedure in [33] to create the dimensionless model. The dimensionless control law for (13) is written as

𝑢( ̄ 𝛼 𝑥𝑎 (𝑡̄ − 𝜏)
̄ 𝑡̄) = 𝑔𝐷 ̄ (22)

where 𝑡̄ = 𝛺𝑡 [−] is the dimensionless time, 𝑔̄ and 𝜏̄ are the corresponding control variables. Then, (15) is transformed to
̄ = 𝑠2 + 2𝜁 𝑠 + 1 − 𝑔𝑠
𝑅(𝑠) ̄ 𝛼 𝑒−𝜏𝑠
̄
. (23)

Similar to the proof of Theorem 1, we obtain the dimensionless form of (16) and (17) as
√
1 2
𝑔( ̄ 𝛼) = 𝛼 (𝜔̄ 2 − 1) + 4𝜁 2 𝜔̄ 2 ,
̄ 𝜔, (24)
𝜔̄

5
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 5. Gain variations with respect to the excitation frequency and feedback order for 𝜁 = 0.2 and 𝜁 = 0.8.

( ( ) )
1 𝛼𝜋 2𝜁 𝜔̄
𝜏( ̄ 𝛼) =
̄ 𝜔, − tan−1 + 2(𝑙 − 1)𝜋 , 𝑙 ∈ N+ . (25)
𝜔̄ 2 1 − 𝜔̄ 2
The conversion of the dimensionless parameters into the dimensional ones are expressed as
𝑘𝑎 1
𝑔= 𝑔̄ = 𝑚𝑎 𝛺2−𝛼 𝑔,
̄ 𝜏 = 𝜏, ̄ (26)
𝛺𝛼 𝛺
which is achieved by comparing (16) to (24) and (17) to (25). With the nondimensionalization, we reduce the analysis of three
free parameters (𝑚𝑎 , 𝑐𝑎 , and 𝑘𝑎 ) to that of one free parameter (𝜁). Next, we explore the parameter variations with respect to the
excitation frequency and feedback order to further visualize the FODR technique.

3.3. Parameter analysis

Let us start with the gain parameter. Fig. 5 illustrates the variations of the gain parameter with respect to the excitation frequency
(on a logarithmic scale) and the feedback order (on a linear scale) for 𝜁 = 0.2 and 𝜁 = 0.8. We see that the FODR technique greatly
broadens the optional range of the parameters since it expands the control variables from three integer-order curves to one surface.
In particular, we denote the axis passing through the point 𝑂0 (𝜔̄ = 1, 𝛼 = 1) and being perpendicular to the 𝜔̄ × 𝛼 plane as 𝑍0 .
Intriguingly, the gain value of the point 𝑂1 (𝜔̄ = 0.5, 𝛼 = 0, 𝑔̄ = 0.7762) and that of the point 𝑂2 (𝜔̄ = 2, 𝛼 = 2, 𝑔̄ = 0.7762) in Fig. 5(a) are
equal when 𝜁 = 0.2. Note that 𝑍0 is the center of symmetry. We draw the same conclusion for the points 𝑂3 (𝜔̄ = 5, 𝛼 = 0, 𝑔̄ = 25.2982)
and 𝑂4 (𝜔̄ = 0.2, 𝛼 = 2, 𝑔̄ = 25.2982) in Fig. 5(b). This symmetricity is explained by the following proposition.

Proposition 1 (Symmetricity Property). The point 𝑂0 (𝜔̄ = 1, 𝛼 = 1) is the center of quasi-symmetry for the gain parameter in the 𝜔̄ × 𝛼
space, where 𝜔̄ is on the logarithmic scale and 𝛼 is on the linear scale.

Proof. The dimensionless gain parameter under this quasi-symmetry is written as

√
(( )2 )2 ( )2
1 1 1 1
̄ , 2 − 𝛼) = ( )
𝑔( − 1 + 4𝜁 2
𝜔̄ 2−𝛼 𝜔̄ 𝜔̄
1
𝜔̄
√ (27)
1 𝜔̄ 2 − 12 + 4𝜁 2 𝜔̄ 2
=
𝜔̄ 𝛼−2 𝜔̄ 4
= 𝑔(
̄ 𝜔,
̄ 𝛼).
That finishes the proof. □

We next observe the limits of the gain parameter with this intriguing symmetricity. When 𝛼 = 0, from (24), we have 𝑔̄ → 1 as
𝜔̄ → 0, while 𝑔̄ → ∞ as 𝜔̄ → ∞. When 𝛼 = 2, it follows that 𝑔̄ → ∞ as 𝜔̄ → 0, while 𝑔̄ → 1 as 𝜔̄ → ∞. For the fractional cases
where 0 < 𝛼 < 2, we have 𝑔̄ → ∞ whether 𝜔̄ → 0 or 𝜔̄ → ∞. Nevertheless, it is trivial that the smaller 𝛼 > 0 we select, the closer
𝑔̄ approaches 1 as 𝜔̄ → 0 for 0 < 𝛼 < 1. We now give the opposite judgment according to Proposition 1 for 1 < 𝛼 < 2, that is, the
larger 𝛼 < 2 we choose, the closer 𝑔̄ tends to 1 as 𝜔̄ → ∞. For practical implementations, the gain parameter should be designed as
reasonably small as possible due to the noise amplification. Therefore, the gain parameter with 0 ≤ 𝛼 < 1 is preferred to track the
low-frequency vibrations. On the contrary, the control logic with 1 < 𝛼 ≤ 2 is more suitable for the high-frequency vibrations since

6
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 6. Dimensional gain variations with respect to the excitation frequency and feedback order. Physical parameters are given in Table 1.

the gain parameter increases slowly as the frequency increases. We note that this conclusion only holds for the case of ‘‘fixed order
tuning’’ since the transformation relationship of the gain in (26) is nonlinear when the feedback order is not fixed. Consequently,
the symmetricity stated in Proposition 1 is distorted and no longer holds for the dimensional gain values. However, as shown in
Fig. 6, the increasing or decreasing trend of the gain with respect to the frequency does not change when fixing the order.
From Figs. 5 and 6, we observe that the gain parameter can exhibit a minimum when it varies with respect to the frequency.
The following proposition reveals the existence condition of the extreme value.

Proposition 2 (Extremum Property [29]). The gain parameter reaches its extremum for some specific damping values if the frequency is
given as
[ )
⎧√ 2 1
⎪ 1 − 2𝜁 , 𝜁 ∈ 0, √2 , 𝛼=0
⎪√
⎪ √ 2
𝜔𝑒 = ⎨ (1−𝛼)2 (2𝜁 2 −1) +𝛼(2−𝛼)−(1−𝛼)(2𝜁 2 −1) (28)
, 𝜁 ∈ [0, 1] , 0<𝛼<2
⎪ [
2−𝛼
)
⎪ 1
⎪√ , 𝜁 ∈ 0, √1 , 𝛼=2
⎩ 1−2𝜁 2 2

We draw the variations of the gain parameter with respect to the frequency for various feedback orders in Fig. 7 to verify this
[ ) [ )
proposition. Note that 𝜁 = 0.2 ∈ 0, √1 in Fig. 7(a) while 𝜁 = 0.8 ∉ 0, √1 in Fig. 7(b) so that only the curve corresponding to
2 2
𝛼 = 0 or 𝛼 = 2 in Fig. 7(b) varies monotonically with respect to the frequency and does not have an extremum. This proposition is
also helpful for the evaluation of the resonator stability, see the discussions in Fig. 4 in [29].
Next, we consider the delay variations and present some associated properties. The transformation relationship of the delay
parameter in (26) is linear, which brings ease for the analysis. The delay limits can be evaluated by the following two equations:
( )
𝛼𝜋 + 4(𝑙 − 1)𝜋
𝜏̄0 = lim 𝜏( ̄ 𝛼) = lim −2𝜁 +
̄ 𝜔, = ∞, 𝑙 ∈ N+ , (29)
𝜔→0
̄ 𝜔→0
̄ 2𝜔̄

𝜏̄∞ = lim 𝜏( ̄ 𝛼) = 0
̄ 𝜔, (30)
𝜔→∞
̄

from which the delay parameter decreases from infinity to zero as the frequency grows, and the delay limits are independent of the
feedback order. The monotonous variation of the delay parameter is revealed by the following proposition.

Proposition 3 (Monotonicity Property). The delay parameter decreases monotonically with respect to the frequency for all the feedback
order values.

