Proceedings of the 2011 International Conference on Machine Learning and Cybernetics, Guilin, 10-13 July, 2011

AIRCRAFT LANDING CONTROL IN WIND SHEAR CONDITION

CHIA-LIN LEE, JIH-GAU JUANG Department of Communications, Navigation and Control Engineering, National Taiwan Ocean University, Keelung, Taiwan E-MAIL: jgjuang@mail.ntou.edu.tw

Abstract:
Most aircraft accidents occurred during final approach or landing. This study proposes cerebellar model articulation controller (CMAC) to improve the performance of automatic landing system (ALS). The atmospheric disturbances affect not only flying qualities of an airplane but also flight safety. If the flight conditions are beyond the preset envelope, the automatic landing system (ALS) is disabled and the pilot takes over. An inexperienced pilot may not be able to guide the aircraft to a safe landing at the airport when wind shear is encountered. An adaptive type-2 fuzzy CMAC (FCMAC) is applied to PID control to construct intelligent landing system which can guide the aircraft to a safe landing in severe wind shear environment.

Keywords:
Automatic landing system; PID control; Fuzzy CMAC; Wind shear

1.

Introduction

Atmospheric disturbance can shake any airplane, no matter how big it is. One of the most dangerous disturbance is wind shear, a large change in wind speed or direction over a short distance. Wind shear is caused by microbursts, which are winds that blast down from showers or thunderstorms and have caused several airline crashes over the years. Wind shear occurs at all altitudes from the ground to the top of the atmosphere and it can be horizontal or vertical. Weather forecasts try to alert pilots to turbulence, but meteorologists often can give only very general forecasts. Pinning down exact locations of the worst turbulence is difficult. On March 1, 2008, at Hamburg airport, a Lufthansa Airbus A320 tried to land in crosswind conditions which exceeded the limit for the aircraft and made the left wing touch ground. The pilots then performed a go around and successfully saved the aircraft from crashing. According to a survey of the National Transportation Safety Board (NTSB) [1], 22.6 percent of aircraft accidents in the years of 1989 to 1999 were weather related. Most aircraft accidents occurred in final approach or landing. Another NTSB report, between 1994 and 2003, there were 19562 aircraft accidents. Weather was a contributing factor in 4159 of these accidents and involved 4167 aircraft.

Of the 4159 weather-related accidents, 2726 were due to wind conditions. In addition, a single accident may involve multiple weather conditions. According to the statistics of Flight International 10-16, January 2006 issue [2], there were 23 accidents/incidents affected by weather, causing total 324 (34 crew and 290 passengers) fatalities. The average accident fatality caused by weather is 14 people. It was apparent that most of cases were in the landing phase. Since 1965, most aircraft have had ALS installed. According to the Federal Aviation Administration (FAA) regulations [3], environmental conditions considered the determination of dispersion limits as being: headwinds up to 25 knots, tailwinds up to 10 knots, crosswinds up to 15 knots, moderate turbulence, and wind shear of 8 knots per 100 feet from 200 feet to touchdown. If the flight conditions are beyond the preset envelope, the ALS is disabled and the pilot takes over. An inexperienced pilot may not be able to guide the aircraft to a safe landing at the airport. The goal of this paper is to show that the proposed intelligent ALS can relieve human operators and guide the aircraft to a safe landing in a severe turbulence environment. Most of the improvements in the ALS system have been on the guidance instruments, such as GNSS Integrity Beacons, Global Positioning System, Microwave Landing System, and Automatic Land Position Sensor [4]-[7]. By using improvement calculation methods and high accuracy instruments, these systems provide more accurate flight data to the ALS to make the landing smoother. However, these researches do not include weather factors such as wind turbulences. Recently, intelligent concepts such as neural networks, fuzzy system, genetic algorithm, and hybrid systems have applied to flight control to increase the flight controllers adaptively to different environments [8]-[11]. This paper proposes an intelligent aircraft automatic landing control that uses type-2 FCMAC [12]-[14] to improve the performance of conventional ALS. Adaptive learning rates are obtained by the use of the Lyapunov stability theory. The convergence of the proposed control scheme is guaranteed. 2. System description At the aircraft landing phase, the pilot descends from the

