Professional Documents
Culture Documents
_
Tolgay Kara1 and Ilyas Eker2
Abstract
This paper presents the design and implementation of adaptive control with approximate input–output linearization for underactuated open-loop
unstable non-linear mechanical systems. Control of a ball and beam (BB) mechanism is selected as a benchmark problem for testing the designed con-
trol. The method of input–output linearization is reviewed and an adaptive input–output linearizing control design procedure is given. An approximate
BB model is developed using Euler–Lagrange equations, and input–output linearization-based adaptive tracking control is designed for the system. The
model is parameterized with respect to ball mass for adaptive tracking, and the proposed control structure is tested via computer simulations and
experiments. The results present the tracking performance of designed control for various ball masses, and reveal the proposed method’s capability to
cover ball mass variations over non-adaptive control. The proposed control exhibits improved error performance in the presence of parametric varia-
tions in the plant. Results of the BB control case reveal successful control of underactuated non-linear mechanisms when a system parameter is
unknown or time varying.
Keywords
Adaptive control, ball and beam, input–output linearization, parameter variation, tracking
applications (Deutscher, 2005; Ho and Hedrick, 2015; Kim The rest of the paper is organized as follows: the next sec-
and Oh, 1999; Leith and Leithead, 2001; Xiang and Wikander, tion presents the basis for the proposed control design, giving
2004). Some of these applications include an adaptive details on the input–output linearization and its adaptive ver-
approach to the method (Ghanadan and Blankenship, 1996; sion. Then, the BB system model is presented and adaptive
Kanellakopoulos et al., 1991; Ko et al., 1999; Kwon and Choi, control design steps are given. Simulation and experimental
2014; Marino, 1997; Mohammed et al., 2012; Raimundez results are presented, followed by conclusions, stressing the
et al., 2014; Sahnehsaraei et al., 2014). The work by Marino basic results and contributions.
(1997) gives a general view of adaptive control of non-linear
systems, with special emphasis on the method of feedback line-
arization. Kanellakopoulos et al. (1991) give the general Adaptive input–output linearizing control
approach to the design of adaptive controllers for feedback lin-
earizable systems. The mentioned works stress the adaptive Input–output linearizing control
control of a non-linear system, presuming that it is feedback
Consider a single-input–single-output (SISO), non-linear, nth
linearizable, and the deduced results cannot be used for sys-
order system of the following form:
tems that are not linearizable. The approach of Hauser et al.
(1992) to control of systems with approximate linearization
presents important conclusions for systems that are not linear- x_ = f ðxÞ + gðxÞu
ð1Þ
izable. Today, adaptive control of non-linear systems with y = hðxÞ
input–output linearization is still an evolving field of study.
The ball and beam (BB) system exhibits strong non-linear where x 2 Rn is the state vector, u 2 R is the input, y 2 R is the
behaviour. Although the non-linearity is involved in the beam output, and f, g and h are smooth functions on Rn . If the sys-
angle and ball position processes, the ball position presents tem in (1) has a strong relative degree r, then it is possible to
the prominent non-linear characteristic of the system. This is transform the model into the following equivalent structure:
a consequence of the fact that a linear mathematical model
8
can express the beam angle process, with the non-linear effect > z_ 1 = z2
>
>
being in a way the disturbing factor on the system. This non- >
< ..
linear effect comes from static and viscous friction, and stick- .
