Professional Documents
Culture Documents
Abstract
Piezoelectric bending actuators have been widely used in a variety of micro- and nano-applications, including atomic force
microscopy, micro assembly, cell manipulation, and in general, micro electromechanical systems. However, their control
algorithms at low frequencies suffer from nonlinearities such as hysteresis in high voltages and creep in long-time static
applications. Also, in high-frequency applications, especially near the actuator natural frequencies, the actuator dynamic is
greatly affected by the material nonlinearity. Therefore, the control approaches based on the linear dynamic modeling
cannot be effective at high frequencies. Thus, the position control of the foregoing actuators become challenging, and it has
been of researchers’ interests in the last decade. In this article, the robust position control of a bimorph piezoelectric
bending actuators is investigated. In this regard, based on the nonlinear constitutive equations and the Euler–Bernoulli beam
theory, a nonlinear dynamic model is presented. Then, to track a desired motion trajectory, an observer-based robust
position control algorithm is proposed. The proposed control methodology is able to accommodate parametric un-
certainties and other un-modeled dynamics. Also, it ensures the elimination of the position tracking error in the presence of
the estimated states. Finally, the tracking ability of the controller is demonstrated in an experimental study. The ex-
perimental results show that the identification of the system is properly conducted with the average error of 5.5%. Also, the
efficiency of the robust controller is proved with the error of 3.7% and 4.9% in the position tracking of the actuator inside
and outside of the identified region, respectively.
Keywords
Piezoelectric bending actuator, material nonlinearity, nonlinear dynamic, position control, high gain observer
position control systems as well (Andreaus et al., 2004; controller and observer, the actuator is precisely tracked and
Dell’Isola et al., 2004; Liu et al., 2016; Liu et al., 2017; desired trajectory inside and outside of the identification
Thomas et al., 2011). However, in such applications, near domain.
the fundamental natural frequency, the actuator shows
a special nonlinear behavior, so called material non-
2. Nonlinear dynamic modeling
linearities (Leadenham and Erturk, 2015b). These non-
linearities demonstrate a softening effect in the frequency A piezoelectric bimorph actuator, which is made of an
responses of the actuator (Leadenham and Erturk, 2019; Tan electrically inactive brass shim of density ρb with thickness
et al., 2018b). As a result, the actuator behavior significantly of tb and symmetric anti-parallel polarized piezoceramic
deviates from the one obtained by conventional linear laminates with density of ρp and thickness of tp , is illustrated
models, and hence, the control approaches which are based in Figure 1. The cross section of the actuator is rectangular
on the linear dynamic model fail to operate properly. with the width of Y .
Recently, many researches aimed at deriving a proper
nonlinear dynamic model to accurately describe the be- 2.1. Material nonlinearity
havior of piezoelectric materials at high frequencies.
Stanton et al. comprehensively investigated the nonlinear A bimorph piezoelectric actuator behaves linearly at fre-
energy harvesting of piezoelectric cantilevered beam quencies far from the resonance while a low-amplitude
(Stanton et al., 2010a). By resorting to energy methods, the input voltage is applied. However, when the applied volt-
authors proposed a dynamic model in which the unknown age attains frequency close to the resonance, the actuator
nonlinear parameters were attained by resonating the pie- behavior is no more predictable with a linear model. This
zoelectric beam near its first natural frequency. As an im- deviation is depicted in Figure 2 where the linear and
portant achievement, it is reported that the linear damping nonlinear frequency responses of a commercial bimorph
model fails to describe the behavior of a piezoelectric piezoelectric actuator (T226-H4-203X) are compared with
cantilevered beam at high frequencies. Alternatively, the experimental results.
a quadratic fluid damping model was used to consider the
energy dissipation at high frequencies (Stanton et al.,
2010a). In a different study, Stanton et al. tried to predict
the nonlinear non-conservative behavior of piezoelectric
energy harvesters with a proof mass at tip of the beam
(Stanton et al., 2012). In this work, the authors considered
a damping model which depends on the beam deflection
(Stanton et al., 2012). Recently, the nonlinear behavior of
a piezoelectric bimorph cantilevered beam was studied by Figure 1. Illustration of a symmetric bimorph actuator con-
Leadenham and Erturk (2015a). The authors investigated nected in series to a power supplier.
the behavior of a piezo beam under electrical and me-
chanical excitations for different applications such as en-
ergy harvesting, sensing, and actuating.