Proof. The proof of this proposition is analogous to that of Proposition 2. Let us take the partial derivative of the delay parameter
with respect to the frequency:
( ( ) )
𝜕 𝜏(
̄ 𝜔,
̄ 𝛼) 1 𝛼𝜋 2𝜁 𝜔̄ 2𝜁(𝜔̄ 2 + 1)
=− − tan−1 + 2(𝑙 − 1)𝜋 −
𝜕 𝜔̄ 𝜔̄ 2 2 1 − 𝜔̄ 2 ̄ 𝜔̄ 4 + 2(2𝜁 2 − 1)𝜔̄ 2 + 1)
𝜔(
( ) (31)
1 2𝜁(𝜔̄ 2 + 1)
=− 𝜏( ̄ 𝛼) +
̄ 𝜔, .
𝜔̄ 𝜔̄ 4 + 2(2𝜁 2 − 1)𝜔̄ 2 + 1

7
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 7. Gain variations with respect to the excitation frequency for various feedback orders and damping values.

Fig. 8. First- and second-branch delay variations with respect to the excitation frequency and feedback order for 𝜁 = 0.2.

One can easily check that 𝜔̄ 4 + 2(2𝜁 2 − 1)𝜔̄ 2 + 1 > 0 holds for 𝜁 ∈ [0, 1]. Since the delay parameter is defined to be positive, we declare
that 𝜕 𝜏( ̄ 𝛼)∕𝜕 𝜔̄ < 0 invariably holds. □
̄ 𝜔,

Figs. 8 and 9 show the variations of the first- and second-branch delay values for different damping ratios, where the delay
indeed decreases monotonically when the frequency increases. Another critical observation is that the first-branch delay is not
always positive, which degenerates to zero for some configurations. We bring forward the following proposition to explain this.

Proposition 4 (Zero-value Delay Boundary). The first-branch delay parameter is positive if the condition
( )
2 2𝜁 𝜔̄
𝛼 > tan−1 (32)
𝜋 1 − 𝜔̄ 2

is satisfied.

Proof. By setting the delay value in (25) to zero with 𝑙 = 1, we have

( ( ))
1 𝛼𝜋 2𝜁 𝜔̄
𝜏( ̄ 𝛼) =
̄ 𝜔, − tan−1 = 0, (33)
𝜔̄ 2 1 − 𝜔̄ 2
which yields (32) if 𝜏̄ > 0. Additionally, we infer from (32) that 𝛼 → 2 as 𝜔̄ → ∞ and 𝛼 → 0 as 𝜔̄ → 0. That completes the proof. □

Fig. 10 displays the zero-value delay boundary in (32) for various damping ratios. As per the monotonic feature described in
Proposition 3, the delay value of the upper left region divided by this boundary is positive, while that of the lower right region is

8
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 9. First- and second-branch delay variations with respect to the excitation frequency and feedback order for 𝜁 = 0.8.

Fig. 10. Zero-value delay boundaries for various feedback order values.

negative. The corresponding dimensionless control logic at this boundary is given as

√
1 2
̄ 𝛼 𝑥𝑎 (𝑡̄), 𝑔̄ = 𝛼 (𝜔̄ 2 − 1) + 4𝜁 2 𝜔̄ 2 ,
𝑢(𝑡̄) = 𝑔𝐷 (34)
𝜔̄
which implies that the fractional non-delayed control scheme is also workable for an entire vibration suppression. The intrinsic
mechanism hidden in this finding is that the pure fractional term,
𝛼𝜋 𝛼𝜋 𝛼𝜋
𝑠𝛼 |𝑠=𝑗𝜔 = 𝜔𝛼 𝑒𝑗 2 = 𝜔𝛼 cos(
) + 𝑗𝜔𝛼 sin( ), (35)
2 2
can also simultaneously satisfy the requirements for the real and imaginary parts in (8). Additionally, we remark that the imaginary
part of 𝑄(𝑗𝜔) in (8) vanishes when 𝜔 = 𝛺 so that only the pure (non-delayed) velocity feedback is required to damp the vibrations.
This explains the fact that all the curves in Fig. 10 intersect at the point 𝑂0 (𝜔̄ = 1, 𝛼 = 1). Furthermore, the pure position
or acceleration feedback fails to achieve an ideal suppression performance since it cannot eliminate the damping term in (8).
Consequently, we have 𝜔̄ → ∞ when 𝛼 → 2 and 𝜔̄ → 0 when 𝛼 → 0 in Fig. 10.
We underline that the FODR technique is more robust and flexible compared to the fractional delay-free feedback in (34). On
the one hand, the pure fractional control logic cannot tolerate the presence of the inherent delay in the feedback, which is stated
in Section 2. On the other hand, the FODR technique regulates the fractional term and the delay-induced transcendental term
simultaneously. Adjusting the order balances the delay effect by
𝛼𝜋 𝛼𝜋 𝛼𝜋
−𝜏𝜔)
𝑠𝛼 𝑒−𝜏𝑠 |𝑠=𝑗𝜔 = 𝜔𝛼 𝑒𝑗( 2 = 𝜔𝛼 cos(
− 𝜏𝜔) + 𝑗𝜔𝛼 sin( − 𝜏𝜔). (36)
2 2
According to (36), we can view the pure fractional or traditional DR control logic as a reduced scenario of the FODR concept.
Comparatively, our proposed scheme tunes the control variables for the desired amplitude and phase angle values in (36), which
leads to considerable enhancement in the control flexibility.

Remark 2 (On Control Objectives in the Parameter Design Procedure). The design of the FODR control parameters in (16) and (17)
necessitates meeting both the primary and advanced objectives. The primary objective revolves around the viability of vibration
absorption, entailing the fulfillment of (8) and the stability of the poles in (5). Subsequently, the gain and delay values can be

9
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 11. Location relationship for the complex values 𝑠+0 , 𝑠−0 , and 𝑠𝑖𝑛𝑣+
0
.

ascertained for the varying fractional orders. As for the advanced objectives, such as the enhancement of workable frequency band,
transient response speed, and stability robustness, we can use the order adjustment strategy to optimize the control parameters.
Interested readers can refer to the ideas proposed in our prior work [29] (in Section 4).

Remark 3. The design of control parameters for an active absorber is purpose-driven, and the proposed FODR design in (16) and (17)
is tailored for harmonic excitations. However, for impact or random excitation scenarios, an additional evaluation of the vibration
suppression performance across the varying control parameters is warranted [34,35]. In light of the active vibration isolation
systems [36,37], the rightmost poles of the coupled system significantly affect the suppression performance in cases involving impact
and random excitations. Consequently, we can establish an optimal design by tuning the order to ensure that the coupled system
carries the supremum of the real parts of the rightmost poles under the requirement of (8). This means that the suppression speed
of the absorber is concurrently guaranteed for harmonic, impact, and random excitations. Due to space limitations, we do not delve
deeper into this matter in cr present article, as it will be a subject for our future work.

4. Sign inverting control and parameter consistency examination

We elaborate on the control parameter variations for the FODR technique in the last section; however, two additional aspects
need to be further clarified. First, we should explore the situation where the polarity of the control output is reversed, i.e., using
the negative gain parameter. Second, we should check the parameter consistency since the fractional feedback differs from the
conventional integer-order feedback in the parameterized form, cf. [9].
Given that the polarity reversion of the feedback only affects the delay-parameter equation in (21), we next provide some basic
deductions for the simplification of the phase-angle function. Readers should pay attention to two types of the periodicity in (25):
the periods of the tangent function and the phase-angle function are 𝜋 and 2𝜋, respectively. Additionally, the four-quadrant inverse
tangent function, i.e., atan2( ⋅ ), with the range [−𝜋, 𝜋] is employed to calculate the delay parameter. Let

𝑠+
0
= (1 − 𝜔̄ 2 ) + 𝑗2𝜁 𝜔,
̄ 𝑠−
0
= (𝜔̄ 2 − 1) + 𝑗2𝜁 𝜔,
̄ 𝑠𝑖𝑛𝑣+
0
= 2𝜁 𝜔̄ + 𝑗(1 − 𝜔̄ 2 ), (37)

be the three special complex values in the Laplace plane, whose principal argument angles are obtained as
( ) ( ) ( )
2𝜁 𝜔̄ 2𝜁 𝜔̄ 1 − 𝜔̄ 2
𝜑+
0
= tan −1
, 𝜑 −
0
= tan −1
, 𝜑𝑖𝑛𝑣+
0
= tan −1
, (38)
1 − 𝜔̄ 2 𝜔̄ 2 − 1 2𝜁 𝜔̄
respectively. Subsequently, we inspect the quadrant variations of these complex values as the frequency changes. We observe from
(38) that 𝑠+
0
is in the first quadrant with 𝜑+ 0
∈ [0, 𝜋2 ], 𝑠−
0
is in the second quadrant with 𝜑− 0
∈ [ 𝜋2 , 𝜋], and 𝑠𝑖𝑛𝑣+
0
is also in the first
𝜋
quadrant with 𝜑𝑖𝑛𝑣+0
∈ [0, 2𝜋
] when 0 < 𝜔
̄ < 1. Accordingly, we get 𝑠+
0
is in the second quadrant with 𝜑+0
∈ [ 𝜋2 , 𝜋], 𝑠−
0
is in the
𝜋
first quadrant with 𝜑+ 0
∈ [0, 2
], and 𝑠 𝑖𝑛𝑣+
0
is in the fourth quadrant with 𝜑𝑖𝑛𝑣+
0
∈ [− 2
, 0] when 𝜔̄ > 1. Fig. 11 describes the location
relationship of 𝜑+0
, 𝜑−
0
, and 𝜑𝑖𝑛𝑣+
0
for both situations. By the trigonometric geometry, it is trivial that

𝜋 − 𝜑+
0
= 𝜑−
0
(39)

and
𝜋
− 𝜑+
0
= 𝜑𝑖𝑛𝑣+
0
. (40)
2
Then, we can give the expression for the delay parameter under the reversed control scheme.

