vibration

Article
Stochastic Stability of a Class of MEMS-Based
Vibratory Gyroscopes under Input Rate Fluctuations
Mohamed Bognash and Samuel F. Asokanthan *
Department of Mechanical and Materials Engineering, The University of Western Ontario,
London, ON N6A 5B9, Canada; mbognash@uwo.ca
* Correspondence: sasokant@uwo.ca; Tel.: +1-519-661-2111 (ext. 88907)




Received: 18 May 2018; Accepted: 16 June 2018; Published: 19 June 2018


Abstract: The influence of stochastic fluctuations in the input angular rate of a class of single
axis mass-spring microelectromechanical (MEM) gyroscopes on the system stability is investigated.
A white noise fluctuation is introduced in the coupled 2-DOF model that represents the system
dynamics for the purposes of stability prediction. Numerical simulations are performed employing
the resulting set of stochastic differential equations (SDEs) that govern the system dynamics. The SDEs
are discretized using the higher-order Milstein scheme for numerical computations. Simulations
via the Euler scheme, as well as the measure of the largest Lyapunov exponent are employed for
validation purposes due to a lack of similar analytical solutions or experimental data. Responses have
been predicted under different noise fluctuation magnitudes and different input angular rates for
stability investigations. A parametric study is performed to estimate the noise intensity stability
threshold for a range of quality factor values at different input angular rates. The predicted results
show a nonlinear dependence of the threshold on the quality factors for different input rates.
Under typical gyroscope operating conditions, a realistic frequency mismatch appears to have
insignificant influence on system stability. It is envisaged that the present quantitative predictions
will aid improvements in performance, reliability, and the design process for this class of devices.

Keywords: gyroscope; stochastic differential equation; dynamic stability; white noise; frequency
mismatch; instability; angular rate fluctuation

1. Introduction
Numerous emerging applications use a micro-machined angular rate sensor or gyroscope as
a stand-alone unit or as part of an inertial measurement unit (IMU). In particular, traction control systems,
ride stabilization, and rollover detection in automotive applications currently use the micro-machined
angular rate gyroscopes. Many other applications, such as digital video camera stabilization systems,
missile guidance systems, and platform stabilization systems take advantage of using this class of device
(see, e.g., [1]). The low manufacturing cost, moderate performance, and miniature nature of these
gyroscopes are the reasons behind their acceptance in these applications [2]. The accuracy of this class
of devices is likely to be influenced by a number of sources, including a combination of temperature,
vibration, acoustic, shock, thermal cycling, and humidity [1]. For instance, microelectromechanical systems
(MEMS) devices can be exposed to shock during fabrication, deployment and operation [3], as well as
vibratory excitation resulting from the environment. The undesirable sources can change the dynamic
behavior of MEMS devices and, hence, affect their performance. For example, vibratory gyroscopic
systems in automotive applications are susceptible to environmental vibrations due to road unevenness
and due to other sources. The excitation from road irregularity is modeled as a stationary random process
with road roughness suggested in the ISO standard. Furthermore, white noise is often used to model the
input of road displacement excitation (see, e.g., [4]). An approximate modeling of rail track unevenness

Vibration 2018, 1, 69–80; doi:10.3390/vibration1010006 www.mdpi.com/journal/vibration

Vibration 2018, 1 70

and road irregularities using a white noise excitation is also reported in the prediction of the random
vibration response of rail and land vehicles [5]. Aerospace, marine, and other applications are expected to
benefit from developments in this class of MEMS gyroscopes. In such applications, the effects of noise
and vibration, which stem from the aerodynamic/wave environment, as well as combustion process in
propellers, need to be investigated in detail.
MEMS-based gyroscopes, at present, are considered as low-accuracy rate-grade sensors compared
to the tactical and inertial-grade sensors which are known to have moderate to high accuracies.
However, MEMS-based gyroscopes offer cost and size advantages compared to the tactical and inertial
grade counterparts. Thus, to enhance the accuracy of MEMS-based gyroscopes to that of a tactical or
inertial grade, further improvements in accuracy, as well as their drift performance, are warranted.
To this end, several recent studies, as well as development efforts, are underway. Micro-machined
gyroscopes have different design and sensing principles, but almost all the configurations utilize the
transfer of energy between two vibration modes of a structure where the Coriolis’ effect is exploited for
the precise sensing of angular rotation rates. These devices, in contrast to the conventional gyroscopes,
do not contain rotating elements; thus, they are suitable for batch micro fabrication and miniaturization.
An external vibratory excitation is used in this class of MEMS device for their operation, but few
designs take advantage of parametric excitation (see, e.g., [6,7]).
In order to achieve further improvement and development of this class of sensors, dynamics and
stability of this type of gyroscopes have been of interest in the recent past. Mechanical coupling
between the drive and detection modes of a single mass-spring micro-machined vibratory gyroscope
was studied by Mochida et al. [8] giving importance to the mechanical coupling, then reducing the
coupling via the introduction of new designs and fabrication structures [8]. A precise mathematical
model for a dual-axis gyroscope was developed by Davis [9], and linear and non-linear suspensions
have been considered in this study. This gyroscope was fabricated in the laboratory, but required
further investigation prior to commercialization. In the same study, a more accurate model for the
single-mass spring gyroscope considering the coupling effect for both the driving and sensing axes has
been developed. However, in all of the above studies, instability investigations were not performed.
The effect of stochastic fluctuations in the input angular rate on the stability of a single-axis mass-spring
vibratory gyroscope has been investigated by Asokanthan and Wang [10]. An approximate analytical
method based on the stochastic averaging procedure has been employed to investigate the stability
of a vibratory MEMS gyroscope system. Closed-form conditions for mean-square stability of the
dynamic response are obtained for the case of exponentially correlated noise. It may be noted that
this study focussed only on narrow band frequency ranges that correspond to certain multiples and
combinations of system natural frequencies, and predictions for white-noise fluctuations were not
presented. Recently, a simplified two degree-of-freedom dynamic model of a ring-based gyroscope is
employed to identify the stability of the system when the input angular speed is exposed to random
fluctuations [11]. The system response is numerically predicted when the input angular rate is subjected
to a white noise fluctuation using the higher-order Milstein scheme that discretizes the governing
stochastic differential equations (SDEs). The stability threshold values for noise intensity have been
identified using largest Lyapunov exponents’ measure for various damping values. The current
study focuses on the stability of the single axis mass-spring gyroscope subjected to stochastic angular
speed fluctuation. It is known that use of white noise in the stability investigations presents a more
practical representation of the speed fluctuation generated by the environment when compared to the
consideration of harmonic vibration or narrow band noise. Thus, the introduction of white noise is
expected to predict a more accurate dynamic response of these devices, as well as the physical systems
they are mounted on. Hence, the effect of wide-band random fluctuation in the input angular rate
on the dynamic stability of the single axis mass-spring structure gyroscope is the main interest of the
present study.
Owing to the fact that obtaining a closed-form analytical solution for a multi-dimensional system
of stochastic differential equations is cumbersome due to their highly non-differentiable character of the
Vibration 2018, 1 71