>
> z_ r1 = zr ð2Þ
slip, which are neglected in many similar studies. However, >
>
:
the relationship between the beam angle and the ball position, z_ r = f1 ðzÞ + g1 ðzÞu
which constitutes the ball position process model, cannot be z_ i = Ci ðzÞ, i = r + 1, . . . , n
expressed in linear terms for it is inherently non-linear. The
BB mechanism is open-loop unstable, underactuated and not
with
input–output linearizable. The ball position control is, there-
fore, a complicated problem. Non-linear adaptive control of
the system has been designed and tested via simulations by f1 ðzÞ = Lrf h ð3Þ
Andreev et al. (2002), Bhushan (2013), Hauser et al. (1992)
and Krishna et al. (2012), and others applied non-linear con- g1 ðzÞ = Lg Lr1
f h ð4Þ
trol methods to the BB system (Chang et al., 1998; Chang
et al., 2013; Cheok and Loh, 1987; Hirschorn, 2002). where Lna bðxÞ is the nth order Lie derivative of the function
In the previous studies mentioned above, various methods bðxÞ with respect to f (Khalil and Grizzle, 2002; Marino et al.,
have been proposed and tested for non-linear and open-loop 1989; Sastry and Bodson, 2011). Note that, if x = 0 is an equi-
unstable systems with ill-defined relative degree using robust librium point of the undriven system, then Ci represents the
and adaptive techniques as well as computational and artifi- so-called zero dynamics of the system in (2). Now, the follow-
cial intelligence methods, and the BB mechanism has been ing control law is introduced for (2):
used as a benchmark problem for justification of the obtained
results. The present paper gives results of the study on non-
1
linear adaptive control of a BB mechanism using input– u= ðf1 ðzÞ + yÞ ð5Þ
g1 ðzÞ
output linearization with an approximation approach. The
proposed approach is novel in the sense that it combines
approximate input–output linearization with an adaptive which yields the following linear input–output relationship in
strategy, making the control system capable of recovering differential equation form:
tracking performance despite parameter variations in the sys-
tem. The proposed control aims to improve the tracking accu- yðrÞ = y ð6Þ
racy in the case of parametric variations in the plant under
control. With this goal, the BB system with varying ball mass The resulting linear system given by (6) permits the control
parameter is tested to track the desired trajectory with mini- design to control the rth order dynamics stably. For a special
mal error using the proposed method, which justifies the pro- case where the strong relative degree is equal to the system
posed approach. The designed control is tested via computer order, zero dynamics do not exist, and the diffeomorphism
simulations and real-time experiments, and the results are defined by (2) reduces to the following:
presented in graphical form.
8
> z_ 1 = z2 depends on whether the non-linear model has been adequately
>
< ..
. developed.
ð7Þ
>
>
: z_ n1 = zn Remark 2. In cases where system characteristics are not ade-
z_ n = f1 ðzÞ + g1 ðzÞu quately modelled, or system uncertainties and parameter var-
iations are significantly large, the non-linearity deviates from
Substituting (5) in (7) yields the following linear differential
an a priori structure. As a result, exact cancellation may not
equation:
be possible and the closed-loop system is not linear. This may
lead to severe damage if uncancelled non-linearity involves
yðnÞ = y ð8Þ
unstable modes.
Remark 2 depicts a problematic issue regarding practical
Remark 1. One can conclude by the resulting closed-loop implementation of the technique of input–output linearizing
dynamics in (8) that it is possible to control all modes of the control. If the source of deviation from the a priori model is
system (1) using the control signal y, providing n = r. For the parametric uncertainty or variation, an adaptive approach is
sake of easy following and comprehension, the order of the a good alternative for maintaining control goals. Considering
system and the strong relative degree will be assumed to be the problem of tracking a reference signal yr for an SISO non-
equal from this point on. The results to be obtained can be linear system of the following form:
generalized for the case where they are not equal, by introdu-
cing a zero dynamics stability constraint (Slotine and Li, 8
< P
p P
p
1991). x_ = f0 ðxÞ + ai fi ðxÞ + g0 ðxÞ + ai gi ðxÞ u,
i=1 i=1 ð15Þ
In order to complete the control design, the input signal y :
y = hðxÞ,
of the linear system must be determined. Define the tracking
error as:
with
e = yr y ð9Þ
X
p X
p
f ðxÞ = f0 ðxÞ + ai fi ðxÞ, gðxÞ = g0 ðxÞ + ai gi ðxÞ, ð16Þ
The control requirements, reducing the steady-state error to i=1 i=1
zero and exhibiting robust transient performance, can be
achieved by defining an error dynamic of the following form: where ai, i=1, ., p are the unknown parameters. Assuming
that the order of the system and the relative degree are equal,
eðnÞ + an1 eðn1Þ + . . . + a1 e_ + a0 e = 0 ð10Þ (15) can be represented by the following normal form (note
that the system model is expressed in mixed z and x state vari-
where sn + an1 sn1 + . . . + a1 s + a0 is a Hurwitz polyno-
ables for easy following and derivation).
mial. Then, the dynamic in (5) guarantees that the system out-
put y approaches the desired output yr at the steady state. 8
Transient performance that determines the speed and shape >
> z_ 1 = z2
>.