10
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Theorem 2. The dimensionless parameters of the FODR technique under the sign inverting control is given as
√
1 2
𝑔( ̄ 𝛼) = − 𝛼 (𝜔̄ 2 − 1) + 4𝜁 2 𝜔̄ 2 ,
̄ 𝜔, (41)
𝜔̄
( ( ) )
1 𝛼𝜋 2𝜁 𝜔̄
𝜏( ̄ 𝛼) =
̄ 𝜔, + tan−1 + 2(𝑙− − 1)𝜋 , 𝑙− ∈ N+ . (42)
𝜔̄ 2 2
𝜔̄ − 1

Proof. If the gain is negative, we directly obtain (41) from the magnitude equality in (23). The dimensionless phase equality in
(23) for the delay parameter is written as:
𝛼𝜋
∠(𝑔̄ 𝜔̄ 𝛼 ) + ∠𝑒( 2
−𝜏̄ 𝜔)
̄
= ∠(1 − 𝜔̄ 2 + 𝑗2𝜁 𝜔).
̄ (43)

Since ∠(𝑔̄ 𝜔̄ 𝛼 )
= −𝜋, we have
𝛼𝜋
− 𝜏̄ 𝜔̄ − 𝜋 + 2(𝑙 − 1)𝜋 = 𝜑+
0
(44)
2
where 𝑙 ∈ N+ . Considering 𝛼𝜋 2
− 𝜋 − 𝜑+ 0
∈ [−2𝜋 + 𝛼𝜋
2
, −𝜋 + 𝛼𝜋
2
] and −𝜋 + 𝛼𝜋
2
< 0 for all 0 < 𝛼 < 2, we should redefine the branch
number to guarantee that the first-branch delay values are not invariably negative. By defining 𝑙− = 𝑙 + 1, we secure the following
derivation based on (44):
( )
1 𝛼𝜋
𝜏̄ = − 𝜋 − 𝜑+
0
+ 2𝜋 + 2(𝑙− − 1)𝜋
𝜔̄ 2
( ) (45)
1 𝛼𝜋
= + 𝜑−
0
+ 2(𝑙 −
− 1)𝜋 ,
𝜔̄ 2
which leads to (42). □

Notice that the dimensional parameters for (41) and (42) are acquired using (26) straightforwardly. By revising (25), (39), and
(42), we know that the delay is shifted by + 2𝜋𝜔
when increasing branch number, while it is shifted by + 𝜔𝜋 when employing the
negative gain. Tuning the delay by both the branch number and sign inversion covers all the half-cycle delay values and widens
the selectable range for fixed frequency tuning. Readers are referred to [38,39] for other benefits brought by the inverse delayed
feedback. Next, we discuss whether the proposed fractional control logic is consistent with the traditional one when 𝛼 = 0, 1, and 2.

Proposition 5 (Parameter Consistency). The control variables in Theorems 1 and 2 are consistent with those of the integer-order ones
summarized in [9].

Proof. The consistency of the gain parameter is easy to verify, and the crux of the analysis lies in the delay parameter. Taking
𝛼 = 0 in (42) and 𝛼 = 2 in (25) yields
( ( ) )
1 2𝜁 𝜔̄
𝜏( ̄ 𝛼 = 0) =
̄ 𝜔, tan−1 + 2(𝑙− − 1)𝜋 , (46)
𝜔̄ 𝜔̄ 2 − 1
( ( ) )
1 2𝜁 𝜔̄
𝜏( ̄ 𝛼 = 2) =
̄ 𝜔, 𝜋 + tan−1 + 2(𝑙 − 1)𝜋 , (47)
𝜔̄ 𝜔̄ 2 − 1
respectively. Indeed, increasing the feedback order by 2 also produces a delay shift of + 𝜔𝜋 . Then, we claim that (47) can be simplified
to (46) by the relationship between 𝜑+ 0
and 𝜑−0
in (39) if 𝑙 = 𝑙− . Consequently, the delay value in the position feedback with a negative
gain is identical with that in the acceleration feedback with a positive gain. Furthermore, as per (40), the following equation holds
when 𝛼 = 1:
( ( ) ) ( ( 2 ) )
1 𝜋 2𝜁 𝜔̄ 1 𝜔̄ − 1
𝜏( ̄ 𝛼 = 1) =
̄ 𝜔, + tan−1 + 2(𝑙 − 1)𝜋 = tan−1 + 2(𝑙 − 1)𝜋 . (48)
𝜔̄ 2 𝜔̄ 2 − 1 𝜔̄ 2𝜁 𝜔̄
From (46), (47), and (48), we conclude that the fractional control logic is reduced to traditional integer-order logic when 𝛼 = 0, 1,
and 2. □

By comparing Theorems 1 and 2, we note that the first-branch delay with the position feedback and positive gain is negative for
the whole frequency range. This is only a question of how to define the first branch and does not cause a substantial effect. Instead,
defining the parameters with the positive and negative gain separately guarantees the consistency.

5. Control framework design with multi-source signals

5.1. Integer-order approximations by Oustaloup filter

The objective in this section is to address the control design problem for the implementation of the abstract fractional operator.
Due to the long-term memory effect, applying the fractional operator by its time-domain definition places considerable demands
on the computational capability of the physical controller. To overcome this obstacle, researchers have developed various finite-
dimensional frequency-domain approximation models to imitate the fractional behavior [30]. This article introduces one of the
widely accepted continuous approximation methods: the Oustaloup filter [40]. Let us denote 𝑁 as the approximation order and 𝜔ℎ

11
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 12. Distribution of the poles and zeros of (49) for different approximation orders with 𝜆 = 0.5, 𝜔𝑏 = 10−3 rad/s, and 𝜔ℎ = 103 rad/s.

Fig. 13. Distribution of the poles and zeros of (49) for different approximation frequency bands with 𝜆 = 0.5 and 𝑁 = 10.

and 𝜔𝑏 as the upper and lower bounds of the approximation frequency band, respectively. Then, the fractional differentiation 𝑠𝜆
with 𝜆 ∈ (0, 1) is approximated by the Oustaloup filter as
∏𝑁
𝑠 + 𝜔′ 𝑘
𝐺𝑁 (𝑠) = 𝐾 ≈ 𝑠𝜆 (49)
𝑘=1
𝑠 + 𝜔𝑘

where
√
(2𝑘−1−𝜆)∕𝑁 (2𝑘−1+𝜆)∕𝑁 𝜔ℎ
𝐾 = 𝜔𝜆ℎ , 𝜔′ 𝑘 = 𝜔𝑏 𝜔𝑢 , 𝜔𝑘 = 𝜔𝑏 𝜔𝑢 , 𝜔𝑢 = , 𝑘 = 1, 2, … , 𝑁. (50)
𝜔𝑏
We observe from (49) that the approximation is realized by 𝑁 first-order linear transfer functions connected in series. This model is
also applicable to the fractional integration 𝑠−𝜆 when 𝜆 is changed to −𝜆 in (49) and (50). Figs. 12 and 13 describe the distribution of
the poles and zeros of the transfer function (49) under various approximation configurations where the real axis is on the logarithmic
scale. We see that the poles and zeros are located on the stable (left) side of the imaginary axis and bounded by the approximation
frequency band, i.e., the left margin 𝑠 = −𝜔ℎ and the right margin 𝑠 = −𝜔𝑏 . Additionally, we should carefully appraise the effect of
the approximation order on the control performance. A higher approximation order typically contributes to a better approximation