realization of the Wiener process [12], a number of iterative approaches to integrate SDE’s numerically
have been developed in the recent past. The most widely used methods are Euler–Maruyama, Euler–Heun,
Milstein, derivative-free Milstein (Runge–Kutta approach), and stochastic Runge–Kutta [13]. In the
present study, the higher-order Milstein scheme is employed to simulate the time response so that the
stochastic response of2,ax FOR
Vibration 2018, single
PEERaxis mass-spring rate gyroscopes can be quantified for certain3 parameters
REVIEW of 12 of
interest. Based on the obtained responses, the behavior of the dynamical system is analyzed. To this end,
have been developed in the recent past. The most widely used methods are Euler–Maruyama,
the characteristic Lyapunov exponents of the stochastic response is evaluated to determine the stability
Euler–Heun, Milstein, derivative-free Milstein (Runge–Kutta approach), and stochastic Runge–Kutta [13].
thresholds.InThe
the effects
present of quality
study, factors andMilstein
the higher-order the magnitude of the angular
scheme is employed speed
to simulate thefluctuation,
time responseassowell as the
that the stochastic
frequency mismatch response
on system of a single
stability haveaxisbeen
mass-spring rate gyroscopes can be quantified for certain
quantified.
parameters of interest. Based on the obtained responses, the behavior of the dynamical system is
2. analyzed.
Governing To this end, the characteristic Lyapunov exponents of the stochastic response is evaluated
Equations
to determine the stability thresholds. The effects of quality factors and the magnitude of the angular
speed axis
The single fluctuation, as well asrate
mass-spring the frequency
gyroscope mismatch
modelon system
used stability
in the have study
present been quantified.
is based on the model
developed by Davis [9] and later presented in the work by Asokanthan and Wang [10]. The developed
2. Governing Equations
governing equations for the gyroscopic system is employed to investigate the stochastic fluctuation
The single axis mass-spring rate gyroscope model used in the present study is based on the
of the input angular rate. The gyroscope configuration consists of a lumped point mass (proof mass)
model developed by Davis [9] and later presented in the work by Asokanthan and Wang [10]. The
at the center and four
developed springs
governing that support
equations the mass
for the gyroscopic as shown
system is employedin Figure 1 with
to investigate the dissipating
stochastic linear
fluctuation
viscous damping. Asofathe input angular
gyroscope, therate.
massTheisgyroscope
excited configuration consistsx-direction
along the driving of a lumped point
(alsomass
referred to as
the driving(proof mass) atby
direction), theancenter and four
external springs
periodic that F
force support
(t) = Fthe mass as shown in Figure 1 with
0 sinωt, where the excitation frequency
dissipating linear viscous damping. As a gyroscope, the mass is excited along the driving x-direction
ω is chosen close to the natural frequency of the system. The gyroscopic system is subjected to
(also referred to as the driving direction), by an external periodic force 𝐹(𝑡) = 𝐹 𝑠𝑖𝑛𝜔𝑡, where the
an input angular rate
excitation Ω about
frequency 𝜔 isthe z-axis
chosen where
close Ω represents
to the natural frequencythe quantity
of the to be
system. The measured.
gyroscopic The mass is
system
constrainedis to
subjected
oscillateto in
an the
input
x–yangular
plane.rate 𝛺 about
Thus, the z-axis
a rotation where
rate, 𝛺 represents
Ω, about the quantity
the z-axis inducestoanbeoscillatory
measured.
Coriolis force in theThe massdirection
y-axis is constrained
(alsotoreferred
oscillate in
tothe
as x–y
the plane.
sensing Thus, a rotation
axis). rate, 𝛺,oscillatory
The y-axis about the motion
z-axis induces an oscillatory Coriolis force in the y-axis direction (also referred to as the sensing axis).
is sensed and used as a basis for the measurement of the angular rate.
The y-axis oscillatory motion is sensed and used as a basis for the measurement of the angular rate.

cy ky x

kx 2 2 cx 

2 2

Mass F0 sin  t

cx kx
2 ky cy 2
2 2

Figure 1. Single-axis mass-spring gyroscope.

Figure 1. Single-axis mass-spring gyroscope.
The system of coupled second-order differential equations that governs the motion of the gyroscope
derived by Davis [9] is presented as follows:
The system of coupled second-order differential equations that governs the motion of the
𝑚 𝑥̈is
gyroscope derived by Davis [9] + presented
𝑐 𝑥̇ + 𝑘 𝑥 −as 𝑥 − 2𝑚𝛺 𝑦̇ − 𝑚𝛺̇ 𝑦 = 𝐹 𝑠𝑖𝑛𝜔𝑡
𝑚𝛺follows:
(1)
.. 𝑚
. 𝑦̈ + 𝑐 𝑦̇ + 𝑘 𝑦 − 𝑚𝛺 𝑦 + 2𝑚𝛺 . . ̇ 𝑥 = 0,
𝑥̇ + 𝑚𝛺
m x + c x x + k x x − mΩ2 x − 2mΩ y − mΩ y = F0 sinωt
where 𝑚 represents the proof mass. The coefficients 𝑐 and 𝑐 denote the . viscous damping constants (1)
.. . .
in the x- and y-directions,
m y + crespectively,
y y + k y y −while
mΩ 𝑘2 and 𝑘 are the
y + 2mΩ x +spring
m Ω xconstants
= 0, in the x- and y-
directions, respectively. Additionally, the gyroscope angular rate measurement about the z-direction
is denoted the
where m represents by 𝛺proof
. The main
mass.interest of this workcis and
The coefficients on examining
cy denote thethe
effects of some
viscous important
damping constants in
x
parameters on the system stability. Hence, the steady state part of the response does not play a role.
the x- and y-directions, respectively, while k x and k y are the spring constants in the x- and y-directions,
To this end, consideration of the homogenous part of the equations of motion in the absence of the
respectively. Additionally,
excitation the gyroscope
force is adequate. angular
Equation (1) can thenrate measurement
be written about the z-direction is denoted
in matrix form:
by Ω. The main interest of this work is on examining the effects of some important parameters on
the system stability. Hence, the steady state part of the response does not play a role. To this end,
Vibration 2018, 1 72

