Documente Academic
Documente Profesional
Documente Cultură
Abstract: A nonlinear speed controller is designed using a fuzzy control strategy based on a Takagi-Sugeno(TS) modelling and
robust control approach for a mini unmanned helicopter 2-cylinder, two-stroke, electronic fuel injection engine. A physical
nonlinear model of the engine is presented and transformed into an uncertain fuzzy TS model to address the nonlinear system
characteristics. Then, a robust nonlinear fuzzy control algorithm is developed via Linear Matrix Inequality (LMI), which can
provide reference speed tracking despite the presence of torque load disturbances and operating condition changes. The simu-
lations show that the proposed controller can meet the performance requirements for the unmanned helicopter engine use,
namely tracking performance, disturbance rejection and robustness.
Key Words: Mini unmanned helicopter, Engine speed control, Takagi-sugeno models, Robust control, LMI
775
2 ENGINE MODEL DEVELOPMENT error, an integral structure is added according to internal
model theory.
Under the above considerations, the structure of the
2.1 Nonlinear Engine Model
augmented speed control loop is shown schematically in
The primary aim of this work is to keep engine speed Figure 1. The engine speed n is controlled by manipu-
constant against disturbances. The engine used in this lation of the commanded throttle angle θ cmd . The desired
study is a 2-cylinder, two-stroke, electronic fuel injection speed is denoted as nr . θ is the throttle-valve position.
piston engine with Electronic Control Unit (ECU). The
air-fuel ratio and the spark advance are maintained at The Te is the engine output torque. The principal distur-
their constant optimal values by appropriate separate bances to be considered in control design are due to col-
controllers. Thus the throttle which regulates the air flow lective pitch changes, flight status variations and ambient
into the engine is primarily intended for engine speed conditions changes. These effects cause a disturbance
control [8,9]. Changes in the air flow result in fueling ad- torque represented by Tl . Let x1 = θ , x2 = n and u = θ cmd .
justments to meet stoichiometric combustion conditions. The output of the integrator is considered as an extra state
This increase of fueling and air charge leads to a larger t
variable, i.e. x3 = ³ e( τ )dτ , e( t ) = nr ( t ) − n( t ) . K c is
torque production from the engine thereby compensating 0
for the additional torque load so as to keep engine speed servo compensator gain; K is stabilizer gain. The simplified
constant. Furthermore, in order to eliminate the static control-oriented plant model is presented as follows [10,11].
776
vicinity of the operating state x0 ; that is to find constant fuzzy set. Ai , Bi , C , D and Dr are constant real matrices.
matrices A and B such that in a neighborhood of x0 ΔAi and ΔBi represent the time-varying uncertain matri-
f (x ) + G (x )u ≈ Ax + Bu , ∀u (6) ces. The uncertainties are norm-bounded satisfying
and [ΔAi ΔBi ]=Di ⋅ Fi (t ) ⋅ [E1i E 2i ]
f (x0 ) + G (x0 )u = Ax0 + Bu , ∀u (7) where Di , E1i , E2 i are some given constant matrices and
Since u is arbitrary, B = G (x0 ) .Then A can be obtained Fi (t ) is an unknown real time-varying matrix function
by solving a convex constrained optimization problem, satisfying FiT (t ) ⋅ Fi (t ) İ I .