12
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 14. Control framework design with multi-source signals by the Oustaloup filter for the FODR technique.

accuracy; however, the accuracy no longer exhibits a significant improvement when 𝑁 is sufficiently large, see Fig. 9 in [29].
Moreover, the increase in the number of poles and zeros can cause a negative effect on the dynamic analysis and real-time control.
Hence, engineers should balance these two aspects and choose a compromised scheme in control.
Creating the control signal by the approximation model with a given measurement is another concern for implementation. The
FODR technique is a generalized concept and no restriction is imposed on the type of sensors. Consequently, the approximated
transfer function (49) is determined by the feedback signal source. When the displacement, velocity, or acceleration sensor is utilized,
the frequency-domain expression for the control signal should be designed as

𝑈 (𝑠) = 𝑔𝑠𝛼 𝑒−𝜏𝑠 𝑋𝑎 (𝑠), (51)

𝛼−1 −𝜏𝑠
𝑈 (𝑠) = 𝑔𝑠 𝑠𝑒 𝑋𝑎 (𝑠), (52)
𝛼−2 2 −𝜏𝑠
𝑈 (𝑠) = 𝑔𝑠 𝑠 𝑒 𝑋𝑎 (𝑠), (53)

respectively. At this point, the coupled dynamics under (51), (52), and (53) share the equivalent spectral characteristics, which
is an obvious but essential conclusion. The identical control results can be achieved by applying the fractional operators 𝑠𝛼 , 𝑠𝛼−1 ,
and 𝑠𝛼−2 to the displacement, velocity, and acceleration signals, respectively. Considering that the Oustaloup filter in (49) is only
applicable to the scenarios with 0 < 𝜆 < 1, one needs to connect an additional first-order differentiator or integrator in series in
the feedback for 1 < 𝜆 < 2. Fig. 14 shows the control framework design with multi-source signals by the Oustaloup filter where
the approximation strategy is developed according to the type of sensors. For the displacement signal, we need a differentiator (49)
with 𝜆 = 𝛼 when 0 < 𝛼 < 1 and with 𝜆 = 𝛼 − 1 when 1 < 𝛼 < 2. For the velocity signal, the filter (49) with 𝜆 = 𝛼 − 1 becomes an
integrator when 0 < 𝛼 < 1, while it turns into a differentiator when 1 < 𝛼 < 2. For the acceleration signal, we secure a differential
filter where 𝜆 = 𝛼 − 1 when 0 < 𝛼 < 1 and 𝜆 = 𝛼 − 2 when 1 < 𝛼 < 2. Due to the easy installation of the accelerometer and the
operational safety of the integrator, we deploy the last scheme with

𝑢(𝑡) = 𝑔𝐷𝛼−2 𝑥̈ 𝑎 (𝑡 − 𝜏) (54)

in the ensuring analysis. Readers can refer to Fig. 10 in [29] for the detailed feedback construction.

5.2. PSO algorithm-based optimization for response match

When utilizing the Oustaloup filter, both the approximation order and frequency band affect the approximation accuracy. The
frequency-response match between the theoretical and approximated models is an essential issue. We specify the transmissibility
of the primary system (11) as the target for the evaluation. Fig. 15 displays the comparison of the theoretical and approximated
transmissibility curves for the first-branch parameters with 𝜔 = 0.9 Ω and 𝛼 = 1.8. For a clearer illustration of the influence of the
approximation order on the transmissibility, the time-domain responses at the absorption frequency corresponding to Fig. 15 are
included in Fig. 16 where we take the excitation force as

𝑓 (𝑡) = 10 sin(35.65𝑡). (55)

We discover that a higher approximation order generally results in a more desirable response match. However, as stated in the last
subsection, this comes at the expense of computational efficiency. Table 2 further reveals the transmissibility value at 𝜔 = 0.9 Ω
under various approximation configurations. The transmissibility match at 𝑁 = 5 and 𝜔 ∈ [0.5, 3000] rad/s is unexpectedly preferable

13
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 15. Normalized transmissibility curves for the theoretical and approximated models with 𝜔 = 0.9 Ω and 𝛼 = 1.8. Approximation frequency band is selected
as 𝜔 ∈ [10−3 , 103 ] rad/s.

Fig. 16. Time-domain responses for the approximated models with various approximation orders in Fig. 15.

Table 2
Transmissibility values at the excitation frequency for various approximation configurations.
𝜔𝑏 [rad/s] 𝜔ℎ [rad/s] 𝑁 𝑇 𝑟(𝜔 = 0.9 Ω)
0.1 800 5 0.0146
0.1 2000 5 0.0182
0.5 3000 5 0.0022
0.1 1000 30 0.0121

to that at 𝑁 = 30 and 𝜔 ∈ [0.1, 1000] rad/s. This means merely enlarging the approximation order is inadvisable since the frequency
band affects the transmissibility as well. Owing to the complexity requirement for real-time control, we can regulate the frequency
response match by simultaneously adjusting 𝜔𝑏 and 𝜔ℎ at a predetermined, reasonable approximation order.
By defining the objective function for minimization as the transmissibility value at the excitation frequency, we convert such
a response match problem to an optimization task. For a given excitation frequency 𝜔̃ and approximation order 𝑁, ̃ the objective
function is set to be
min 𝑇𝑟 (𝜔𝑏 , 𝜔ℎ )||𝑠𝜆 =𝐺 (𝑠),𝜔=𝜔,𝑁=
̃ 𝑁̃
𝑁
(56)
s.t: 0 < 𝜔𝑏 < 𝜔ℎ , 𝜔𝑏 ∈ [𝜔̄ 𝑏 , 𝜔𝑏 ], 𝜔ℎ ∈ [𝜔̄ ℎ , 𝜔ℎ ]
where 𝜔𝑏 ∈ [𝜔̄ 𝑏 , 𝜔𝑏 ] and 𝜔ℎ ∈ [𝜔̄ ℎ , 𝜔ℎ ] are the design variables with lower and upper bounds. Since this is a typical optimization
problem, we deploy the well-known particle swarm optimization (PSO) algorithm herein to determine the optimal approximation
frequency band. For instance, the 5-order Oustaloup filter becomes
( )( )( )( )( )
𝑠 + 𝜔0.88
𝑏
𝜔0.12
ℎ
𝑠 + 𝜔0.68
𝑏
𝜔0.32
ℎ
𝑠 + 𝜔0.48
𝑏
𝜔0.52
ℎ
𝑠 + 𝜔0.28
𝑏
𝜔0.72
ℎ
𝑠 + 𝜔0.08
𝑏
𝜔0.92
ℎ
𝐺𝑁 (𝑠) = ( ) ( ) ( ) ( ) ( ) (57)
𝜔0.2
ℎ
𝑠 + 𝜔0.92
𝑏
𝜔0.08
ℎ
𝑠 + 𝜔0.72
𝑏
𝜔0.28
ℎ
𝑠 + 𝜔0.48
𝑏
𝜔0.48
ℎ
𝑠 + 𝜔0.32
𝑏
𝜔0.68
ℎ
𝑠 + 𝜔0.12
𝑏
𝜔0.88
ℎ
when we take 𝛼 = 1.8 (𝜆 = −0.2) in (54). The first-branch delay and gain are given as 𝜏 = 0.0625 and 𝑔 = 0.1294, respectively. Then,
the transmissibility value at 𝑠 = 𝑗0.9 Ω is a real function with respect to only 𝜔𝑏 and 𝜔ℎ . We next set 𝜔𝑏 ∈ [0.005, 10] rad/s and

14
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 17. Comparison of the theoretical transmissibility curve and the optimized one.

Fig. 18. Time-domain responses for the coupled dynamics when the feedback order is tuned at 1.8 and the approximation model is given in (58).