This article mainly targets at the high-frequency position
control of a piezoelectric bending actuator. As the corre-
sponding nonlinear constitutive equations depend signifi-
cantly on the dimensions, the electrical input amplitude, the
strain magnitude and, etc., they are obtained empirically.
Next, based on the obtained equations, the Euler–Bernoulli
beam theory is used for the nonlinear dynamic model near
the first natural frequency. Then, a sliding mode position
control is proposed to follow a desired motion trajectory in
the presence of un-modeled dynamics and disturbances.
Because of the necessity of actuator velocity estimation in
a control problem, this state is evaluated at high frequencies
by means of a high-gain reduced-order observer. By re-
sorting to the Lyapunov stability theorem, the overall
closed-loop stability is also theoretically proven in the
presence of estimated states. Finally, the nonlinear mod-
eling accuracy and high-frequency position control per-
formance are evaluated experimentally. With the designed Figure 2. Piezoelectric actuator frequency response for V = 3 V.
Shahabi et al. 3
As it is obvious in Figure 2, a linear model cannot predict actuator is composed of the linear bending energy of
the actual behavior of the actuator near the resonance substrate, and the energy related to the nonlinear stress–
frequency. In such cases, it is predicted that because of high strain behavior of piezoceramic laminates, namely
strains, the material linear models are no longer valid, and
thus, a nonlinear model should be considered. Z l 2 2 Z " 2 4 2 6 #
1 ∂w ∂w ∂w
In case of piezoelectric materials, the conservative po- U ¼ CIe dx α þβ dx
2 0 ∂x 2 ∂x 2 ∂x2
tential density is expressed in terms of an electric enthalpy
Z 2
function. In linear models, because of the linear constitutive 1X 3
2 l ∂w
relations for the electromechanical behavior of piezoelectric e31i Yi tio tiu2
E3 dx
2 i¼1 0 ∂x2
actuators, the electric enthalpy function is quadratic, such as
X 3 Z l
1 1 1S
He linear ¼ cE11 S12 e31 E3 S1 εS33 E32 (1) þ ε33i ½tio tiu wi E3i2
dx
2 2 i¼1
2 0
linear strain-displacement relation is considered, namely the nonconservative work W comprises the works done by
the applied voltage and the dissipative frictional forces,
2
∂w namely
S1 ¼ z (3)
∂x2
Z X3 Z hio Z Y Z l
l
∂w ∂U
where w ¼ wðx; tÞ denotes the transverse deflection of the W ¼ μ wdx σ dxdydz
actuator. The total kinetic energy of the actuator can be also 0 ∂t i¼1 hiu 0 0 ∂z
cast in the following format (10)
Z l 2
1 ∂w where μ denotes the coefficient of friction and σ is the
T ¼ ρAe dx (4) surface charge density. Substituting equation (10) into (9),
2 0 ∂t
applying the calculus of variations and using the integration
where ρAe ¼ Y ðρb tb þ 2ρp tp Þ is the mass per unit length of by parts, the nonlinear equation of motion of the actuator is
the actuator. The total potential energy stored within the obtained as
4 Journal of Vibration and Control 0(0)
CIe w0000 þ ρAe w€ þ μw_ þ 12α 2w00 w0002 þ w0000 w002 q€1 þ 2ξ 1 ω1 q_1 þ ω21 q1 þ k1 q31 þ k2 q51
Z l
d d (14)
þ 30β 6w003 w0002 þ w0000 w004 ¼ Mp V ðtÞ Gðx lÞ ¼ Mp V ðtÞ φj ðxÞ Gðx lÞ
dx 0 dx
(11)
in which
in which Mp ¼ Ye31 ðtp þ tb Þ and GðxÞ refers to the Dirac Z l