<
of the response to reference changes is determined by selection ..
ð17Þ
of appropriate coefficients for the dynamics in (10). Using (8) >
> z_ n1 = zn
>
: z_ = Ln hðxÞ + uL Ln1 hðxÞ
and (9), (10) can be written as: n f g f
ðn1Þ
y = y(n)
r + an1 e + . . . + a1 e_ + a0 e ð11Þ Let ^ai denote the estimates of ai for i=1, ., p, and for simpli-
city of notation, define the following:
The control law in (11) depends not only on the tracking
error, but directly on the reference signal as well. Substituting X
p
(10) in all error terms of (11), the following combination of a ^f ðxÞ = f0 ðxÞ + ^ai fi ðxÞ, ð18Þ
feedforward and a feedback control laws is obtained (Marino i=1
et al., 1989):
X
p
y = yff + y fb ð12Þ g^ðxÞ = g0 ðxÞ + ^ai gi ðxÞ: ð19Þ
i=1
where
Then, using the certainty equivalence principle, linearizing
yff = y(n) + an1 y(n1)
+ . . . + a1 y_ r + a0 yr ð13Þ control is as follows:
r r
y fb = an1 y(n1) + . . . + a1 y_ + a0 y ð14Þ L^f Ln1
f0 hðxÞ + y
u= ð20Þ
Lg^ Ln1
f0 hðxÞ
Adaptive tracking with input–output linearization Note that the parameters ai are unknown, so we can design
The method of input–output linearization relies on exact can- the controller in terms of estimates of these parameters. Using
cellation of system non-linearity. The accuracy of the method (20), input signal y can be expressed as:
process, the accuracy of the model can also be improved by Remark 3. The denominator term in (44) is zero for x1 = 0 or
considering the motor dynamics (Nordin and Gutman, 2002). x4 = 0, which means the relative degree of the BB system is
Using Euler–Lagrange equations, the BB mechanical system not well defined and the linearizing law in (25) is not applica-
model is obtained as follows (Ogata, 2004): ble, as it becomes infinite whenever the ball position or rota-
tional speed of the beam becomes zero.
ð42Þ
Bg cosðx3 Þ
aðxÞ = ð50Þ
where B = m
, J T = J + Jb . mx21 + JT
J
m + b2
R
An approximate system for input–output linearization is
obtained by ignoring the centrifugal term (48) and the follow-
Approximate input–output linearization ing system in normal form is obtained (Hauser et al., 1992):
A Jacobian linearization of (42) is not possible, as the system 8
does not satisfy linearizability conditions (Hauser et al., >
> z_ 1 = z2
<
1992). Another option is input–output linearization, which z_ 2 = z3
ð51Þ
relies on defining a control input n that is expressed in terms >
> z_ = z4
: 3
of input u and the states, and expressing a linear dynamic z_ 4 = bðxÞ + aðxÞu
relationship between the new input n and output y. To apply
The system output is given by yð xÞ = x1 = z1 . Successive dif-
the method to a non-linear system, the system must have a
ferentiation of the output expression results in the following
well-defined relative degree. For the sake of ease of calcula-
fourth-order non-linear differential equation:
tion, neglecting the viscous friction terms yields the model
given by (43) in state-space form with applied torque as input
yð4Þ = bðxÞ + aðxÞu ð52Þ
and ball position as output. When input–output linearization
is applied to (43), control input in (44) is obtained with rela- The system in (52) is input–output linearizable with relative
tive degree 3. degree 4 and the linearizing input is given by
8
>
> x_ 1 = x2 bðxÞ + y
>
> u= ð53Þ
< x_ 2 = Bx1 x24 Bg sinðx3 Þ aðxÞ
x_ 3 = x4 ð43Þ
>
> 2m mg u with (49) and (50). Substituting (53) in (52) gives the following