𝜔ℎ ∈ [10, 104 ] rad/s. By the default optimization configuration, the minimum transmissibility value in (56) is achieved as 1.09 × 10−8
at 𝜔𝑏 = 0.4366 rad/s and 𝜔ℎ = 2911.6 rad/s. At this point, (57) is written as
0.2028𝑠5 + 352.5𝑠4 + 8.98 × 104 𝑠3 + 3.818 × 106 𝑠2 + 2.71 × 107 𝑠 + 2.82 × 107
𝐺𝑁 (𝑠) = . (58)
𝑠5 + 1222𝑠4 + 2.189 × 105 𝑠3 + 6.544 × 106 𝑠2 + 3.266 × 107 𝑠 + 2.389 × 107
To demonstrate the overall fitness of the results, we compare the theoretical and optimized transmissibility curves in Fig. 17, where
𝑇 𝑟𝑒 denotes their difference. The plot shows that the optimized approximation model owns a considerably high approximation
accuracy.
The time-domain responses corresponding to approximated model in Fig. 17 are displayed in Fig. 18 where the active control
process starts at 𝑡 = 5 s. When 0 < 𝑡 < 5 s, residual vibrations certainly exist in the primary system owing to the non-zero
transmissibility of the passive scheme. However, a dramatic decay of the vibrations will be observed once the active control
is exerted. As shown in Fig. 18(c), taking the 0.2-order integration by the Oustaloup filter leads to an abstract signal with a
scaled amplitude and phase shift, which is turned into a physically meaningful force signal by further multiplying the (abstract)
dimensionless gain. Readers are directed to Figs. 14 and 15 in our earlier study [29] for this understanding. Also, we state in Remark
7 in [29] that the amplitudes of the control force 𝐴𝑢 and the resonator displacement 𝐴𝑎 are independent of the control logic when
the oscillations are suppressed. For this setting, we have
√
(𝑘𝑎 − 𝑚𝑎 𝜔)2 + (𝑐𝑎 𝜔)2
𝐴𝑢 = 𝐴𝑓 = 2.845 N, (59)
𝑚𝑎 𝜔2
1
𝐴𝑎 = 𝐴𝑓 = 0.0354 m. (60)
𝑚𝑎 𝜔 2
For comparison, we exhibit the time-domain behavior of the primary system for the 5-order optimized configuration and the
20-order non-optimized one when 8 < 𝑡 < 16 s in Fig. 19. This figure indicates that the effect of increasing the approximation order
on the precision improvement is limited. In contrast, optimizing the approximation frequency band is more effective to achieve an
ideal approximated behavior. As a result, the validity of the frequency-domain optimization model described in Fig. 17 is verified
by time-domain responses.

15
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 19. Comparison of the time response of the primary system for the optimized and non-optimized configurations.

5.3. Discretized model analysis

Next, we use the well-known Tustin operator to derive the discretized model of the approximated transfer function [41], which
is crucial for the implementation of digital controller. Under the Tustin operator, the discrete transfer function for (49) is generated
by:

𝐺𝑁 (𝑧) = 𝐺𝑁 (𝑠)||𝑠= 2 𝑧−1 (61)

𝑇𝑠 𝑧+1

where 𝑧 ∈ C is the complex variable with Z-transform and 𝑇𝑠 is the sampling period. For the 5-order approximation, (61) can be
written as
𝑏0 + 𝑏1 𝑧−1 + 𝑏2 𝑧−2 + 𝑏3 𝑧−3 + 𝑏4 𝑧−4 + 𝑏5 𝑧−5
𝐺𝑁 (𝑧) = (62)
1 + 𝑎1 𝑧−1 + 𝑎2 𝑧−2 + 𝑎3 𝑧−3 + 𝑎4 𝑧−4 + 𝑎5 𝑧−5
where 𝑎𝑖 and 𝑏𝑘 , 𝑖 = 1, … , 5, 𝑘 = 0, … , 5, are the corresponding real coefficients. The time-domain expression of (62) is
𝑦𝐺 (𝑛) =𝑏0 𝑢𝐺 (𝑛) + 𝑏1 𝑢𝐺 (𝑛 − 1) + 𝑏2 𝑢𝐺 (𝑛 − 2) + 𝑏3 𝑢𝐺 (𝑛 − 3) + 𝑏4 𝑢𝐺 (𝑛 − 4) + 𝑏5 𝑢𝐺 (𝑛 − 5)
(63)
+ 𝑎1 𝑦𝐺 (𝑛 − 1) + 𝑎2 𝑦𝐺 (𝑛 − 2) + 𝑎3 𝑦𝐺 (𝑛 − 3) + 𝑎4 𝑦𝐺 (𝑛 − 4) + 𝑎5 𝑦𝐺 (𝑛 − 5)
where 𝑢𝐺 and 𝑦𝐺 are the input and output of the transfer function, respectively, and 𝑛 is the discrete time signal. Note that this
discretized expression only applies to the fractional cases. For the realization of (63), the physical controller should store and refresh
the input and output signals from the last five control cycles. The current output 𝑦𝐺 (𝑛) is formed by the linear multiplications and
additions of these stored signals. Besides, the amount of the stored data in the buffer is unchanged if the approximation order is
predetermined. Accordingly, a lower approximation order can reduce the memory burden on the physical controller and facilitate
real-time calculations. For instance, we assume the frequency of the control loop as 1 kHz, i.e., 𝑇𝑠 = 0.001 s, such that the coefficients
in (63) for (58) are given as
𝑏0 = 0.2412, 𝑏1 = −0.9391, 𝑏2 = 1.401, 𝑏3 = −0.9793, 𝑏4 = 0.305, 𝑏5 = −0.02913,
(64)
𝑎1 = −4.132, 𝑎2 = 6.667, 𝑎3 = −5.202, 𝑎4 = 1.934, 𝑎5 = −0.2657.
Actually, we can design these coefficients by the PSO algorithm in advance for operational convenience. For the first-branch
case, Fig. 20 illustrates the variations of the approximation frequency bounds, transmissibility error at the excitation frequency,
discretized control coefficients, and gain-delay parameters under the order adjustment control for 𝜔 = 0.9 Ω. Furthermore, Fig. 21
shows the variations of these first-branch control parameters under the frequency tracking control with the feedback order fixed at
1.8. We see from Fig. 20 that some fluctuations would occur in the discretized coefficients when tuning the order. These fluctuations
are caused by the discontinuity of the 𝜆 value in (49) from 𝛼 ∈ (0, 1) to 𝛼 ∈ (1, 2) as well as the randomness of the PSO algorithm.
Due to the robustness concern, we can sacrifice the control performance in favor of the smoother parameter variations. Nevertheless,
in most cases, these parameters vary continuously, and the approximation error is well controlled.

Remark 4 (On Practical Realization of the Fractional Controller). The implementation of the abstract fractional controller constitutes
a significant challenge, and it is more pragmatic and straightforward to approximate the controller by the frequency-domain
transfer functions, i.e., (49) for continuous systems and (61) for discrete systems. When employing advanced controllers, such as the
CompactRIO controller with the LabVIEW software platform provided by National Instruments, the integrated continuous/discrete
transfer function modules are directly available. In this context, it suffices to furnish the approximation order and polynomial
coefficients of the transfer function, derived from the optimal settings in Section 5.2. For other control systems, such as standard
microcontrollers or digital signal processing (DSP) systems, we can configure the input and output buffers using registers, thus
effectuating the approximation of the fractional controller based on the time-domain expression of discrete transfer function,
e.g., (63).

16
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 20. Variations of the discretized first-branch control parameters under the ‘‘order adjustment control’’ for 𝜔 = 0.9 Ω.

Fig. 21. Variations of the discretized first-branch control parameters under the ‘‘frequency tracking control’’ for 𝛼 = 1.8.

6. Stability preservation analysis

Stability is one of the fundamental concerns of control systems. Exerting an active control force without stability examination
is incredibly risky. In our earlier research [29], an analytical procedure is presented for the complete stability of the resonator
subsystem and the coupled system under the fractional delayed feedback. The analysis follows the work for the integer-order
DR models in [8,9], and [33], which can exactly determine the workable frequency bounds and evaluate the stability robustness
against parameter uncertainties. However, we note that the stability states of the theoretical and approximated dynamics are not
perfectly matched. Employing a theoretically stable configuration can result in an unstable resonator or coupled structure in practical

17
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 22. Stability map of the coupled system. Black curves represent the theoretical stability hypersurfaces and green curves represent the control parameters
in (16) and (17). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

realization, which is even worse than the frequency-response mismatch as discussed in the last section. In this section, we focus on
the effect of the approximated design on the system stability, i.e., the stability preservation analysis, to address this concern.
We denote the characteristic equation of the coupled system, i.e., the denominator of (5), as

𝐶(𝑠) = (𝑚𝑝 𝑠2 + 𝑐𝑝 𝑠 + 𝑘𝑝 )(𝑚𝑎 𝑠2 + 𝑐𝑎 𝑠 + 𝑘𝑎 − 𝑄(𝑠)) + 𝑚𝑎 (𝑐𝑎 𝑠 + 𝑘𝑎 )𝑠2 = 0 (65)

where 𝑄(𝑠) is determined by the approximation model. For the theoretical fractional delayed feedback, we have

𝑄(𝑠) = 𝑔𝑠𝛼 𝑒−𝜏𝑠 . (66)