delta function. k1 ¼ 12α φ φ002 φ0000 þ 2φ00 φ0002 dx (15)
0
X
n Equation (14) can be written in the following format
wðx; tÞ ¼ φi ðxÞ qi ðtÞ (12)
i¼0
q€ þ 2ξωn q_ þ ω2n q þ k1 q3 þ k2 q5 ¼ mpi V ðtÞ (18)
Parameters PZT-5H Brass By rearranging equation (44), the observer error dynamic
can be simplified as
Length ðlÞ (mm) 24.53 24.53
Width ðYÞ (mm) 6.4 6.4 ~x_ 2 þ h~x2 ¼ δðx; ~xÞ (46)
Thickness ðtÞ (mm) 0.265 0.140
Mass density ðρÞ (kg/m3) 7500 9000 According to the conditions presented in equations (23)
Elastic modulus ðcE ; cb Þ (GPa) 51.51 105 and (24), the error function δðx; ~xÞ stays bounded, which
Piezoelectric constant ðe31 Þ ðC=m2 Þ 13.11 — means that the state variables can be estimated accurately. In
addition, if the observer variable h is chosen to be large
Permittivity constant ðε33 Þ (nF/m) 25.5 —
enough, the effect of error modeling on the performance of
Figure 4. Theoretical agreement upon optimizing nonlinear parameters to reflect experimental observations.
the observer is decreased, and consequently, it increases the signal and its derivatives. However, an observer can change
precision of the estimated velocity. the dynamic of a system; therefore, the stability of system
with this new dynamic should be investigated. In this re-
gard, the following dynamic is considered
3.2. The global stability analysis
yðnÞ ¼ f ð Þ þ gð Þum (47)
In the presence of a designed sliding mode controller, the
state variables should converge to the sliding surface in where u is an input and y is the measurable output of
a finite time, and accordingly, they have to follow a desired a system. The functions f and g may depend on u, y, their
8 Journal of Vibration and Control 0(0)
derivatives, and disturbances. If it is assumed that z1 ¼ u, If the following assumptions hold, it can be concluded
z2 ¼ u1 ,…, zm ¼ um1 , and x1 ¼ y; x2 ¼ y1 ,…, xn ¼ yn1 , that the control system stays stable with the presence of an
and also, v ¼ um is considered as a control input of the observer.
augmented system, the augmented system in the state space
can be expressed as Assumption 1. For all ðx; z; dÞ 2 U × Γ1 , where Γ1 is
a known compact subset of Rp
x_i ¼ xiþ1 1≤i≤n 1 (48)
jgðx; z; dÞj ≥ kg > 0 (53)
x_n ¼ f ðx; z; dÞ þ gðx; z; dÞv (49) This assumption implies that the system has to be in
minimum phase and it does not comprise an internal dy-
z_j ¼ zjþ1 1≤j≤m 1 (50) namic. To investigate this assumption for the foregoing
control system, it can be written that
z_m ¼ v (51) y ¼ x1 (54)
y€ ¼ x_2 ¼ μ þ μ1 x21 ^x2 ω2n x1 k1 x31 k2 x51 þ mpi u
(56)
Figure 5. The phase plots obtained from experiment and the x_n ¼ f ðx; η; dÞ þ gðx; η; dÞv (58)
theoretical framework.
Figure 6. The phase plots from the linear and nonlinear models in different input voltages.
Shahabi et al. 9
(T226-H4-203X). The actuator consists of a brass substrate Then, the experiments were conducted to guarantee the
which is covered by two lead zirconate titanate identical performance of the designed controller in the tracking of
layers (series connection). The physical properties of the different position trajectories.
layers are listed in Table 1, and the fundamental frequency
of the actuator is equal to 2637 rad/s.