consideration of the homogenous part of the equations of motion in the absence of the excitation force
is adequate. Equation (1) can then be written in matrix form:
" # " ωx
# " . #
1 0 .. Qx −2Ω . ωx2 − Ω2 −Ω
q+ ωy q+ . q=0 (2)
0 1 2Ω Qy Ω ω y − Ω2
2

kx ky mω x mωy
with ωx2 = m, ωy2 = m, Qx = cx , and Qy = cy . The vector q represents the vector of the
T
system generalized coordinates, i.e.,[ x y] , while ωx and ωy , respectively, are the undamped natural
frequencies associated with the x- and y-directions. The quality factors representing damping in the
x and y directions are denoted by Q x and Qy , respectively. The quality factor is a dimensionless
parameter that indicates the energy losses within a resonant element. In general, energy dissipation in
vibratory MEMS is governed by several mechanisms, including viscous damping, dissipation through
the substrate, thermo-elastic dissipation, and resonator surface effects [14]. The quality factor indicates
energy loss relative to the amount of energy stored within the system. Thus, the higher the quality
factor the lower the rate of energy loss and, hence, oscillations will decay more slowly. It may be
noted that the stiffness matrix includes the centrifugal force term Ω2 , which takes a negative value
for the present system. Hence, overall system stiffness decreases with higher angular velocity that
may lead to lower system stability. The damping matrix, apart from representing viscous dissipation,
includes the gyroscopic coupling term 2Ω, which is dependent on the input angular velocity. Moreover,
.
for a constant angular rate, the term Ω = 0. This case is not practical in the presence of fluctuations
in the angular rate. However, for the system under investigation, the contributions of the associated
. . .
terms, Ωq1 and −Ωq2 , are considered negligible when compared to the gyroscopic terms −2Ωq1
.
and 2Ωq2 at high angular rates where instability becomes an issue, it is sufficient to approximate
. .
Ωq1 = −Ωq2 = 0 for the purpose of stability analysis. Implications and limitations due to this
assumption are highlighted in the Results and Discussion section.
The numerical schemes to be used in the present study require the system equations to be
transformed to a set of first-order stochastic differential equation in incremental form. For this purpose,
Equation (1) is first transformed into a system of four first-order differential equations to accommodate
. .
these methods. A set of four state variables is defined as x1 = q1 , x2 = q1 , x3 = q2 , and x4 = q2 ,
and the random fluctuation in the input angular rate is then incorporated in the system to form
the governing equations of motion in the standard first-order SDE form. Equation (1) is written in
state-space form as:
. . ωx
 
x1 = x2 , x2 = − ωx2 − Ω2 x1 − x2 + 2Ω x4
Qx
. .
  ωy (3)
x3 = x4 , x4 = −2Ωx2 − ωy2 − Ω2 x3 − x4 .
Qy
The random fluctuations in the input angular velocity are assumed to be represented by a white
noise process. A Brownian motion function W (t) is employed for this purpose to simulate the random
fluctuations taking advantage of its first-time derivatives as Gaussian white noise (see, e.g., [15]).
Introducing white noise, ξ (t) = dW/dt, with a noise intensity magnitude, µ0 , to the nominal input
angular velocity, Ω0 , for representing the random fluctuations, the input angular velocity is written as:

Ω = Ω0 + µ0 ξ ( t ). (4)

The centrifugal component in the equations of motion which are governed by the terms Ω2 in
Equations (3) can be evaluated using Equation (4) as:

Ω2 = Ω20 + 2µ0 Ω0 ξ (t) + µ20 ξ 2 (t), (5)

Vibration 2018, 1 73

since the random fluctuation term, µ0 ξ (t), is considered small relative to the nominal angular rate, Ω0 ,
the last term in Equation (5), µ20 ξ 2 (t), is negligible due to its lower-order of smallness. In addition,
for the purposes of simulations, a noise intensity ratio, µ, is introduced to characterize the noise
intensity magnitude, µ0 , as:
µ0 max(ξ (t))
µ= . (6)
max(Ω0 )
Combining (Equations (3)–(5)), a system of standard SDEs that represents the gyroscope motion
is obtained as:      
dx1 a1 [X(t)] b1 [X(t)]
 dx   a X t   b X t 
2   2 [ ( )]   2 [ ( )] 
= dt +  dW, (7)

 dx3   a3 [X(t)]   b3 [X(t)] 


dx4 a4 [X(t)] b4 [X(t)]

where the vectors a[X(t)] and b[X(t)], respectively, are known as the drift and the diffusion vectors
(see, e.g., [12,15]). For the present system, the vector elements written in terms of the state vector
X(t) = [ x1 x2 x3 x4 ] T are given by:
  ωx
a1 [X(t)] = x2 , a2 [X(t)] = − ωx2 − Ω20 x1 − x2 + 2Ω0 x4 ,
Qx
  ωy
a3 [X(t)] = x4 , a4 [X(t)] = −2Ω0 x2 − ωy2 − Ω20 x3 − x ,
Qy 4
b1 [X(t)] = 0, b2 [X(t)] = 2µ0 Ω0 x1 + 2µ0 x4 ,

b3 [X(t)] = 0, b4 [X(t)] = −2µ0 x2 + 2µ0 Ω0 x3 .

3. Numerical Simulation Scheme

It is known that obtaining analytical solutions of the standard first-order stochastic differential
equations, which govern the gyroscope motion and were formulated in Section 2, is cumbersome,
the present study attempts to employ numerical solutions of such systems. To this end, the discretization
of SDE using Itô–Taylor expansion, which leads to the higher-order Milstein scheme [16], is employed.
The Milstein scheme is a numerical scheme for solving stochastic deferential equations with a strong order
of convergence. The method uses Itô’s lemma to increase the accuracy of the approximation by adding
the second-order term to approximate numerical solution of a stochastic differential equation. In this
technique, the sequence of values of the Milstein approximation at the instants of the time discretization
can be computed in a similar way to those of the deterministic case. The main difference is that we now
need to generate the random increments of the Wiener process. For a given time discretization, the Milstein
scheme determines values of the approximating process at the discretization times only. In order to clarify
the approach, the following first-order standard SDE for a scalar dependent variable is used:

dX (t) = a[ X (t)]dt + b[ X (t)]dW (t), (8)

along with an Itô–Taylor expansion, where a[ X (t)] and b[ X (t)], respectively, denote the drift and the
diffusion terms while dW (t) represents the driving Wiener process. Use of Itô’s Lemma leads to:

d f [ X (t)] = L0 a[ X (t)]dt + L1 b[ X (t)]dW (t), (9)

where:
∂ ∂ 1 ∂2 ∂
L0 ≡ +a + b2 2 , and L1 ≡ b .
∂t ∂X 2 ∂X ∂X
When Itô’s Lemma is iterated to obtain constant integrands for the higher order terms,
and assuming that a and b are not direct functions of t, the integrated form of Equation (9) becomes:
Vibration 2018, 1 74