and the ith row aiT of the matrix A is computed as follows Using a singleton fuzzifier, product fuzzy inference and
f i (x0 )-x0T ∇f i (x0 ) weighted average defuzzifer, the final output of the fuzzy
ai = ∇f i (x0 )+ 2
⋅ x0 , x0 ≠ 0 (8) system is inferred as follows
x0
x (t ) = ¦ r hi (x (t ))[(Ai + ΔAi )x (t ) + (Bi + ΔBi )u(t )
where the ith component of f is f i :R → R ; ∇f i (x0 ):Rn n
°° i =1
® + Dw (t )+Dr yr (t )] (11)
→ R n is the gradient, a column vector, of fi evaluated ° y (t ) = Cx (t )
at x0 . °̄
777
and F T (t )F (t ) İ I , then Y +DF (t )E +E T F T (t )D T < 0 , if γ disturbance attenuation, if there are matrices X > 0 , Wi
and only if there exists a scalar ε > 0 satisfying and scalars ε ii > 0 , ε ij >0(i < j ) such that for all admissible
Y +ε DD +ε E E < 0
T −1 T
uncertainties the following two LMI conditions hold:
Theorem1 Given a real number γ > 0 , the uncertain
closed-loop fuzzy system(14) is robustly stable with
ª Ai X + BiWi + (Ai X + BiWi )T + ε ii Di Di T D XC T (E1i X + E2iWi )T º
« »
« DT -γ 2 I 0 0 » < 0 (1 İ i İ r ) (15)
« CX 0 -I 0 »
« »
¬« E1i X + E2iWi 0 0 −ε ii I ¼»
where Λij =(Ai X + BiW j )T + (Ai X + BiW j )+(A j X + B jWi )T + (A j X + B jWi )+ε ij (Di Di T + D j D j T ) .Furthermore, if matrices
exist which satisfies these inequalities, and then the feedback gains are given by K i = Wi X −1 . Proof See Appendix A.
ª 0 º ª0º
4 ENGINE SPEED CONTROL D = « −14.4686 » , Dr = «« 0 »»
« »
«¬ 0 »¼ «¬1 »¼
4.1 Engine Speed Control Design The matrices of uncertainties are defined using the mod-
In this section, the previous robust fuzzy control ap- eling error bounds. As a consequence, variations of about
proach is applied to engine speed control. In light of the 3% on Te (x1 ,x2 ) and about 5% on the throttle valve
real condition of the engine three operating points are model parameter τ thr are considered. Let Di = 0.1I 3 .
selected. The corresponding three plant rules are given as The application of theorem 1 to calculate the control
follows. Membership functions are chosen as trimf type. gains with γ = 1 obtains the following results
Rule 1 Ȁ1 = [-2.3391 -8.8146 0.1151]
IF x1 is about 20% and x2 is about 3300rpm
K 2 = [-4.6637 -16.5646 0.2152]
x = (A1 + ΔA1 )x + (B1 + ΔB1 )u + Dw + Dr yr
THEN ® K 3 = [-5.4305 -19.1732 0.2492]
¯ y = Cx
Rule 2 4.2 Nonlinear Simulation Validations
IF x1 is about 35% and x2 is about 4400rpm
x = (A2 + ΔA2 )x + (B2 + ΔB2 )u + Dw + Dr yr To evaluate the designed TS fuzzy controller, the com-
THEN ® puter simulation of the closed-loop system is performed
¯ y = Cx with the nonlinear dynamic engine model(4). In this
Rule 3 simulation, the engine speed is selected as output of the
IF x1 is about 50% and x2 is about 5500rpm plant and the external torque load input is selected as a
x = (A3 + ΔA3 )x + (B3 + ΔB3 )u + Dw + Dr yr disturbance to the plant. Two types of simulations have
THEN ® been conducted.
¯ y = Cx
The first group of the simulations is to investigate the
Where
speed regulation characteristic of the engine. The refer-
ª −33.3333 0 0º ª33.3333º ence engine speed is selected as a step increase from
A1 = « 7.8496 0.0267 0 » , B1 = B2 = B3 = «« 0 »»
« »
3300rpm to 3800rpm then to 4300 rpm and then a drop to
«¬ 0 −1 0 »¼ «¬ 0 »¼ 3500rpm. The corresponding closed-loop engine speed
and the throttle position responses are shown in Figure 2
ª −33.3333 0 0º
and Figure 3 respectively. It is observed that the engine
«
A2 = « 19.8004 −0.0721 0 »» speeds follow quite closely the desired reference speeds.