4.2. Nonlinear dynamic identification
The experimental setup includes an amplifier (EPA-104-
230), and two data acquisition cards (PCI-1710 and PCI- Now, the proposed dynamic model can be validated, and the
1716) which can capture data with a frequency of 20 kHz nonlinear parameters can be evaluated by means of the
(Figure 3). Moreover, the deflection of the actuator’s tip is available experimental setup. In this regard, a sinusoidal
measured by means of a laser displacement sensor (op- input voltage is applied on the piezoelectric actuator. The
toNCDT 2300) with a resolution of 10 nm. frequency of the input signal should be near the first res-
The achieved nonlinear dynamic model was experimen- onance frequency, such as 2512, 2580, 2637, 2680, and
tally identified to find the actuator nonlinear parameters. 2736 rad/s. Whereas the laser displacement sensor is set on
Figure 11. Position tracking results (a) Tracking of xd ¼ 20 sinð2500tÞ. (b) Tracking of xd ¼ 30 sinð2630tÞ. (c) Tracking of xd ¼ 40 sin
ð2700tÞ. (d) Tracking of xd ¼ 60 sinð2800tÞ.
12 Journal of Vibration and Control 0(0)
the tip of the actuator, the steady-state displacements are Table 3. Steady state tracking errors.
recorded at nine different input voltages with the sampling
Desired trajectory Error (%)
time of 0.0001 s. The average of all peak displacements over
the time response of the actuator is depicted as an amplitude r ¼ 20 sinð2500tÞ 5.6
on the frequency response plot. Most of the previous re- r ¼ 30 sinð2630tÞ 4.3
searches used the amplitude of the frequency responses to r ¼ 40 sinð2700tÞ 1.8
identify the nonlinear model. Indeed, the phase responses r ¼ 60 sinð2800tÞ 3.1
were generally ignored. However, the main target of this r ¼ 95 sinð2500tÞ 4.9
study is to evaluate the nonlinear parameters by considering
the amplitude and the phase frequency responses, simul-
taneously. Therefore, the phase values should be obtained
experimentally at different frequencies and amplitudes. As 4.3. Observer-based position control
a result, the following values are obtained for the unknown In this section, the obtained mathematical model of ac-
nonlinear parameters tuator, with the designed controller and observer, is im-
plemented in MATLAB Simulink, and unknown gains are
k1 ¼ 6:94 × 1017
determined in accordance with the constraints by con-
ducting numerical iterations. These parameters are pre-
k2 ¼ 1:443 × 1028 sented in Table 2.
Figure 8 depicts the experimental setup with the de-
μ1 ¼ 6:28 × 1013 signed observer and controller in the loop.
To evaluate the performance of observer, the observer
The plots which are shown in Figure 4 illustrate that the output signal is compared with the time derivative of
nonlinear parameters are identified correctly, and also, the actuator position, which is shown in Figures 9 and 10.
suggested model properly predicts the vibration ampli- After implementing the controller with a proper gain,
tude for different input voltages. different signal trajectories with frequencies near the
As it is observed in Figure 5, the experimental data are resonance are applied to the actuator. Figure 11 shows the
also in a good agreement with the theoretical results to corresponding results for the tracking of different desired
estimate the response phase. trajectories.
The results showed that, within the identification do- Moreover, to evaluate the accuracy of dynamic model
main, the suggested mathematical framework accurately and the performance of designed controller and observer,
predicts the overall dynamic behavior of the actuator near the tracking is also conducted outside of the identification
the resonant actuation with the maximum error of 7% and range; the corresponding result is shown in Figure 12.
the average error of 5.5%. As it is observed, the tracking error in this set of ex-
To clarify the difference between the linear and non- periments is in the range of 1–6% of maximum amplitude.
linear models, the corresponding phase responses are also Also Table 3 shows the steady state errors of the foregoing
shown in Figure 6. experiments.