Zt Zt
X (t) = X (t0 ) + a[ X (t0 )] ds1 + b[ X (t0 )] dW (s1 )
t0 t0 (10)
 
1 1
+ b[ X (t0 )]b́[ X (t0 )] [W (t) − W (t0 )]2 − (t − t0 ) + O((δt)3/2 )
2 2

where O((δt)3/2 ) represents terms that include (δt)3/2 , or terms of higher order, and (´) denotes the
derivative with respect to variable X. This equation forms the theoretical basis for both Euler and
Milstein schemes [16]. It may be noted that the Euler scheme is constructed using the first three terms
of this expansion, while incorporation of the fourth term yields the Milstein scheme.
Considering the time interval [ti , ti+1 ] by choosing t0 = ti , t = ti+1 , ∆t = ti+1 − ti and ∆Wi =
W (ti+1 ) − W(ti ), the discretized form of the Milstein method is formulated as:

1
X (ti+1 ) = X (ti ) + a[ X (ti )]∆t + b[ X (ti )]∆Wi + b[ X (ti )]b́[ X (ti )]{(∆Wi )2 − ∆t}. (11)
2

Equation (11), when extended to multi-dimensional systems, yields the uth component of the state
vector employing the Milstein scheme for numerical computations and takes the general form:
m m
∑ bu,j [X(ti )]∆Wi + ∑
j
Xu (ti+1 ) = Xu (ti ) + au [ X (ti )]∆t + L j1 bu,j2 [X(ti )]I( j1 ,j2 ) (12)
j =1 j1 ,j2 =1

where the drift and diffusion terms, the driving Wiener process and the variables, are written in vector
form. In Equation (12):

ti+1 Zs1
d Z
∂
∑b
j1 k,j1 j j
L = [X(ti )] k , I( j1 ,j2 ) = dWs21 dWs12
k =1 ∂X
ti ti

where b[X(ti )] is the diffusion coefficient matrix, d is the number of dimensions, and m represents the number
of independent Weiner processes [15]. In the special case when j1 = j2, the following integral is obtained:

1 j 2
I( j1 ,j2 ) = {(∆Wi1 ) − ∆t}.
2
The vector-based scheme presented in Equation (12), considering the system drift and diffusion
coefficient matrices, is employed for the purposes of performing numerical computations to solve the
system of equations that govern the gyroscope response. To this end, considering Equation (7) and
setting d to 4 and m to 1 in Equation (12), the response takes the form:

Xu (ti+1 ) = Xu (ti ) + au [X(ti )]∆t + bu,1 [X(ti )]∆Wi

4
1 k,1 ∂bu,1 [X(ti )] (13)
+ ∑ 2
b [X(ti )]
∂X k
{(∆Wi )2 − ∆t}, u = 1, 2, . . . , d.
k =1

The resulting four equations are employed in the prediction of the gyroscope response.

4. Results and Discussion

In the present numerical study, a smooth increase of the input angular rate Ω0 from zero to
different practical values is employed for the purposes of response predictions considering a noise
intensity ratio µ as defined in Equation (6). For the purposes of numerical simulations, the parameters
of the single-axis gyro parameters as shown in Table 1 have been used.
Conforming to the goal of the present study, namely the stability investigation, the time response of
the gyroscope when subjected to appropriate initial disturbance is examined. In the simulations, an initial
Vibration 2018, 1 75

displacement of the driving coordinate, q1, is imposed with a value of 10−5 m and the displacement of sensing
coordinate q2 is then computed to evaluate the gyroscope response. However, for high quality factors, which are
typical in this class of devices, the convergence of the Milstein scheme is very slow even for a deterministic system.
In order to accelerate the convergence, the first two terms in Equation (13) which represent the deterministic part
of the numerical solution are evaluated using a fourth-order Runge–Kutta scheme and the values at each time
step are used in each iteration for obtaining the complete solution incorporating random terms.
Vibration 2018, 2, x FOR PEER REVIEW 7 of 12
Table 1. Single-axis gyroscope parameters [9].
using a fourth-order Runge–Kutta scheme and the values at each time step are used in each iteration
for obtaining the complete solution incorporating random
Parameter terms.
Notation Value
Gyroscope mass m 3.6 × 10−10 kg
Table 1. Single-axis gyroscope parameters [9].
Nominal X-axis natural frequency ωx 26.2 kHz
Nominal Y-axisParameter
natural frequency Notation
ωy Value
26.2 kHz
Gyroscope
X-axis mass
quality factor Q
𝑚x 3.6 ×20–1000
10 kg
Y-axis
Nominal quality
X-axis factor
natural frequency Q
𝜔y 20–1000
26.2 kHz
Nominal Y-axis natural frequency 𝜔 26.2 kHz
X-axis quality factor 𝑄 20 − 1000
The convergence of the simulation algorithm has been
Y-axis quality factor 𝑄
tested for many quality
20 − 1000
factors, Qx and Qy
ranging from 20 to 1000 for different relative noise intensity ratio values µ in the range of 6 × 10−6 to 1.55 in
order to guaranteeThe convergence
the accuracy, of theas simulation
well as thealgorithm has been tested
appropriateness, of the forschemes.
many quality Veryfactors,
good 𝑄convergence
and 𝑄 ranging
has been reached for afromtime20stepto 1000
of 1for
×different
10−5 s;relative
however, noise 1intensity
× 10−6ratio s isvalues
used𝜇 for in the rangeaccuracy
more of and
6 × 10 to 1.55 in order to guarantee the accuracy, as well as the appropriateness, of the schemes.
a simulation time of 0.1 s is chosen for the simulations, which are adequate for the remainder of the study.
Very good convergence has been reached for a time step of 1 × 10 s; however, 1 × 10 s is used
The higher-order
for more Milstein
accuracy and scheme is known
a simulation time ofto 0.1
achieve higher
s is chosen accuracy
for the compared
simulations, which areto adequate
the Euler method;
for the remainder of the study. The higher-order Milstein scheme is
therefore, the first approach is chosen in the present investigations for stability predictions. known to achieve higher accuracy
Time responses
compared to the Euler method; therefore, the first approach is chosen
of the two methods for the purposes of verifying the response predictions are generated and depicted in the present investigations
for stability predictions. Time responses of the two methods for the purposes of verifying the
in Figuresresponse
2–4. The figures show the responses of the gyroscope at different fluctuation magnitudes in
predictions are generated and depicted in Figures 2–4. The figures show the responses of
input angular rate at 2πatrad/s
the gyroscope demonstrating
different various stability
fluctuation magnitudes behaviors.
in input angular rate atMaximum relative noise intensity,
2𝜋 rad/s demonstrating
as definedvarious stability
in Equation (6)behaviors.
has beenMaximum
used as arelative
magnitudenoise intensity,
measureasfor defined in Equation
representing (6) has beenfluctuation.
environment
In Figure 2,used
theassystem
a magnitude measure
response forquality
for representing environment
factor of 500 and fluctuation.
maximum In Figure 2, the system
relative response measure
noise intensity
for quality factor of 500 and maximum relative noise intensity measure 𝜇 = 0.0126 demonstrates a
µ = 0.0126stable
demonstrates a stable behavior for this sufficient level of damping. Increasing the noise intensity
behavior for this sufficient level of damping. Increasing the noise intensity to a sufficiently high
to a sufficiently high value to cause
value to cause a noticeable a noticeable
disturbance in thedisturbance
system, alongin withthehigh
system,
enough along with
quality high
factors enough
that do quality
not suppress the oscillation, causes oscillatory motion as shown in Figure 3.
factors that do not suppress the oscillation, causes oscillatory motion as shown in Figure 3. It may be noted It may be noted that a
certain
that a certain threshold
threshold intensity measure
intensity measure for foreach
eachquality factor
quality is associated
factor with thewith
is associated transition to instability
the transition to instability
and this measure can be computed using the time responses. Thus, crossing this threshold with a
and this measure can be computed using the time responses. Thus, crossing
high quality factor or, alternatively, with higher noise intensity, leads to system instability, as shown
this threshold with a high
quality factor or, alternatively,
in Figure 4. with higher noise intensity, leads to system instability, as shown in Figure 4.
Amplitude, (m)