>
> _
x 4 = 2 x 1 x 2 x 4 x cosðx3 Þ +
JT + mx21 1 JT + mx21
: JT + mx1
y = x1 linear differential equation, which reveals that the dynamic
2m mg
Bx2 x24 + Bgx4 sinðx3 Þ + 2Bx1 x4 x 1 x 2 x 4 + x 1 cosð x 3 Þ +y
mx21 + JT mx21 + JT
u= ð44Þ
2Bx1 x4
mx21 + JT
yð4Þ = y ð54Þ JT x1
f1 ðxÞ = 2 ð2x2 x4 + g cosðx3 ÞÞ ð58Þ
m0 x21 + JT
As the approximate plant model in (45) without centrifugal
term is used for input–output linearization, the validity of this 2m0 x21 + JT
g0 ðxÞ = 2 ð59Þ
approximation should be verified using simulation and experi- m0 x21 + JT
mental tests. Verification results are presented below. It is
worth noting that the tests are carried out using a plant model x21
g1 ðxÞ = 2 ð60Þ
where the centrifugal term is present. m0 x21 + JT
The system in (56) has the same structure as the model in (15)
Adaptive tracking controller with the following correspondence:
The BB mechanism has been investigated in detail by several T
f0 ðxÞ = x2 Bx1 x24 + Bg sinðx3 Þ x4 f0 ðxÞ ð61Þ
researchers in past years. The models used in these studies
have a common property of being independent of the ball
f1 ðxÞ = ½ 0 0 0 f1 ðxÞ T ð62Þ
mass. In other words, control design is assumed to be unaf-
fected by the variation of ball mass (Chang et al., 1998; g0 ðxÞ = ½ 0 0 0 g0 ðxÞ T ð63Þ
Hauser et al., 1992; Leith and Leithead, 2001). This is a con-
sequence of the approach of transforming the g-vector field in g1 ðxÞ = ½ 0 0 0 g1 ðxÞ T ð64Þ
the system equation using an invertible transformation
(Hauser et al., 1992). Control design is performed according with p = 1 and a1 = m. Following the results above, input–
to this modified model; therefore, control performance is output linearizing control is obtained using the certainty
expected to remain unchanged with a ball of any unpredicted equivalence principle as:
mass, which is not possible in practice. Variation of ball mass
has a strong interaction with the system behaviour, as ^bðxÞ + y
u= ð65Þ
depicted by the modelling procedure above. ^aðxÞ
Remark 4. In order to apply adaptive tracking, the system of where
concern must have the form in (15). This implies that system
equations must be linear in unknown parameters. The ^ 1 cosðx3 Þ
^bðxÞ = Bmgx ð2x2 x4 + g cosðx3 ÞÞ + Bgx24 sinðx3 Þ ð66Þ
unknown parameter for the normal system in (51) is mass m ^ 21 + JT
mx
and (49) obviously reveals that the system model is non-linear
in m. Bg cosðx3 Þ
^aðxÞ = ð67Þ
As a result of the discussion in the above remark, linear ^ 21 + JT
mx
parameterization of the system model in the ball mass is
required. The following procedure is followed to put the sys- and m
^ is the estimate for the unknown ball mass. The control
tem in the required form for adaptive control by linearizing input signal y is generated as:
the terms in (49) that depend on m about an average mass m0 .
ð3 Þ
y = y(4)
r a3 z4 y r a2 ðz3 €yr Þ
8 ð68Þ
>
> m ðm0 x1 Þ2 + JT m a1 ðz2 y_ r Þ a0 ðz1 yr Þ
>
> ffi 2
< mx21 + JT m0 x21 + JT
ð55Þ with polynomial ðs4 + a3 s3 + a2 s2 + a1 s + a0 Þ being strictly
>
> 1 x21 ð2m0 mÞ + JT
>
> Hurwitz. The adaptive tracking control design is complete
: mx2 + J ffi 2
1 T m0 x21 + JT with the adaptation law given below:
Substituting (55) in (43) and rearranging the model gives the ^_ = gWT e
m ð69Þ
following approximate system:
where g is the scalar adaptation gain, e and W are respectively
8 defined in (23) and (30).