The deviation of the linear model from the nonlinear It should be noted that the maximum and the average
one are also illustrated at some frequencies before and steady-state errors for different input signals within the
after resonance in Figure 7. identification range are obtained as 5.6% and 3.7%,
Shahabi et al. 13
respectively. Moreover, outside of the identification range, the suggested control approach was resulted in 4.9% steady-
the suggested control approach is able to track a desired state error which is quite acceptable.
trajectory with the steady state error of 4.9%. Finally, the
control effort for the tracking of one of the desired tra-
jectories such as xd ¼ 95 sinð2500tÞ is shown in Figure 13. Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with
respect to the research, authorship, and/or publication of this
5. Conclusion article.
In this work, a nonlinear mathematical framework was first
developed and experimentally validated for a piezoelectric Funding
bimorph actuator which is excited at high frequencies near The author(s) received no financial support for the research, au-
its first bending mode natural frequency. This nonlinear thorship, and/or publication of this article.
mathematical framework was obtained by resorting to
Hamilton’s principles, Galerkin’s method, and the single
mode assumption. The nonlinear parameter identification ORCID iDs
was conducted by means of amplitude and phase frequency Pouyan Shahabi https://orcid.org/0000-0002-6665-3907
responses, simultaneously. Then, to control the actuator Hamed Ghafarirad https://orcid.org/0000-0001-7292-143X
position at high frequencies, a boundary layer sliding mode
controller was used. The corresponding control law was References
derived based on the Lyapunov theorem so that the stability
Andreaus U, Dell’Isola F and Porfiri M (2004) Piezoelectric
of the system can be guaranteed. Then, the stability of the passive distributed controllers for beam flexural vibrations.
system was evaluated in the presence of a high-gain Journal of Vibration and Control 10: 625–659.
reduced-order observer as well. Finally, the precision and Chao PC-P, Liao P-Y, Tsai M-Y, et al. (2011) Robust control design
feasibility of the suggested control method were tested by for precision positioning of a generic piezoelectric system with
means of tracking different trajectories, even outside of the consideration of microscopic hysteresis effects. Microsystem
identification range. Technologies 17: 1009–1023.
The results showed that within the identification domain, dell’Isola F, Maurini C and Porfiri M (2004) Passive damping of
the suggested mathematical framework accurately predicts beam vibrations through distributed electric networks and
the overall dynamic behavior of the actuator near the res- piezoelectric transducers: prototype design and experimental
onant actuation with the maximum error of 7% and the validation. Smart Materials and Structures 13: 299.
Ghafarirad H, Rezaei SM, Sarhan AAD, et al. (2014a) Continuous
average error of 5.5%. Also, the experimental control results
dynamic modelling of bimorph piezoelectric cantilevered ac-
indicated that the suggested control system not only has tuators considering hysteresis effect and dynamic behaviour
satisfactory performance but also attenuate the chattering analysis. Mathematical and Computer Modelling of Dynamical
effects in the control effort. The maximum and average Systems 21: 130–152.
steady state errors for different tested input signals within Ghafarirad H, Rezaei SM, Zareinejad M, et al. (2014b) Disturbance
the identification range were obtained as 5.6% and 3.7%, rejection-based robust control for micropositioning of piezo-
respectively. Moreover, for the out of identification range, electric actuators. Comptes Rendus Mécanique 342: 32–45.
14 Journal of Vibration and Control 0(0)
Jain RK, Majumder S, Ghosh B, et al. (2015) Design and Qingsong X and Minping J (2014) Model reference adaptive
manufacturing of mobile micro manipulation system with control with perturbation estimation for a micropositioning
a compliant piezoelectric actuator based micro gripper. Journal system. IEEE Transactions on Control Systems Technology 22:
of Manufacturing Systems 35: 76–91. 352–359.
Jemai A, Najar F and Chafra M (2017) Parametric analysis of Rakotondrabe M, Clévy C and Lutz P (2010) Complete open loop
multilayered unimorph piezoelectric vibration energy har- control of hysteretic, creeped, and oscillating piezoelectric
vesters. Journal of Vibration and Control 23: 2538–2553. cantilevers. IEEE Transactions on Automation Science and
Jingang Y, Chang S and Yantao S. (2009) Disturbance-observer- Engineering 7: 440–450.
based hysteresis compensation for piezoelectric actuators. Sadek I, Kucuk I, Zeini E, et al. (2009) Optimal boundary control
IEEE/ASME Transactions on Mechatronics 14: 456–464. of dynamics responses of piezo actuating micro-beams. Ap-
Khadraoui S, Rakotondrabe M and Lutz P (2014) Interval force/ plied Mathematical Modelling 33: 3343–3353.
position modeling and control of a microgripper composed of Shirazi MJ, Salarieh H, Alasty A, et al. (2012) Tip tracking
two collaborative piezoelectric actuators and its automation. control of a micro-cantilever Timoshenko beam via piezo-
International Journal of Control, Automation and Systems 12: electric actuator. Journal of Vibration and Control 19:
358–371. 1561–1574.