Figure 2. Example of the stable time response.

Figure 2. Example of the stable time response.
Vibration 2018, 1 76
Vibration
Vibration2018,
2018,2,
2,xxFOR
FORPEER
PEERREVIEW
REVIEW 88 of 12
of 12

(m)
Amplitude,(m)
Amplitude,

Figure
Figure3.
Figure 3.Example
3. Exampleof
Example ofthe
of themarginally
the marginallystable
marginally stable time
stable time response.
time response.
response.
(m)
Amplitude,(m)
Amplitude,

Figure
Figure 4. Example of the unstable time response.
Figure 4.
4. Example
Example of
of the
the unstable
unstable time
time response.
response.

Further,
Further, itit may
may bebe noted
noted thatthat the
the predicted
predicted responses
responses from
from the
the Euler
Euler andand Milstein
Milstein schemes
schemes are are
similarFurther, it may be noted that the predicted responses from the Euler and Milstein schemes are
similar with slightly larger amplitude values for Milstein than the values predicted by the
with slightly larger amplitude values for Milstein than the values predicted by the Euler
Euler
similar with
scheme, which slightly
does notlarger
take amplitude
the valuesterms
higher-order for Milstein
into than the
account. values
Thus, the predicted
Euler scheme by produces
the Euler
scheme,
scheme, which does not
which does not takeand take the higher-order
the higher-order terms into account.
terms into account. Thus, the
Thus, the EulerEuler scheme produces
scheme produces
an
anunder-predicted
under-predictedresponse response andexhibitsexhibitsincreased
increasedsystem
systemstability
stabilitycompared
comparedto tothe
theMilstein
Milsteinscheme
scheme
an under-predicted
predictions. response andcertain exhibits increased of system stability comparedwhich to the Milstein scheme
predictions. Accordingly,
Accordingly, under under certain conditions
conditions of noise
noise intensity
intensity values,
values, which are are close
close toto the
the
predictions.
threshold, a Accordingly,
stable system under
behaviorcertain
is conditions
predicted of
via noise
the intensity
Euler scheme values,
while which
an are
unstableclose to the
system
threshold, a stable system behavior is predicted via the Euler scheme while an unstable system
threshold,isapredicted
behavior stable systemthe behavior is predicted via the Eulerinscheme while an unstable system behavior
behavior is predicted by by the Milstein
Milstein scheme,
scheme, as as depicted
depicted in Figure
Figure 5.
5. Therefore,
Therefore, due due to to aa lack
lack of
of any
any
is predicted
exact by the Milstein scheme, as depicted in Figure 5. Therefore, due to a lack of any exact
exact analytical
analytical or or experimental
experimental results results to to verify
verify the
the acquired
acquired data
data against,
against, the the accuracy
accuracy of of the
the
analytical model
employed or experimental
should be results
assessed tobefore
verifyperforming
the acquired data
the against,sweep
parameter the accuracy
which of the employed
includes varying
employed model should be assessed before performing the parameter sweep which includes varying
model
the should be assessed before performing the parameter sweep which includes varying the relevant
the relevant
relevant system
system parameters
parameters to to characterize
characterize the the system
system behavior.
behavior. ForFor this
this purpose,
purpose, aa measure
measure for for
system
stability parameters
via the to
largest characterize
Lyapunov the system
exponent behavior.
(LLE) method For
is this purpose,
employed to a measure
study the for
system stability
dynamic via
stability via the largest Lyapunov exponent (LLE) method is employed to study the system dynamic
the largest
stability Lyapunov exponent (LLE) method is employed to study the system dynamic stability of
stability of
of the
the two two schemes.
schemes. The The concept
concept of of Lyapunov
Lyapunov exponents
exponents is is one
one ofof the
the most
most powerful
powerful tools
tools
for
foranalyzing
analyzingnonlinear
nonlinearand andstochastic
stochasticsystems.
systems.Lyapunov
Lyapunovexponents
exponentsquantify
quantifythe therate
rateat atwhich
whichorbits
orbits
Vibration 2018, 1 77