978-1-4577-0308-9/11/$26.00 2011 IEEE 1180

Proceedings of the 2011 International Conference on Machine Learning and Cybernetics, Guilin, 10-13 July, 2011

cruise altitude to an altitude of approximately 1200 feet above the ground. The pilot then positions the aircraft so that the aircraft is on a heading towards the runway centerline. When the aircraft approaches the outer airport marker, which is about 4 nautical miles from the runway, the glide path signal is intercepted, as shown in Fig. 1. As the airplane descends along the glide path, its pitch, attitude, and speed must be controlled. The descent rate is about 10 ft/sec and the pitch angle is between -5 to +5 degrees. Finally, as the airplane descends 20 to 70 feet above the ground, the glide path control system is disengaged and a flare maneuver is executed. The vertical descent rate is decreased to 2ft/sec so that the landing gear may be able to dissipate the energy of the impact at landing. The pitch angle of the airplane is then adjusted, between 0 to 5 degrees for most aircraft, which allows a soft touchdown on the runway surface. A simplified model of a commercial aircraft that moves only in the longitudinal and vertical plane is used in the simulations for implementation ease [10].
Altitude

xb a+b where a = 300 b = 4300 , c = , h* = 1000 . x is the 2 horizontal position (ft) relative to ground of the aircraft. a, b, and c are various horizontal distances (ft) measured from the initial position. h is the vertical altitude (ft) of the aircraft. h* is the reference altitude (ft). k is the wind shear intensity (ft/sec). Fig. 2 shows a wind shear profile with a wind speed of 30 ft/sec.
0
Wind Shear velocity components: Horizontal (Solid) & Vertical (Dashed) 30

20

10 ft/sec

-10

-20

1200 ft ~ ~

glide path

-30 0 5 10 15 20 25 Time (sec.) 30 35 40

Figure 2. Wind shear profile

50 ft 0 ft touchdown flare path Runway position

3.

Controller design

Figure 1. Glide path and flare path

There are several wind shear models [15]. The most used one is given as follow h Wx = k A( x) Wh = k ( ) B( x) (1) h* The distribution of the horizontal wind versus the horizontal distance is given by
1 xa 1 + 2 ba

PID controllers are the most used controller in engineering industry. One of the applications is the aircraft landing system. Controller inputs consist of altitude and altitude rate commands along with aircraft altitude and altitude rate. The pitch command c is obtained from the PID controller. Then, the pitch autopilot is controlled by pitch command. In order to enable aircraft to land more steady when an aircraft arrives to the flare path, a constant pitch angle will be added to the controller. In general, the PID controller is simple and effective but there are some drawbacks such as apparent overshoot and sensitive to external noise and disturbance. When wind disturbance is encountered the PID controller may not be able to guide the aircraft to land safely. With CMAC compensator the proposed controller can overcome these disadvantages. It uses a traditional PID controller to stabilize the system and train the CMAC to provide precise control. The gains of PID controller are adjusted based on experiences, what it provides are tolerable solutions, not desired solutions. The CMAC can effectively meliorate these conditions. The overall control scheme is described in Fig. 3, in which the control signal U is the sum of the PID controller output and the type-2 fuzzy

xa
a< x<b
(2)

A( x) =

1 x>b and the distribution of the vertical wind versus the horizontal distance is given by

B( x) =

0 xa ca bx bc

xa axc
cxb

3)

1181

Proceedings of the 2011 International Conference on Machine Learning and Cybernetics, Guilin, 10-13 July, 2011

CMAC output. The inputs for the CMAC and PID controller are: altitude, altitude command, altitude rate, and altitude rate command. The PID controller provides tolerable solutions. In each time step k, the CMAC involves a recall process and a learning process. In the recall process, it uses the desired system output of the next time step and the actual system output as the address to generate the control signal U CMAC . In the learning process, the control signal of the pitch autopilot, U, is treated as a desired output. It is used to modify the weights of CMAC stored at location which is addressed by the actual system output and the system output of the next time step. The output of the CMAC is the compensation for pitch command. When the wind shear is too strong, the ALS can not control the aircraft to land safely. Here we use a type-2 fuzzy CMAC control scheme to improve the ability of wind shear resistance of the ALS. CMAC is a type of artificial neural network proposed in the literatures [16]. The CMAC represents one kind of associative memory technique. In the addressing technique, each input space (state variable) is quantized and the output space is divided into discrete states. A quantized input vector specifies a discrete state and is used to generate addresses for retrieving information from memory for this state.