If the feedback is generated by the Oustaloup filter under (54), we have

{
𝑠𝐺𝑁 (𝑠), 0 < 𝛼 < 1
𝑄(𝑠) = 𝑔𝑒−𝜏𝑠 𝐺𝑁𝐶 (𝑠), 𝐺𝑁𝐶 (𝑠) = (67)
𝑠2 𝐺𝑁 (𝑠), 1 < 𝛼 < 2
where 𝐺𝑁 (𝑠) is given in (49). Dividing both sides of (65) by 𝑚𝑝 𝑠2 + 𝑐𝑝 𝑠 + 𝑘𝑝 yields

𝐶(𝑠) (𝑚𝑝 𝑠2 + (𝑐𝑎 + 𝑐𝑝 )𝑠 + 𝑘𝑎 + 𝑘𝑝 )

𝐶𝑑𝑖𝑣 (𝑠) = = 𝑚𝑎 𝑠2 + 𝑐𝑎 𝑠 + 𝑘𝑎 − 𝑄(𝑠). (68)
𝑚𝑝 𝑠2 + 𝑐𝑝 𝑠 + 𝑘𝑝 𝑚 𝑝 𝑠2 + 𝑐 𝑝 𝑠 + 𝑘 𝑝
From the magnitude and phase equality of (68), we obtain the potential stability switching hypersurfaces for the coupled system as
| 𝜙(𝑠)𝑚 𝑠2 + 𝑐 + 𝑘 | |
| 𝑎 𝑎 𝑎| |
𝑔𝑐𝑠 (𝜔, 𝛼) = | | |, (69)
| 𝐺𝑁𝐶 (𝑠) | 𝜋 |
| |𝑠=𝜔𝑒𝑗 2 |
( )
1
2
𝜙(𝑠)𝑚𝑎 𝑠 + 𝑐𝑎 + 𝑘𝑎 | |
𝜏𝑐𝑠 (𝜔, 𝛼) = −∠ | + 2 (𝑙 − 1) 𝜋 (70)
𝜔 𝐺𝑁𝐶 (𝑠) | 𝜋
|𝑠=𝜔𝑒𝑗 2
where 𝑙 ∈ N+ and
(𝑚𝑝 𝑠2 + (𝑐𝑎 + 𝑐𝑝 )𝑠 + 𝑘𝑎 + 𝑘𝑝 )
𝜙(𝑠) = . (71)
𝑚𝑝 𝑠2 + 𝑐𝑝 𝑠 + 𝑘𝑝
Note that the theoretical stability hypersurfaces are secured if one forces 𝐺𝑁𝐶 (𝑠) = 𝑠𝛼 in (69) and (70). Fig. 22 shows the
distribution of these hypersurfaces for 𝛼 = 1.8 where the parameter curves in (16) and (17) are also plotted. The origin in Fig. 22,
i.e., (𝑔, 𝜏) = (0, 0), invariably stands for a stable system with the vanishment of the control input. Therefore, the shaded region in
this figure dictates an asymptotically stable overall structure according to the D-subdivision theorem [42]. The control parameters
(points on the green lines) are valid when they are tuned in the stable region. When the parameters cross the stability hypercurves
(black curves) to the unshaded region, the poles move across the imaginary axis to the right half of the complex plane so that the
system becomes unstable. Then, some small but inevitable uncertainties in the feedback loop can cause the instability of the coupled
dynamics since the parameter curves are so close to the stability hypersurfaces. For more discussions of this concern, readers are
directed to Sections 3.2 and 4.2 in [29].
Nevertheless, the stability robustness (or the distribution of stability hypersurfaces) of the coupled dynamics is affected by the
approximation configurations when the Oustaloup filter is employed. For instance, we set the approximation frequency band as
𝜔 ∈ [10−3 , 103 ] rad/s and plot the variations of the stability hypersurfaces with respect to the approximation order in Fig. 23.
Indeed, the stability margin slightly shifts with respect to the approximation order and frequency band. However, it is difficult to
prove whether such an offset necessarily diminishes or enhances the stability robustness. For this, one should carefully recheck the
effect of the approximation model on the stability robustness of the coupled system in some stability-sensitive scenarios.
Next, we utilize the Quasi-Polynomial mapping-based Root-finder (QPmR) [43] routine to demonstrate how the approximation
model affects the zero and pole distributions of the transfer function (5). Figs. 24 and 25 display the spectrum of the coupled
system for the first- and second-branch parameters under various approximation configurations. We observe that the approximated

18
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 23. Variations of the stability hypersurfaces of the coupled system for various approximation orders.

Fig. 24. Spectrum of the coupled system obtained by the QPmR algorithm for the first-branch parameters. Configuration 1: 𝜔 ∈ [10−3 , 103 ] rad/s, 𝑁 = 3.
Configuration 2: 𝜔 ∈ [10−3 , 103 ] rad/s, 𝑁 = 5. Configuration 3: 𝜔 ∈ [5 × 10−3 , 5 × 103 ] rad/s, 𝑁 = 5.

Fig. 25. Spectrum of the coupled system obtained by the QPmR algorithm for the second-branch parameters. Configuration 1: 𝜔 ∈ [10−3 , 103 ] rad/s, 𝑁 = 3.
Configuration 2: 𝜔 ∈ [10−3 , 103 ] rad/s, 𝑁 = 5. Configuration 3: 𝜔 ∈ [5 × 10−3 , 5 × 103 ] rad/s, 𝑁 = 5.

spectrum is slightly shifted compared to the theoretical one. On the one hand, the offset of zeros makes the vibration-absorption
performance unsatisfactory, and even destabilizes the absorber subsystem, which is not beneficial to the system robustness. On the
other hand, the pole shift affects the overall stability and the transient response speed, i.e., the vibration-absorption speed. Therefore,
the approximated transfer function should be designed cautiously when the dominant poles are close to the imaginary axis. For the
safety concern, we can even sacrifice the control performance in exchange for the system stability. In conclusion, we should design
a proper approximation model to solve the stability preservation problem.

7. Frequency sensitivity analysis under delay discretization

In what follows, we consider a practical issue that has not been explored to date. We note that the actual excitation frequency
is typically acquired by the discrete Fourier transform (DFT) analysis. The frequency resolution is used to represent the frequency
spacing between DFT coefficients in the frequency domain, which is given by 𝛥𝜔𝑟 = 𝜔𝑠 ∕𝑁 𝑠 where 𝜔𝑠 is the sampling frequency and
𝑁𝑠 is the number of samples. Due to the delay discretization, the precise delay value is written as

𝜏 = 𝑘𝑇𝑠 + 𝜏𝑡𝑟𝑢𝑛 (72)

where 𝑘 ∈ N+ ,
𝑇𝑠 is the sampling period, and 𝜏𝑡𝑟𝑢𝑛 is the truncated delay. According to (72), only the discretized delay 𝜏 = 𝑘𝑇𝑠 is
valid (or precise) from the practical point of view. Consequently, the small nonlinear frequency intervals 𝛥𝜔𝑠 , which are induced
by the discretization period 𝑇𝑠 , prevent us from tracking every identified frequency. Evidently, a smaller frequency interval leads

19
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 26. Variation of the discretized delay with respect to the excitation frequency for various feedback orders under the 1 kHz control frequency.

Fig. 27. (a) Frequency sensitivity curves for various feedback orders. (b) Delay parameter variation.