>
> x_ 1 = x2
>
>
>
> 2
< _ 2 = Bx1 x4 Bg sinðx3 Þ
x
x_ 3 = x4 ð56Þ Simulation and experimental results
>
>
>
>
> x_ 4 = f0 ðxÞ + mf1 ðxÞ + g0 ðxÞ + mg1 ðxÞ u
> Tracking performances of the designed controllers are tested
:
y = x1 via a number of simulation experiments. The two control
approaches relying on approximate input–output lineariza-
where tion and its adaptive version are investigated for different ball
masses. Simulations are implemented on the BB system model
6 29:43
0:714x2 x24 + 7:01x4 sinðx3 Þ + 0:04x1 x4 x1 x2 x4 + x1 cosðx3 Þ + y
2 + 3x21 2 + 3x21
u= ð71Þ
4x1 x4
2 + 3x21
Hertz. Tracking performances of non-adaptive and adaptive
The experimental set-up used in the BB system control systems are given in Figures 7 and 8, respectively. A compari-
experiments is given in open loop by the diagram in Figure 3. son of the tracking errors of non-adaptive and adaptive con-
Figure 4 is a photograph of the experimental set-up. The trollers for the two ball masses is graphically presented in
results of simulation tests are given in Figures 5 and 6. Figure Figure 9. Inspection of plots in the figures leads to the conclu-
5 is a clear indication that the tracking performance of a non- sion that the track of the ball significantly deviates from the
adaptive approximate input–output linearizing controller desired trajectory when the ball mass changes in the non-
depends on the value of ball mass, and only a small change in adaptive approach, but the proposed adaptive control suc-
mass leads to a great deviation from the desired trajectory. ceeds in keeping the ball around the desired trajectory after a
Figure 5. Reference trajectory (solid) and ball position track (dotted) and tracking error for two different ball masses of (a) 0.02 kg and (b) 0.035
kg, using non-adaptive input–output linearization
Figure 6. Reference trajectory (solid) and ball position track (dotted), tracking error and ball mass estimate for two different ball masses of (a) 0.02
kg and (b) 0.035 kg, using adaptive input–output linearization.
transient of about 25 s. In Figure 9, it is clear that the pro- transient period, whereas the non-adaptive approach fails to
posed adaptive approach is capable of keeping the ball on the maintain the tracking accuracy for the ball mass value of
desired track for both ball masses with a small error after the 0.035 kg.
Figure 7. Experimental results: reference trajectory (solid) and actual ball position (dashed) for the ball masses of (a) 0.02 kg and (b) 0.035 kg using
non-adaptive approximate input–output linearization.
Figure 8. Experimental results: reference trajectory (solid) and actual ball position (dashed) for the ball masses of (a) 0.02 kg and (b) 0.035 kg using
adaptive approximate input–output linearization.
Figure 9. Experimental results: tracking error variations for the ball masses of (a) 0.02 kg and (b) 0.035 kg using non-adaptive (solid) and adaptive
(dashed) approximate input–output linearization.
tracking control law making use of approximate input– Bhushan B (2013) Indirect adaptive control of nonlinear systems.
output linearization is presented. The input–output lineariz- International Journal of Advanced Research in Computer Science
ing control law, feedforward and feedback control laws, and and Software Engineering 3(10): 1359–1365
parameter adaptation rule are derived. Convergence of the Cao L and Chen X (2015) Input–output linearization minimum slid-
ing mode error feedback control for synchronization of chaotic
tracking error and parameter adaptation despite variations in
system. Proceedings of the Institution of Mechanical Engineers,
ball mass is verified.