CMAC could be computed as:

c j ( x) = c j1 ( x1 ) * c j2 ( x 2 ) * ......c jn ( xn ) =

c
i =1 n i =1

ji

(xi ) (x i )

(4) (5)

c j ( x) = c j1 ( x1 ) * c j2 ( x 2 ) * ......c jn ( x n ) =

ji

The type-reduced set of the type-2 fuzzy CMAC using the center of sets type reduction :
y cos = [ y l , y r ] = 1
w [ w1 , w 1 ] n

....

w N [ w M , w N ]

... 1

c [ c1 , c 1 ]

.....

c M [ c M , c N ]

c c
j j =1 j =1

(6)
j

wj

It is an interval type-1 set determined by its left and right end points yl and yr , which can be written as follows [13]:
yr =

c
j =1 N j =1

wj =
j

c
j =1

wj +
j

j = R +1

c c
N

wj
(7)

c
j =1

j = R +1

yl =

c
j =1 N j =1

w
j

c
j =1 L

w +
j j

j = L +1

c c
N

(8)
j

c
j =1

j = L +1

w and w are the corresponding weights of c and c , respectively. L and R can be obtained as follows: Step 1. Assume that the pre-computed w j are arranged in ascending order, i.e., w 1 w 2 ... w N

Step 2. Compute y r by initially setting for j = 1, 2,....N and let y r = y r

Figure 3. The type 2 CMAC control scheme

c j = (c j + c j ) / 2

Step 3. Find R(1 R N 1) such that w R y w R +1 r Step 4. Compute y r with c j = c j for j R and c j = c j for j > R and let y r = y r Step 5. If y r y r then go to step 6. If y r = y r then stop and set y r y r Step 6. Set y r = y r and return to Step 3.

The type-2 fuzzy theorem is utilized into CMAC structure in order to promote more accurate resolution than conventional fuzzy CMAC. The mapping procedure of type-2 fuzzy CMAC is similar as type-1 fuzzy CMAC. Each phase of mapping is described as follows. The X is an n-dimensional input space. Type-2 fuzzy CMAC uses the interval type-2 fuzzification method of the fuzzy theorem as its addressing scheme. After the input vector to the interval type-2 fuzzy set is being fuzzified, the input state values are transformed to upper firing strength and lower firing strength, which is based on corresponding interval type-2 membership functions. We choose the product inference method as the t-norm operator. The jth rules upper firing strength c j and lower firing strength firing strength c
j

The procedure for computing yl is very similar to the one just given for yr . In Step 3, find L(1 L N 1) such that w L y l w L +1 . Additionally, in Step 2 compute y l initially setting c j = (c j + c j ) / 2 for j = 1, 2,....N and in Step 4 compute y l with c j = c j for j L and c j = c j for j > L . The defuzzified output is simply the average of yr

in type-2 fuzzy

1182

Proceedings of the 2011 International Conference on Machine Learning and Cybernetics, Guilin, 10-13 July, 2011
V (t ) V (t ) e(t ) e(t ) (i ) = = e(t ) = wj w j (t ) e(t ) w j (t ) w j (t ) = ( y d y )c j ( x ) / c j ( x )
j =1 N

and yl as y = y r + yl (9) The work on learning of type-2 fuzzy CMAC is to update the memory weight according to the error between the desired output and the actual output. The learning rules for type-2 fuzzy CMAC is as follows: N (10) w j( i ) = w j( i 1 ) + 1 ( y d y ) c j ( x ) / c j ( x ) m j =1
wj = wj
(i) ( i 1 )