to a higher frequency sensitivity. To precisely track the frequency, we should guarantee that the difference between 𝛥𝜔𝑠 and 𝛥𝜔𝑟
is minimized.
We set the control loop frequency as 1 kHz (𝑇𝑠 = 1 ms) to clarify this understanding. Fig. 26 shows the variation of the discretized
delay with respect to the excitation frequency for various feedback orders where the discretized points represent the valid delay
values, i.e., 𝜏 = 𝑘𝑇𝑠 . We discover that the frequency interval is much larger than the frequency resolution in practice. Additionally,
the delay closer to the zero-value boundary (33) has a greater decay rate with respect to the frequency as well as a higher frequency
sensitivity. As per (12), we can regard the delay parameter as a combination of the inherent delay and loop delay. When the inherent
delay is large or the delay parameter experiences a severe perturbation, a fast delay decay rate can bring about the violation of (12),
i.e., 𝜏 < 𝜏𝑠𝑦𝑠 . Therefore, the control variables should be tuned far away from this zero-value boundary in view of the control safety.
However, the inherent delay is small, and the control parameters fluctuate slightly when the control system is robust enough. Thus,
employing the parameters close to zero-value boundary significantly improves the frequency sensitivity.
Fig. 27 illustrates the effect of the order adjustment control on the frequency intervals, which indicates that the increase in
the feedback order enhances the frequency sensitivity. The curves in Fig. 27, apart from the one with acceleration feedback,
are truncated by the cut-off frequency due to the zero-delay boundary. Furthermore, the frequency sensitivity corresponding to
acceleration feedback drops dramatically for the high-frequency vibrations owing to the gradually slowing delay decay rate, which
is unacceptable. Hence, only adjusting the order has a limited effect on improving the frequency sensitivity. Optionally, we can
raise the control frequency to solve this issue, e.g., from 1 kHz to 2 kHz, then, the 𝛥𝜔𝑠 curves in Fig. 27 would decrease overall.
However, it requires a balance between the hardware and software for every control session and comes at a greater cost.
We can think differently to take the feedback order as an additional free parameter for this problem. Since the FODR control
logic has three free control variables, we fix the delay parameter to a certain value (𝜏 = 𝑘𝑇𝑠 ) for delay discretization. Then, as shown
in Fig. 28, the gain and feedback order are tuned simultaneously to suppress the vibrations with the varying frequency. Another
advantage of fixing the delay is that the artificially compensated delay in (12) can be adjusted in real-time to improve the robustness
against delay uncertainties. However, the order tuning is feasible within the range of [0,2], which is the only limitation. According
to Fig. 28, the delay should be kept smaller to guarantee a wider feasible frequency band. Additionally, the delay increases from zero
to infinity as the feedback order decreases from two to zero, which is proved by Proposition 3. Thus, the cut-off frequency in Fig. 28
tends to zero when utilizing the position feedback. We obtain the corresponding first-branch gain value as 𝑔𝑐𝑢𝑡 = 𝑘𝑎 = 350 kg s−2 by
setting 𝜔 → 0 and 𝛼 = 0 in (17).
The ensuring simulation example is provided to emphasize the effectiveness of the idea. We take the frequency and amplitude
of the excitation force as 𝜔 = 45 rad/s (𝜔̄ = 1.1359) and 𝐴𝑓 = 10 N, respectively. According to (32), the first-branch delay parameter
is feasible when 𝛼 > 1.5731. We consider the experimental platform built in [7,8] where the control frequency and inherent delay
are 1 kHz and 4 ms, respectively. For the FODR technique, we can compensate 1 ms delay in the control loop such that 𝜏 = 0.005
s, then the feedback order and gain are successively calculated as 𝛼 = 1.8163 and 𝑔 = 0.1159 kg s𝛼−2 , respectively. The optimized
approximation parameters for 𝑁 = 5 are given as 𝜆 = −0.1837, 𝜔𝑏 = 0.5538 rad/s, 𝜔ℎ = 3656.6 rad/s. Fig. 29 shows the numerical

20
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

Fig. 28. Parameter variations under the fractional order adjustment control when the delay is fixed.

Fig. 29. Comparison of the suppressed behavior of the primary system under delay discretization.

(continuous) simulations for the primary structure under such a configuration for 2 < 𝑡 < 6 s where the active control starts at
𝑡𝑐 = 0 s. By contrast, the truncated delay for the acceleration delayed feedback is given as 𝜏 = 0.011 s with 𝜏𝑡𝑟𝑢𝑛 = 0.0004 s, which
leads to either an unsatisfactory absorption performance (Fig. 29) or a poor frequency sensitivity (Fig. 27).

8. Conclusion

This paper investigates a classical vibration-suppression problem from a new perspective. We bring about the FODR technique
and address the associated design, analysis, and optimization control problems. We wish to elaborate on our unique contributions
as follows.
(1) We analyze the control parameter variations in detail to visualize the FODR technique. The symmetricity and extremum
properties of the gain, and the monotonicity and zero-value properties of the delay are proved. Also, the parameter consistency and
the delay variation under polarity reversal are examined.
(2) We design a generalized control framework with multi-source signals by the frequency-domain approximation model
(Oustaloup filter). To achieve the desired suppression performance, we calculate the optimal approximation frequency band by
the PSO algorithm. Based on this, the discretized expression of the optimized controller is discussed.
(3) We perform a complete analysis for the asymptotic stability of the coupled system and further inspect the stability preservation
issue. The results reveal that the approximation model has an indeterminate effect on the pole and zero distributions.
(4) We propose a control strategy named ‘‘order adjustment’’ that treats the feedback order as a freely tunable parameter to
enhance the frequency sensitivity. This facilitates the vibration suppression with a varying external excitation frequency.
Finally, we stress that the vibration control task is much more complicated in practice. The oscillations can be multi-frequency,
multi-dimensional, or frequency stochastic. Besides, the vibratory system can be multi-degree-of-freedom, non-linear, or in a non-
collocated configuration with the absorber. Therefore, for the future work, we would analyze the effectiveness of the FODR technique
in the context of specific practical applications.

21
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

CRediT authorship contribution statement

Jiazhi Cai: Conceptualization, Methodology, Software, Investigation, Formal analysis, Writing – original draft, Writing – review
& editing. Qingbin Gao: Project administration, Funding acquisition, Writing – review & editing. Yifan Liu: Conceptualization,
Software, Validation. Nejat Olgac: Supervision, Writing – review & editing.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared
to influence the work reported in this paper.

Data availability

Data will be made available on request.

Acknowledgment

This work is supported in part by the National Natural Science Foundation of China (No. 62103119), in part by the Basic Research
Plan of Shenzhen, China (No. JCYJ20200109113429208), in part by the Basic and Applied Basic Research Foundation of Guangdong
Province, China (No. 2022A1515011790), and in part by the Shenzhen Excellent Scientific and Technological Innovation Talent
Training Project, China (No. RCJC20200714114436040). The authors are also grateful to the anonymous reviewers for their valuable
comments.

References

[1] Y. Shen, H. Peng, X. Li, S. Yang, Analytically optimal parameters of dynamic vibration absorber with negative stiffness, Mech. Syst. Signal Process. 85
(2017) 193–203, http://dx.doi.org/10.1016/j.ymssp.2016.08.018.
[2] Y. Shen, Z. Xing, S. Yang, J. Sun, Parameters optimization for a novel dynamic vibration absorber, Mech. Syst. Signal Process. 133 (2019) 106282,
http://dx.doi.org/10.1016/j.ymssp.2019.106282.
[3] M. Kassem, Z. Yang, Y. Gu, W. Wang, E. Safwat, Active dynamic vibration absorber for flutter suppression, J. Sound Vib. 469 (2020) 115110.
[4] T.G. Larsen, Z. Zhang, J. Høgsberg, Vibration damping of an offshore wind turbine by optimally calibrated pendulum absorber with shunted electromagnetic
transducer, J. Sound Vib. 505 (2021) 116144.
[5] Y. Liu, J. Cai, H. Li, Q. Gao, Optimal design and sensitivity analysis of the dynamic vibration absorber with amplifying mechanism, J. Comput. Inf. Sci.
Eng. 23 (5) (2023) 051005, http://dx.doi.org/10.1115/1.4056920.
[6] N. Olgac, B.T. Holm-Hansen, A novel active vibration absorption technique: Delayed resonator, J. Sound Vib. 176 (1) (1994) 93–104, http://dx.doi.org/
10.1006/jsvi.1994.1360.
[7] D. Pilbauer, T. Vyhlídal, N. Olgac, Delayed resonator with distributed delay in acceleration feedback—design and experimental verification, IEEE/ASME
Trans. Mechatronics 21 (4) (2016) 2120–2131, http://dx.doi.org/10.1109/TMECH.2016.2516763.
[8] V. Kučera, D. Pilbauer, T. Vyhlídal, N. Olgac, Extended delayed resonators–Design and experimental verification, Mechatronics 41 (2017) 29–44,
http://dx.doi.org/10.1016/j.mechatronics.2016.10.019.
[9] T. Vyhlídal, D. Pilbauer, B. Alikoç, W. Michiels, Analysis and design aspects of delayed resonator absorber with position, velocity or acceleration feedback,
J. Sound Vib. 459 (2019) 114831, http://dx.doi.org/10.1016/j.jsv.2019.06.038.
[10] D. Pilbauer, T. Vyhlídal, W. Michiels, Optimized design of robust resonator with distributed time-delay, J. Sound Vib. 443 (2019) 576–590, http:
//dx.doi.org/10.1016/j.jsv.2018.12.002.
[11] M. Valášek, N. Olgac, Z. Neusser, Real-time tunable single-degree of freedom, multiple-frequency vibration absorber, Mech. Syst. Signal Process. 133 (2019)
106244, http://dx.doi.org/10.1016/j.ymssp.2019.07.025.
[12] N. Olgac, R. Jenkins, Actively tuned noncollocated vibration absorption: An unexplored venue in vibration science and a benchmark problem, IEEE Trans.
Control Syst. Technol. 29 (1) (2021) 294–304, http://dx.doi.org/10.1109/TCST.2020.2973603.
[13] Z. Šika, T. Vyhlídal, Z. Neusser, Two-dimensional delayed resonator for entire vibration absorption, J. Sound Vib. 500 (2021) 116010, http://dx.doi.org/
10.1016/j.jsv.2021.116010.
[14] T. Vyhlídal, W. Michiels, Z. Neusser, J. Bušek, Z. Šika, Analysis and optimized design of an actively controlled two-dimensional delayed resonator, Mech.
Syst. Signal Process. 178 (2022) 109195, http://dx.doi.org/10.1016/j.ymssp.2022.109195.
[15] Y. Liu, J. Cai, N. Olgac, Q. Gao, A robust delayed resonator construction using amplifying mechanism, J. Vib. Acoust. 145 (1) (2023) 011010,
http://dx.doi.org/10.1115/1.4055559.
[16] Q. Gao, Y. Liu, J. Cai, H. Wu, Z. Long, Complete stability analysis and optimization of the extended delayed resonator with virtual natural frequency
adjustment, J. Dyn. Syst. Meas. Control 145 (1) (2023) 011002, http://dx.doi.org/10.1115/1.4055800.
[17] H. Sun, Y. Zhang, D. Baleanu, W. Chen, Y. Chen, A new collection of real world applications of fractional calculus in science and engineering, Commun.
Nonlinear Sci. Numer. Simul. 64 (2018) 213–231, http://dx.doi.org/10.1016/j.cnsns.2018.04.019.
[18] Q. Gao, J. Cai, Y. Liu, Y. Chen, L. Shi, W. Xu, Power mapping-based stability analysis and order adjustment control for fractional-order multiple delayed
systems, ISA Trans. 138 (2023) 10–19, http://dx.doi.org/10.1016/j.isatra.2023.02.019.
[19] H. Wang, G.I. Mustafa, Y. Tian, Model-free fractional-order sliding mode control for an active vehicle suspension system, Adv. Eng. Softw. 115 (2018)
452–461, http://dx.doi.org/10.1016/j.advengsoft.2017.11.001.
[20] S.D. Nguyen, B.D. Lam, S.-B. Choi, Smart dampers-based vibration control–Part 2: Fractional-order sliding control for vehicle suspension system, Mech.
Syst. Signal Process. 148 (2021) 107145, http://dx.doi.org/10.1016/j.ymssp.2020.107145.
[21] Z. Yu, Y. Zhang, B. Jiang, C.-Y. Su, J. Fu, Y. Jin, T. Chai, Nussbaum-based finite-time fractional-order backstepping fault-tolerant flight control of fixed-wing
UAV against input saturation with hardware-in-the-loop validation, Mech. Syst. Signal Process. 153 (2021) 107406, http://dx.doi.org/10.1016/j.ymssp.
2020.107406.
[22] B. Shang, J. Liu, Y. Zhang, C. Wu, Y. Chen, Fractional-order flight control of quadrotor UAS on vision-based precision hovering with larger sampling
period, Nonlinear Dynam. 97 (2019) 1735–1746, http://dx.doi.org/10.1007/s11071-019-05103-5.