the two schemes. The concept of Lyapunov exponents is one of the most powerful tools for analyzing
nonlinear and
Vibration stochastic
2018, systems.
2, x FOR PEER REVIEWLyapunov exponents quantify the rate at which orbits on an 9 attractor
of 12
converge or diverge as the system evolves in time and provide a direct measure of the stability of
those on an attractor
orbits. converge or
One exponent is diverge
definedasfortheeach
system evolves in time
dimension, and provide
representing thea direct
averagemeasure
rate ofof growth
the
stability
or decay along of each
those of orbits. One exponent
the principal axesisin
defined
the d Efor each dimension,
-dimensional staterepresenting
space. Thethe average
largest rate
Lyapunov
of growth or decay along each of the principal axes in the 𝑑 -dimensional state space. The largest
exponent specifies the maximum average rate of divergence, or convergence of the orbits. Any system
Lyapunov exponent specifies the maximum average rate of divergence, or convergence of the orbits.
with at least one positive Lyapunov exponent will inevitably become unstable, with the magnitude
Any system with at least one positive Lyapunov exponent will inevitably become unstable, with the
of themagnitude
exponent of reflecting on the
the exponent time scale
reflecting that
on the thescale
time systemthatdynamics
the systemwill diverge.
dynamics willFor complicated
diverge. For
systems, determination of Lyapunov exponents analytically is, in general, impossible
complicated systems, determination of Lyapunov exponents analytically is, in general, impossible but calculating
them but calculatingusing
numerically them numerically
a time series using a time series
is extremely is extremely attractive/easier
attractive/easier [17]. Therefore,
[17]. Therefore, it
it is sufficient
is sufficient
to calculate to calculate
the largest the largest
Lyapunov Lyapunov
exponent exponent for characterizing
for characterizing system
system stability. Instability. In the
the present study,
present
a practical usestudy,
of the a practical
LLE search use algorithm,
of the LLE search
basedalgorithm,
on the methodbased of
on time
the method
delay as of described
time delay in as [18],
described
is utilized. Figurein [18], is utilized.
6 shows Figure 6 shows
the obtained the obtained
threshold value ofthreshold
the noisevalue of theatnoise
intensity whichintensity at of
the onset
which the onset of system instability commences
system instability commences at an input angular rate of 2π rad/s. at an input angular rate of 2𝜋 rad/s.
Amplitude, (m)

Figure
Figure 5. Under-predictionof
5. Under-prediction of results
results by
byEuler
Eulerscheme.
scheme.

The compatibility of the results from the Milstein and Euler schemes, as well as the error of
The compatibility
omitting of the terms
the higher-order resultsinfrom the Milstein
the Euler scheme, and have Euler schemes,
been further as well as
investigated the error
through a of
number
omitting of points utilizing
the higher-order thein
terms LLE.
theFor
Eulerthisscheme,
purpose,have
three been
pointsfurther
in a highly stable region
investigated with quality
through a number
factorutilizing
of points values ofthe104,LLE.
413, For
and this
706, and the respective
purpose, maximum
three points in a relative
highly noise
stableintensity
region values of
with quality
0.06505, 0.02670, and 0.01845 are chosen and displayed in Figure 6.
factor values of 104, 413, and 706, and the respective maximum relative noise intensity values of The used algorithms yield
negative
0.06505, LLE
0.02670, values
and for these
0.01845 points and
are chosen indicating
displayeda stable system6.under
in Figure The usedthesealgorithms
conditions. yield
However,
negative
LLE values for these points indicating a stable system under these conditions. However,Milstein
LLE values evaluated by Euler’s scheme have been found to be lower than those predicted by LLE values
scheme by 1.7–5.4%. It may be noted that the percentage difference is larger when the conditions are
evaluated by Euler’s scheme have been found to be lower than those predicted by Milstein scheme by
closer to the threshold, while it is minimal when the system is highly stable or unstable. This character
1.7–5.4%. It may be noted that the percentage difference is larger when the conditions are closer to the
has also been demonstrated via the time responses in Figures 2, 4, and 5. As expected, the time response
threshold, while
predicted byitthe
is Euler
minimal when
scheme tendsthetosystem is highlythestable
under-estimate or unstable.
response when comparedThis character
with that has also been
predicted
demonstrated via thescheme.
via the Milstein time responses
This mayinbe Figures 2, 4 to
attributed andthe5. higher-order
As expected,term the time responsewhich
participation, predicted
are by
included in the latter scheme. Further, the compatibility of the results of the two schemes in the
the Euler scheme tends to under-estimate the response when compared with that predicted via the
marginal
Milstein scheme.stability
This region
may be has been examined
attributed to thetohigher-order
show that theterm two participation,
systems produce approximately
which are included
in theconsistent instability
latter scheme. thresholds.
Further, Six marginal points
the compatibility of theas shown
resultsinofFigure
the two 6 have been chosen
schemes in theformarginal
this
purpose,
stability region three
haspoints slightly above
been examined the threshold
to show that the points
two within
systems theproduce
unstable approximately
region and three consistent
points
slightly
instability below within
thresholds. Sixthe stable region.
marginal pointsThe LLE analysis
as shown for both
in Figure schemes
6 have beenpredict
chosen similar signs
for this at
purpose,
each point, which indicates a correct prediction of the stability at this marginal region. At this low angular
three points slightly above the threshold points within the unstable region and three points slightly
speed, a small noise intensity of value 2.5% of angular speed is seen to affect the system stability for
below within the stable region. The LLE analysis for both schemes predict similar signs at each point,
high quality factors. Thus, values greater than this threshold are likely to destabilize the system.
which indicates a correct prediction of the stability at this marginal region. At this low angular speed,
Vibration 2018, 1 78

a small noise2018,
Vibration intensity of value
2, x FOR PEER 2.5% of angular speed is seen to affect the system stability
REVIEW 10 for
of 12high
quality factors. Thus, values greater than this threshold are likely to destabilize the system.