Part I: Journal of Systems and Control Engineering 229(8):
A complete mathematical model is developed for the BB 685–699.
mechanism, accounting for the dependence of output para- Chang BC, Kwatny H and Hu SS (1998) An application of robust
meters on variation of ball mass. An adaptive approximate feedback linearization to a ball and beam control problem. In:
input–output linearizing controller is designed for the BB Proceedings of the 1998 IEEE International Conference on Control
mechanism. The BB model is modified for control design pur- Applications, Vol. 1, pp. 694–698. Washington, DC: IEEE.
poses. Tracking performance of proposed control system is Chang YH, Chan WS and Chang CW (2013) TS fuzzy model-based
tested with computer simulations and experiments. Results adaptive dynamic surface control for ball and beam system. IEEE
reveal that the proposed adaptive control law reduces the Transactions on Industrial Electronics 60(6): 2251–2263.
Chhabra R and Emami MR (2016) A unified approach to input–
dependence of tracking error on ball mass variation. The pres-
output linearization and concurrent control of underactuated
ent study permits deduction of results about the performance
open-chain multi-body systems with holonomic and nonholo-
of the adaptive approach to approximate input–output linear- nomic constraints. Journal of Dynamical and Control Systems
ization when model parameters are unknown or time varying. 22(1): 129–168.
The case study of BB validates that the proposed control suc- Chen DY, Shi L, Chen HT, et al. (2012) Analysis and control of a
cessfully compensates for plant variations. hyperchaotic system with only one nonlinear term. Nonlinear
Dynamics 67(3): 1745–1752.
Chen D, Zhao W, Ma X, et al. (2013a) Control for a class of four-
Declaration of conflicting interest dimensional chaotic systems with random-varying parameters and
The authors declare that there is no conflict of interest. noise disturbance. Journal of Vibration and Control 19(7):
1080–1086.
Chen D, Zhao W, Sprott JC, et al. (2013b) Application of Takagi–
Funding Sugeno fuzzy model to a class of chaotic synchronization and
anti-synchronization. Nonlinear Dynamics 73(3): 1495–1505.
This research received no specific grant from any funding
Cheok KC and Loh NK (1987) A ball balancing demonstration of
agency in the public, commercial or not-for-profit sectors. optimal and disturbance-accommodating control. Control Systems
Magazine 7(1): 54–57.
Deng H, Li HX and Wu YH (2008) Feedback-linearization-based
References
neural adaptive control for unknown nonaffine nonlinear discrete-
Andreev F, Auckly D, Gosavi S, et al. (2002) Matching, linear sys- time systems. IEEE Transactions on Neural Networks 19(9):
tems, and the ball and beam. Automatica 38(12): 2147–2152. 1615–1625.
Deutscher J (2005) Input–output linearization of nonlinear systems Marino R (1997) Adaptive control of nonlinear systems: basic results
using multivariable Legendre polynomials. Automatica 41(2): and applications. Annual Reviews in Control 21: 55–66.
299–304. Marino R, Kanellakopoulos I and Kokotovic P (1989) Adaptive
Freidovich LB and Khalil HK (2008) Performance recovery of tracking for feedback linearizable SISO systems. In: Proceedings
feedback-linearization-based designs. IEEE Transactions on Auto- of the 28th IEEE Conference on Decision and Control, pp. 1002–
matic Control 53(10): 2324–2334. 1007. Washington, DC: IEEE.
Ghanadan R and Blankenship GL (1996) Adaptive control of non- Mohammed S, Poignet P, Fraisse P, et al. (2012) Toward lower limbs
linear systems via approximate linearization. IEEE Transactions movement restoration with input–output feedback linearization
on Automatic Control 41(4): 618–625. and model predictive control through functional electrical stimu-
Hauser J, Sastry S and Kokotovic P (1992) Nonlinear control via lation. Control Engineering Practice 20(2): 182–195.
approximate input–output linearization: the ball and beam exam- Nordin M and Gutman PO (2002) Controlling mechanical systems
ple. IEEE Transactions on Automatic Control 37(3): 392–398. with backlash – a survey. Automatica 38(10): 1633–1649.
Hirschorn RM (2002) Incremental sliding mode control of the ball Ogata K (2004) System Dynamics. Englewood Cliffs, NJ: Pearson/
and beam. IEEE Transactions on Automatic Control 47(10): Prentice Hall.