«¬ 0 −1 0 »¼ The second group of simulations is to investigate the
speed stability of the engine while engine torque load is
ª −33.3333 0 0º
suddenly changed. The loading torque is selected as a
A3 = «« 23.5328 −0.1128 0 »» , C = [ 0 1 0] step change from 14 Nm to 20Nm at t=200s, when the
«¬ 0 −1 0 »¼ speed is steady at 4300 rpm. The corresponding engine
speed and the throttle position responses are shown in
778
Figure 4 and Figure 5 respectively. As can be seen, the 5 CONCLUSION
maximal speed drop to the load disturbances is less than
15rpm. At steady state, the engine speed fluctuation is The purpose of this work is to develop a robust control
less than 5rpm. law for speed control of a two-stroke SI engine. A sim-
4400
output speed
plified control-oriented nonlinear engine model is de-
reference speed veloped and transformed into a TS model. Some fuzzy
4200
TS control tools are developed which concern the condi-
engine speed response (rpm)
4300
compensation of nonlinear systems by Takagi-Sugeno fuzzy
model[C]// Proc of the 4th IEEE Conference on Fuzzy Sys-
4295 tems. Yokohama, 1995: 531-538.
[7] BOYD S P, GHAOUI L E, FERON E. et al. Linear Matrix
4290
Inequalities in Systems and Control Theory [M]. Philadel-
phia: SIAM, 1994.
4285
[8] SOBOLAK S J. Simulation of the Ford vehicle speed control
system [J]. SAE: 820777.
4280
200 220 240 260 280 300 320 340 360 380 400
t(s)
[9] CAVINA N, MINELLI G, CAGGIANO M.et al. Model-based
idle speed control for a high performance engine [J]. SAE
Fig.4 Dynamic responses of engine speed
Technical Paper N.2003-01-0358.
under torque load disturbances at 4300rpm
30.5
[10] CHOI S B, HEDRICK J K. Robust throttle control of auto-
motive engines: theory and experiment [J]. ASME Journal of
30 Dynamic Systems, Measurement, and Control, 1996, 118:
92-97.
29.5
[11] COOK J A, POWELL B K. Modeling of an internal com-
throttle position (%)
779
x (t )T Px (t ) , where P > 0 ,and γ > 0 such that, for all t ªQij T P + PQij PD C T º
« »
V (x (t )) + y T (t )y (t )-γ 2 w T (t )w (t ) < 0 (17) where ȥij = « DT P −γ 2 I 0 » , so˄21˅is satisfied
for system (14) with yr (t ) = 0 .By integrating «¬ C 0 − I »¼
䫭䇃ʽᡒࠄᓩ⫼⑤DŽ from 0 to T, under the zero initial if
condition it leads to ȥii < 0 , i = 1, 2 ," ,r (24)
T
V (x (T )) + ³0 (y T (t )y (t )-γ 2 w T (t )w (t ))dt < 0 (18) ȥij +ȥ ji < 0, 1 İ i < j İ r (25)
Since V (x (T )) ı 0 , (18) implies y (t ) 2 İ γ w (t ) 2 . So the ª (Ai + Bi K j )T P + P (Ai + Bi K j ) PD CT º
« »
L2 gain of the fuzzy model is less than γ if (17) holds. Let Yij = « DT P −γ 2 I 0 »,
From(17), it is easy to obtain « C 0 − I »»
r r ¬« ¼
¦ ¦i =1
h h j [x T Qij T Px + x T PQij x + x T PDw
j =1 i inequality ˄24˅can be rewritten as
(19)
+ w T D T Px + x T C T Cx − γ 2 w T (t )w (t )]<0 ª PDi º
where Qij = Ai + Bi K j + Di Fi (t )(E1i + E2i K j ) , (19) can be Yii + «« 0 »» Fi (t ) [ E1i + E2i K i 0 0]
rewritten as «¬ 0 »¼
ªQij T P + PQij +C T C PD º ª x º T
˄26˅
¦ ¦
r r
h h [x T
w T
] « » « » <0 (20) ª PDi º
−γ 2 I ¼» ¬ w ¼
i =1 j =1 i j
DT P
¬« + [ E1i + E2i K i 0 0] Fi (t ) «« 0 »» < 0
T T
Pre-multiplying and post-multiplying ˄29˅ by diag { P −1 , I , I , I } let X = P −1 ˈ Wi = K i P −1 , which leads to inequality ˄15˅
in Theorem 1.
Similarly, inequality ˄16˅ in Theorem 1 is proved.
780