0.18
Stability Threshold Point
0.16 O LLE Study Point Close to Stability Threshould
LLE Study Point in Highly Stable Region
0.14 0
= 2 rad/s

0.12

0.1

0.08 Unstable

0.06

0.04

Stable
0.02

0
0 100 200 300 400 500 600 700 800 900 1000
Quality Factor , Q

Figure
Figure 6. Stability
6. Stability boundary
boundary inin theµ𝜇−−ζ𝜁 space
the space (Ω
(𝛺 ==2𝜋
2πrad/s).
rad/s).
0

It is known that the gyroscopic systems are not immune to the external noise effects that can
It is known that the gyroscopic systems are not immune to the external noise effects that can
alter their dynamic behavior. Such noise usually stems from environmental and operational sources
alter their
and candynamic behavior.
be exerted Such noise
on the system at anyusually
frequency stems
rangefrom environmental
depending and operational
on the source. Furthermore,sources
the
and can be exerted
allowance on the
of noise system
effects on theat MEMS
any frequency
system isrange depending
an important designonfactor,
the source. Furthermore,
especially at low
damping of
the allowance ratios
noise(high quality
effects on factor) whichsystem
the MEMS is a characteristic feature design
is an important of modern MEMS
factor, gyroscopic
especially at low
damping systems.
ratiosTherefore, a parametric
(high quality factor)study
whichis performed to assess feature
is a characteristic the noiseofintensity
modern stability
MEMSthreshold
gyroscopic
systems.for aTherefore,
number of aquality factor values
parametric studyatisdifferent
performed inputtoangular
assess rates.
the noise To this end, an increment
intensity of the
stability threshold
system quality
for a number factorfactor
of quality of approximately 50 is employed
values at different and the system
input angular rates. To noise
thisstability
end, an threshold
increment is of
obtained via the bisection search method. Further refinement is employed
the system quality factor of approximately 50 is employed and the system noise stability threshold is in the low quality factor
region for analyzing the predicted time response. Figure 7 shows that the noise intensity threshold is
obtained via the bisection search method. Further refinement is employed in the low quality factor
decreased at higher quality factors. Increasing the angular speed, for high quality factors, reduces the
region for analyzing the predicted time response. Figure 7 shows that the noise intensity threshold is
fluctuation noise intensity threshold value to a value less than 1% of the angular speed. In such cases,
decreased
where atthe
higher
noisequality factors.
effects are Increasing
of concern, use ofthe angular
an active speed, for
disturbance high quality
rejection controlfactors,
has beenreduces
shown the
fluctuation noise intensity threshold value to a value less than 1% of the angular
to eliminate/reduce the noise effects (see, e.g., [19]). It may be noted that the present study is limitedspeed. In such cases,
whereto thethe predictions
noise of of
effects are threshold
concern,values,
use of while the disturbance
an active actual applications,
rejectionascontrol
well ashas thebeen
operating
shown to
conditions may
eliminate/reduce thedictate the absolute
noise effects levels[19]).
(see, e.g., of fluctuation
It may berequired
noted that to cause system study
the present instability. It is to
is limited
worth noting
the predictions that such predictions
of threshold can only
values, while be achieved
the actual via laboratory
applications, as well or in
assitu
themeasurements. The
operating conditions
employed approach uses the assumption of noise in angular rate in the absence
may dictate the absolute levels of fluctuation required to cause system instability. It is worth noting of angular acceleration.
Owing
that such to the non-differentiable
predictions nature of white
can only be achieved noise, this term
via laboratory or incannot be evaluated for The
situ measurements. the solution
employed
process followed for numerically solving the resulting SDE. This assumption limits the inclusion of
approach uses the assumption of noise in angular rate in the absence of angular acceleration. Owing to
the angular acceleration terms in the analysis; however, to overcome this difficulty, one can use the
the non-differentiable nature of white noise, this term cannot be evaluated for the solution process
fluctuation in the angular acceleration, for low angular velocities, which is typical in the case of
followed for numerically
gyroscopes. solving
Alternatively, the resulting
a narrow SDE.assumption
band noise This assumption limits the
for the angular inclusion
velocity of the angular
to simulate the
acceleration
velocityterms in the is
fluctuation analysis; however,
also feasible. Thesetoapproaches
overcome are thislikely
difficulty, one can
to increase theuse the fluctuation
prediction accuracies, in the
angular butacceleration,
are beyond the forscope of the present
low angular study.
velocities, which is typical in the case of gyroscopes. Alternatively,
a narrow band noise assumption for the angular velocity to simulate the velocity fluctuation is also
feasible. These approaches are likely to increase the prediction accuracies, but are beyond the scope of
the present study.
Vibration 2018, 2, x FOR PEER REVIEW 11 of 12
Vibration 2018, 1 79

0.8

0
=3 rad/s
0.7 =2 rad/s
0

0
= rad/s
0.6 = /2 rad/s
0
= /4 rad/s
0
0.5

0.4

0.3

0.2

0.1

0
0 100 200 300 400 500 600 700 800 900 1000
Quality Factor , Q

Figure7.7.Stability
Figure Stabilityboundary
boundaryin
inthe
the 𝜇 space.
µ − 𝜁ζ space.

In addition, studies performed on varying input angular rates of practical significance for typical
In addition, studies performed on varying input angular rates of practical significance for typical
gyroscope applications revealed that the increase in angular velocity decreases noise thresholds.
gyroscope applications revealed that the increase in angular velocity decreases noise thresholds.
Further studies on the effect of the frequency mismatch, which originates form asymmetry in both of
Further studies on the effect of the frequency mismatch, which originates form asymmetry in both
supporting springs stiffness and/or proof- mass are performed. A change of frequency mismatch for
of supporting springs stiffness and/or proof- mass are performed. A change of frequency mismatch
low and moderate values in the range up to 10% seems to have insignificant effects on the noise
for low and moderate values in the range up to 10% seems to have insignificant effects on the noise
threshold. However, unrealistic frequency mismatch values close to 20% causes a reduction in the
threshold. However, unrealistic frequency mismatch values close to 20% causes a reduction in the
noise threshold. It is worth noting that such values for the frequency mismatch are not of practical
noise threshold. It is worth noting that such values for the frequency mismatch are not of practical
significance under typical gyroscope operating conditions. It is expected that these results will lead
significance under typical gyroscope operating conditions. It is expected that these results will lead
to the identification of critical stability regions for this class of devices, which can then improve the
to the identification of critical stability regions for this class of devices, which can then improve the
device performance, reliability, and the design process.
device performance, reliability, and the design process.
5. Conclusions
5. Conclusions
The dynamic
The dynamic behavior
behavior ofof aa mass-spring
mass-spring typetype MEMS-based
MEMS-based single-axis
single-axis vibratory
vibratory gyroscope
gyroscope is is
investigated when subjected to stochastic fluctuations in the input angular rate. The
investigated when subjected to stochastic fluctuations in the input angular rate. The effect of random effect of random
fluctuations is
fluctuations is introduced
introduced into
into the
the gyroscope
gyroscope governing
governing equations
equations to to form
form stochastic
stochastic differential
differential
equationsthat
equations thatarearediscretized
discretized using
using thethe Milstien
Milstien scheme
scheme to predict
to predict the response
the response numerically.
numerically. The
The Euler
scheme, as well as the largest Lyapunov exponents, have also been used in this study as a validationa
Euler scheme, as well as the largest Lyapunov exponents, have also been used in this study as
validation
tool toollack
due to the dueof toanalytical
the lack ofsolutions
analyticalandsolutions and experimental
experimental data. A
data. A nonlinear nonlineartrend
decreasing decreasing
of the
obtained threshold values for the noise intensity at different quality factor values has beenvalues
trend of the obtained threshold values for the noise intensity at different quality factor observedhas
been observed while a similar decreasing trend of noise thresholds with the angular
while a similar decreasing trend of noise thresholds with the angular rate increase is also predicted. rate increase is
also predicted.
Variations in theVariations
frequencyinmismatch
the frequency mismatch
have shown have shown
insignificant insignificant
influence influence
on system on system
stability under
stability under typical gyroscope operating conditions. It is envisaged that the
typical gyroscope operating conditions. It is envisaged that the present quantitative predictions willpresent quantitative
predictions
lead will lead toinimprovements
to improvements performance,in performance,
reliability, and the reliability, and the for
design process design
thisprocess
class offor this class
devices via
of devices via identification of critical
identification of critical stability regions. stability regions.