1696–1700. Raimúndez C, Villaverde AF and Barreiro A (2014) Adaptive track-
Ho D and Hedrick JK (2015) Control of nonlinear non-minimum ing in mobile robots with input–output linearization. Journal of
phase systems with input–output linearization. In: American Con- Dynamic Systems, Measurement, and Control 136(5): 054503.
trol Conference (ACC), pp. 4016–4023. Washington, DC: IEEE. Sahnehsaraei MA, Mahmoodabadi MJ and Bagheri A (2014) Pareto
Kabuli MG and Kosut RL (1992) Adaptive feedback linearization: optimum control of a 2-DOF inverted pendulum using approxi-
implementability and robustness. In: Proceedings of the 31st IEEE mate feedback linearization and sliding mode control. Transac-
Conference on Decision and Control, pp. 251–256. Washington, tions of the Institute of Measurement and Control 36(4): 496–505.
DC: IEEE. Sastry S and Bodson M (2011) Adaptive Control: Stability, Conver-
Kanellakopoulos I, Kokotovic PV and Morse S (1991) Systematic gence and Robustness. New York: Courier Corporation.
design of adaptive controllers for feedback linearizable systems. Sastry SS and Isidori A (1989) Adaptive control of linearizable sys-
IEEE Transactions on Automatic Control 36(11): 1241–1253. tems. IEEE Transactions on Automatic Control 34(11): 1123–1131.
Khalil HK and Grizzle JW (2002) Nonlinear Systems. Englewood Slotine JJE and Li W (1991) Applied Nonlinear Control, Vol. 199, No.
Cliffs, NJ: Prentice Hall. 1. Englewood Cliffs, NJ: Prentice-Hall.
Kim DH and Oh JH (1999) Tracking control of a two-wheeled mobile Sofianos NA and Boutalis YS (2016) Robust adaptive multiple mod-
robot using input–output linearization. Control Engineering Prac- els based fuzzy control of nonlinear systems. Neurocomputing 173:
tice 7(3): 369–373. 1733–1742.
Ko J, Strganac TW and Kurdila AJ (1999) Adaptive feedback lineari- Wu J and Li J (2016) Adaptive fuzzy control for perturbed strict-
zation for the control of a typical wing section with structural non- feedback nonlinear systems with predefined tracking accuracy.
linearity. Nonlinear Dynamics 18(3): 289–301. Nonlinear Dynamics 83(3): 1185–1197.
Kostarigka AK and Rovithakis GA (2012) Adaptive dynamic output Xiang F and Wikander J (2004) Block-oriented approximate feedback
feedback neural network control of uncertain MIMO nonlinear linearization for control of pneumatic actuator system. Control
systems with prescribed performance. IEEE Transactions on Engineering Practice 12(4): 387–399.
Neural Networks and Learning Systems 23(1): 138–149. Yao X, Tao G and Jiang B (2016) Adaptive actuator failure compen-
Krishna B, Gangopadhyay S and George J (2012) Design and simula- sation for multivariable feedback linearizable systems. Interna-
tion of gain scheduling PID controller for ball and beam system. tional Journal of Robust and Nonlinear Control 26(2): 252–285.
In: International Conference on Systems, Signal Processing and Yuan X, Chen Z, Yuan Y, et al. (2016) Sliding mode controller of
Electronics Engineering (ICSSEE’2012), Dubai, pp. 199–203. hydraulic generator regulating system based on the input/output
Kwon HY and Choi HL (2014) Gain scheduling control of nonlinear
feedback linearization method. Mathematics and Computers in
systems based on approximate input–output linearization. Inter-
Simulation 119: 18–34.
national Journal of Control, Automation and Systems 12(5):
Zhao ST, Gao XW and Che CJ (2015) Robust adaptive feedback lin-
1131–1137.
earization control for a class of MIMO uncertain nonlinear sys-
Leith DJ and Leithead WE (2001) Input–output linearisation of non-
tems. In: 27th Chinese Control and Decision Conference (CCDC),
linear systems with ill-defined relative degree: the ball and beam
pp. 1624–1629. Washington, DC: IEEE.
revisited. In: Proceedings of the American Control Conference, Vol.
4 pp. 2811–2816. Washington, DC: IEEE.