Khalil HK (2014) Nonlinear Control. London: Pearson Higher Stanton SC, Erturk A, Mann BP, et al. (2010a) Nonlinear pie-
Education. zoelectricity in electroelastic energy harvesters: modeling and
Leadenham S and Erturk A (2015a) Nonlinear M-shaped broad- experimental identification. Journal of Applied Physics 108:
band piezoelectric energy harvester for very low base accel- 074903.
erations: primary and secondary resonances. Smart Materials Stanton SC, Erturk A, Mann BP, et al. (2010b) Resonant mani-
and Structures 24: 055021. festation of intrinsic nonlinearity within electroelastic micro-
Leadenham S and Erturk A (2015b) Unified nonlinear elec- power generators. Applied Physics Letters 97: 254101.
troelastic dynamics of a bimorph piezoelectric cantilever for Stanton SC, Erturk A, Mann BP, et al. (2012) Nonlinear non-
energy harvesting, sensing, and actuation. Nonlinear Dynamics conservative behavior and modeling of piezoelectric energy
79: 1727–1743. harvesters including proof mass effects. Journal of Intelligent
Leadenham S and Erturk A (2019) Mechanically and electrically Material Systems and Structures 23: 183–199.
nonlinear non-ideal piezoelectric energy harvesting framework Tan D, Yavarow P and Erturk A (2018a) Nonlinear elastodynamics
with experimental validations. Nonlinear Dynamics: 1–17. of piezoelectric macro-fiber composites with interdigitated
Liu J, Chen X, Gao J, et al. (2016) Multiple-source multiple- electrodes for resonant actuation. Composite Structures 187:
harmonic active vibration control of variable section cylindrical 137–143.
structures: a numerical study. Mechanical Systems and Signal Tan D, Yavarow P and Erturk A (2018b) Resonant nonlinearities of
Processing 81: 461–474. piezoelectric macro-fiber composite cantilevers with in-
Liu J, Chen X, Yang L, et al. (2017) Analysis and compensation of terdigitated electrodes in energy harvesting. Nonlinear Dy-
reference frequency mismatch in multiple-frequency feedfor- namics 92: 1935–1945.
ward active noise and vibration control system. Journal of Thomas O, Ducarne J and Deü J-F. (2011) Performance of pie-
Sound and Vibration 409: 145–164. zoelectric shunts for vibration reduction. Smart Materials and
Mehrdad Pourkiaee S, Khadem SE and Shahgholi M (2017) Structures 21: 015008.
Nonlinear vibration and stability analysis of an electrically Vidoli S and Dell’Isola F (2001) Vibration control in plates by
actuated piezoelectric nanobeam considering surface effects uniformly distributed PZT actuators interconnected via electric
and intermolecular interactions. Journal of Vibration and networks. European Journal of Mechanics – A/Solids 20:
Control 23: 1873–1889. 435–456.
Mirafzal SH, Khorasani AM and Ghasemi AH (2016) Optimizing Yu S and He S (2013) Accurate free vibration analysis of cantilever
time delay feedback for active vibration control of a cantilever piezoelectric panel carrying a rigid mass. Journal of Vibration
beam using a genetic algorithm. Journal of Vibration and and Control 19: 1187–1198.
Control 22: 4047–4061. Zhang Y and Xu Q (2017) Adaptive sliding mode control with
Oh S and Khalil HK (1997) Nonlinear output-feedback tracking parameter estimation and Kalman filter for precision motion
using high-gain observer and variable structure control. Au- control of a piezo-driven microgripper. IEEE Transactions on
tomatica 33: 1845–1856. Control Systems Technology 25: 728–735.