(19)

Thus
w j (t ) =

1
m

( y d y )c j ( x ) / c j ( x ), j = 1,2,....., N
j =1

(20) (21)

2
m

( y d y )c j ( x ) / c j ( x )
j =1

(11)

w j (t ) =

2
m

( y d y )c j ( x) / c j ( x ), j = 1,2,....., N
j =1

where 1 and 2 are the learning rates, m is the size of floor (called generalization) , y d is the desired output. The conventional fuzzy CMAC weight updating rule uses fixed learning rate, which might make the learning process updating too slow or too fast. In order to improve the shortcoming of conventional fuzzy CMAC on the learning process, an adaptive learning rate is introduced. Here, we use the discrete-time Lyapunov function to define the adaptive learning rate . Let the tracking error e(t) be e(t ) = y d y (t ) (12) where t is the time index. A discrete-type Lyapunov function can be expressed as 1 V = e 2 (t ) (13) 2 Thus, the change in the Lyapunov function is obtained by 1 V = V (t + 1) V (t ) = e 2 (t + 1) e 2 (t ) (14) 2 The error difference can be represented by [17] e(t ) (15) e(t ) W (t ) W Using gradient descent method, we have N w j (t ) = 1 ( y d y )c j ( x) / c j ( x ) (16) m j =1

N c 1 ( x) / c j ( x) j =1 w1 (t ) N w 2 (t ) 1 c 2 ( x ) / c j ( x) ( y y (t )) = W (t ) = j =1 # m d # N w N (t ) c N ( x ) / c j ( x ) j =1

(22)

1
m

(C ( x) / c j ( x))( y d y (t )
j =1

N c 1 ( x) / c j ( x) j =1 w1 (t ) N w (t ) 2 = 2 c 2 ( x) / c j ( x) ( y y (t )) W (t ) = j =1 # m d # N w N (t ) c N ( x ) / c j ( x ) j =1

(23)

2
m

(C ( x) / c j ( x))( y d y (t )
j =1

w j (t ) =

2
m

( y d y )c j ( x ) / c j ( x )
j =1

From (4), (5), (10), and (11) we have N e(t ) = c j ( x ) / c j ( x ) w j (t ) j =1

e(t ) W = C ( x) T / c j ( x)
j =1 N

(17)

(24)

Since V (t ) V (t ) e(t ) e(t ) = = e(t ) w j (t ) e(t ) w j (t ) w j (t )

= ( y d y )c j ( x ) / c j ( x )
j =1 N

N e(t ) = c j ( x ) / c j ( x ) w j (t ) j =1

(18)

N e(t ) = C ( x) T / c j ( x) W j =1

(25)

From (12) to (25), with respect to W , we have

1183

Proceedings of the 2011 International Conference on Machine Learning and Cybernetics, Guilin, 10-13 July, 2011

V =

1 2 1 e (t + 1) e 2 (t ) = [e(t + 1) e(t )][e(t + 1) + e(t )] 2 2 1 1 = [e(t )][2e(t ) + e(t )] = e(t ) e(t ) + e(t ) 2 2 e(t ) 1 = (C ( x) / W m

for t . The convergence of the adaptive type-2 fuzzy CMAC learning process is then guaranteed. The aircraft landing control system is locally asymptotically stable.
4. Simulations

c j ( x))e(t ) j =1
N

1 e(t ) 1 (C ( x) / e(t ) + 2 W m

c ( x))e(t )
j j =1

(26)

Since

e(t ) W

= C ( x) T / c j ( x) , we have
j =1

The aircraft starts the initial states of the ALS as follows: the flight height is 500 ft, the horizontal position before touching the ground is 9240 ft, the flight angle is -3 degrees, the speed of the aircraft is 234.7 ft/sec. Successful touchdown landing conditions are defined as follows [10]: 200 x (T ) ft/sec 270, -3 h (T ) ft/sec 0, -300 x (T ) ft 1000, -10 (T ) degree 5, where T is the time at touchdown, h(T ) is vertical speed of the aircraft at touchdown, x(T ) is the horizontal position at touchdown, x(T ) is the horizontal speed, (T ) is the pitch angle at touchdown. Maximal wind shear speeds from using conventional PID controller and CMAC with genaral basis function controller are 11ft/sec and 46 ft/sec, respecttively. The type-2 fuzzy CMAC PID controller can successfully guide an aircraft flying through wind shear up to 59 ft/sec. Figures 4 to 6 show the results by using the adaptive type-2 fuzzy CMAC PID controller at 59 ft/sec wind shear speed.
Pitch (Solid) & Pitch Command (Dashed) 2