Author Contributions:
Author Contributions: M.B.
M.B.and
andS.F.A.
S.F.A.formulated the
formulated system
the model;
system M.B.
model; performed
M.B. the the
performed numerical simulations;
numerical and
simulations;
M.B.M.B.
and andand
S.F.A. analyzed
S.F.A. the the
analyzed results andand
results wrote the the
wrote manuscript.
manuscript.

Funding: This
Funding: This research
research was
was funded
funded partially
partially by
by Natural
Natural Science
Science and
andEngineering
EngineeringResearch
ResearchCouncil
Council (NSERC)
(NSERC) of
of
Canada discovery grant, grant# RGPIN/250432-2012 and scholarship awarded to the first author by Libyan-North
Canada discovery grant, grant# RGPIN/250432-2012 and scholarship awarded to the first author by Libyan-North
American Scholarship program.
American Scholarship program.
Vibration 2018, 1 80

Acknowledgments: The support of the Natural Science and Engineering Research Council (NSERC) of Canada
discovery grant and the Libyan-North American Scholarship program are acknowledged.
Conflicts of Interest: The authors declare no conflict of interest. The funding sponsors had no role in the design of
the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision
to publish the results.

References
1. Acar, C.; Schofield, A.; Trusov, A.; Costlow, L.; Shkel, A. Environmentally robust MEMS vibratory gyroscopes
for automotive applications. IEEE Sens. J. 2009, 9, 1895–1906. [CrossRef]
2. Xing, H.; Hou, B.; Lin, Z.; Guo, M. Modeling and compensation of random drift of MEMS gyroscopes based on
least squares support vector machine optimized by chaotic particle swarm optimization. Sensors 2017, 17, 2335.
[CrossRef] [PubMed]
3. Younis, M.; Jordy, D.; Pitarresi, J. Computationally efficient approaches to characterize the dynamic response
of microstructures under mechanical shock. J. Microelectromech. Syst. 2007, 16, 628–638. [CrossRef]
4. Zuo, L.; Pei-Sheng, Z. Energy harvesting, ride comfort, and road handling of regenerative vehicle suspensions.
J. Vib. Acoust. 2013, 135. [CrossRef]
5. Schiehlen, W. White noise excitation of road vehicle structures. Sadhana 2006, 31, 487–503. [CrossRef]
6. Passaro, V.M.N.; Cuccovillo, A.; Vaiani, L.; De Carlo, M.; Campanella, C.E. Gyroscope technology and
applications: A review in the industrial perspective. Sensors 2017, 17. [CrossRef] [PubMed]
7. Hosseini-Pishrobat, M.; Keighobadi, J. Robust vibration control and angular velocity estimation of
a single-axis MEMS gyroscope using perturbation compensation. J. Intell. Robot. Syst. 2018, 90, 1–19.
[CrossRef]
8. Mochida, Y.; Tamura, M.; Ohwada, K. A micro machined vibrating rate gyroscope with independent beams
for the drive and detection modes. Sens. Actuator A Phys. 2000, 80, 170–178. [CrossRef]
9. Davis, W.O. Mechanical Analysis and Design of Vibratory Micro Machined Gyroscopes. Ph.D. Thesis,
University of California, Berkeley, CA, USA, 2001.
10. Asokanthan, S.F.; Wang, T. Dynamic instabilities in a single-axis gyroscope subjected to stochastic angular
rate perturbations. Probab. Eng. Mech. 2009, 24, 600–607.
11. Asokanthan, S.F.; Arghavan, S.; Bognash, M. Stability of ring-type MEMS gyroscopes subjected to stochastic
angular speed fluctuation. ASME. J. Vib. Acoust. 2017, 139, 040904–040907. [CrossRef]
12. Higham, D.J. An algorithmic introduction to numerical simulation of stochastic differential equations.
SIAM Rev. 2001, 43, 525–546. [CrossRef]
13. Schaffter, C.T. Numerical Integration of SDEs: A Short Tutorial; Swiss Federal Institute of Technology in
Lausanne (EPFL): Lausanne, Switzerland, Unpublished work; 2010.
14. Zotov, A.S.; Simon, B.R.; Prikhodko, I.P.; Trusov, A.A.; Shkel, A.M. Quality factor maximization through
dynamic balancing of tuning fork resonator. IEEE Sens. J. 2014, 14, 2706–2714. [CrossRef]
15. Kloeden, P.E.; Platen, E. Numerical Solution of Stochastic Differential Equations; Springer: Berlin, Germany, 1999.
16. Higham, D.J.; Kloeden, P.E. MAPLE and MATLAB for Stochastic Differential Equations in Finance.
Programming Languages and Systems in Computational Economics and Finance; Nielson, S.B., Ed.; Springer Science
+ Business Media: Dordrecht, The Netherlands, 2002.
17. Yang, C.; Wu, Q. On stability analysis via Lyapunov exponents calculated from a time series using nonlinear
mapping-a case study. Nonlinear Dyn. 2010, 59, 239–257. [CrossRef]
18. Kliková, B.; Raidl, A. Reconstruction of phase space of dynamical systems using method of time delay. In WDS’11
Proceedings of the Contributed Papers: Part III—Physics; Matfyz Press: Prague, Czech Republic, 2011.
19. Zheng, Q.; Dong, L.; Lee, D.H.; Gao, Z. Active disturbance rejection control for MEMS gyroscopes. IEEE Trans.
Control Syst. Technol. 2009, 17, 1432–1438. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).