22
J. Cai et al. Journal of Sound and Vibration 571 (2024) 118083

[23] Z. Wang, H. Wu, Stabilization in finite time for fractional-order hyperchaotic electromechanical gyrostat systems, Mech. Syst. Signal Process. 111 (2018)
628–642, http://dx.doi.org/10.1016/j.ymssp.2018.04.009.
[24] P. Veeraian, U. Gandhi, U. Mangalanathan, Design and analysis of fractional order seismic transducer for displacement and acceleration measurements, J.
Sound Vib. 419 (2018) 123–139, http://dx.doi.org/10.1016/j.jsv.2018.01.007.
[25] M.P. Aghababa, A fractional-order controller for vibration suppression of uncertain structures, ISA Trans. 52 (6) (2013) 881–887, http://dx.doi.org/10.
1016/j.isatra.2013.07.010.
[26] A.-A. Zamani, S. Tavakoli, S. Etedali, Fractional order PID control design for semi-active control of smart base-isolated structures: A multi-objective cuckoo
search approach, ISA Trans. 67 (2017) 222–232, http://dx.doi.org/10.1016/j.isatra.2017.01.012.
[27] W. Niu, B. Li, T. Xin, W. Wang, Vibration active control of structure with parameter perturbation using fractional order positive position feedback controller,
J. Sound Vib. 430 (2018) 101–114, http://dx.doi.org/10.1016/j.jsv.2018.05.038.
[28] L. Marinangeli, F. Alijani, S.H. HosseinNia, Fractional-order positive position feedback compensator for active vibration control of a smart composite plate,
J. Sound Vib. 412 (2018) 1–16, http://dx.doi.org/10.1016/j.jsv.2017.09.009.
[29] J. Cai, Y. Liu, Q. Gao, Y. Chen, Spectrum-based stability analysis for fractional-order delayed resonator with order scheduling, J. Sound Vib. 546 (2023)
117440, http://dx.doi.org/10.1016/j.jsv.2022.117440.
[30] F.N. Deniz, B.B. Alagoz, N. Tan, M. Koseoglu, Revisiting four approximation methods for fractional order transfer function implementations: Stability
preservation, time and frequency response matching analyses, Annu. Rev. Control 49 (2020) 239–257, http://dx.doi.org/10.1016/j.arcontrol.2020.03.003.
[31] C.W. de Silva, Modeling of Dynamic Systems with Engineering Applications, CRC Press, Taylor & Francis, Boca Raton, FL, 2018.
[32] C.A. Monje, Y. Chen, B.M. Vinagre, D. Xue, V. Feliu-Batlle, Fractional-Order Systems and Controls: Fundamentals and Applications, Springer Science &
Business Media, London, 2010.
[33] T. Vyhlídal, N. Olgac, V. Kučera, Delayed resonator with acceleration feedback–Complete stability analysis by spectral methods and vibration absorber
design, J. Sound Vib. 333 (25) (2014) 6781–6795, http://dx.doi.org/10.1016/j.jsv.2014.08.002.
[34] M. Tso, J. Yuan, W. Wong, Suppression of random vibration in flexible structures using a hybrid vibration absorber, J. Sound Vib. 331 (5) (2012) 974–986,
http://dx.doi.org/10.1016/j.jsv.2011.10.017.
[35] Y. Chang, J. Zhou, K. Wang, D. Xu, A quasi-zero-stiffness dynamic vibration absorber, J. Sound Vib. 494 (2021) 115859, http://dx.doi.org/10.1016/j.jsv.
2020.115859.
[36] X. Sun, J. Xu, J. Fu, The effect and design of time delay in feedback control for a nonlinear isolation system, Mech. Syst. Signal Process. 87 (2017)
206–217, http://dx.doi.org/10.1016/j.ymssp.2016.10.022.
[37] X. Sun, J. Xu, X. Jing, L. Cheng, Beneficial performance of a quasi-zero-stiffness vibration isolator with time-delayed active control, Int. J. Mech. Sci. 82
(2014) 32–40, http://dx.doi.org/10.1016/j.ijmecsci.2014.03.002.
[38] Q. Gao, A.S. Kammer, U. Zalluhoglu, N. Olgac, Combination of sign inverting and delay scheduling control concepts for multiple-delay dynamics, Systems
Control Lett. 77 (2015) 55–62, http://dx.doi.org/10.1016/j.sysconle.2015.01.001.
[39] Q. Gao, N. Olgac, Optimal sign inverting control for time-delayed systems, a concept study with experiments, Internat. J. Control 88 (1) (2015) 113–122,
http://dx.doi.org/10.1080/00207179.2014.941409.
[40] A. Oustaloup, F. Levron, B. Mathieu, F.M. Nanot, Frequency-band complex noninteger differentiator: Characterization and synthesis, IEEE Trans. Circuits
Syst. I 47 (1) (2000) 25–39, http://dx.doi.org/10.1109/81.817385.
[41] B.M. Vinagre, Y.Q. Chen, I. Petráš, Two direct Tustin discretization methods for fractional-order differentiator/integrator, J. Franklin Inst. B 340 (5) (2003)
349–362, http://dx.doi.org/10.1016/j.jfranklin.2003.08.001.
[42] V.B. Kolmanovskii, V.R. Nosov, Stability of Functional Differential Equations, Academic Press, Cambridge, 1986.
[43] T. Vyhlidal, P. Zítek, Mapping based algorithm for large-scale computation of quasi-polynomial zeros, IEEE Trans. Automat. Control 54 (1) (2009) 171–177,
http://dx.doi.org/10.1109/TAC.2008.2008345.

23