N V = (C ( x)T / c j ( x)) 1 (C ( x) / c j ( x))e(t ) m j =1 j =1 N N N 1 e(t ) + (C ( x)T / c j ( x)) 1 (C ( x) / c j ( x))e(t ) m 2 j =1 j =1

2 T C ( x) 1 1 e(t ) C ( x) C ( x) 2e(t ) 1 e(t ) = N N 2 m m 2 2 ( c j ( x)) ( c j ( x )) j =1 j =1 2 2 C ( x) C ( x) 1 1 2 2 1 e (t ) N = 2 m m N 2 2 ( c j ( x)) ( c j ( x )) j =1 j =1

(27)

-2

-4 deg. -6 -8 -10 -12 0

2 C ( x) > 0 then V < 0 , i.e., we can Let 2 1 N m ( c j ( x )) 2 j =1 select the learning rate 1 in the following range

10

15

2m (

c
j =1

20 25 Time (sec.)

30

35

40

45

( x )) 2
2

C ( x)

> 1 > 0

Figure 4.

Aircraft pitch and pitch command

(28)
5. Conclusions

Similar to the learning rate 2 , with respect to W , we can obtain the following range 2 m(

c
j =1

( x ))
2

C ( x)

>2 > 0

(29)

V becomes negative definite. This implies that e(t ) 0

The purpose of this paper is to investigate the use of the intelligent controller, which includes the CMAC with general basis function and a type-2 fuzzy CMAC in the aircraft automatic landing control and to make the automatic landing system more intelligent. Stability and adaptive capability are proved by Lyapunov stability theory. The conventional PID

1184

Proceedings of the 2011 International Conference on Machine Learning and Cybernetics, Guilin, 10-13 July, 2011

controller can only overcome up to 11 ft/sec wind shear speed. The CMAC_GBF PID controller has better adaptive capability than conventional PID controller; it can tolerate the wind shear strength to 46 ft/sec. The adaptive type-2 fuzzy CMAC PID controller can successfully overcome 59 ft/sec wind shear speed and guide an aircraft to land safely.
Vertical Velocity (Solid) & Vertical Velocity Command (Dashed) 0 -2 -4 -6 ft./sec. -8 -10 -12 -14 -16 -18

10

15

20 25 Time (sec.)

30

35

40

45

Figure 5.
500 450 400 350 300 ft. 250 200 150 100 50 0 0 5 10

Aircraft vertical velocity and command

Altitude (Solid) & Altitude Command (Dashed)

15

20 25 Time (sec.)

30

35

40

45

Figure 6.

Aircraft altitude and command

References
[1] NASDAC Review of NTSB Weather-Related Accidents, http://www.nasdac.faa.gov/aviation_studies/weather_ study/studyindex.html, 2000. [2] Flight Safety Foundation-Taiwan, http://www. flightsafety.org.tw/news1.php?Code=1&main_id=3&pages =5, Flight International 10-16, Jan. 2006. [3] Federal Aviation Administration, Automatic Landing Systems, AC20-57A, 1997.

[4] C.E. Cohen et al., Automatic Landing of a 737 Using GNSS Integrity Beacons, Proc. ISPA, 1995. [5] DDC-I, Advanced Auto Landing System from Swiss Federal Aircraft Factory, Real-Time Journal, Sprint, 1995. [6] S. Asai et al., Development of Flight Control System for Automatic Landing Flight Experiment, Mitsubishi Heavy Industries Technical Review, vol. 34, no. 3, 1997. [7] D.N. Kaufmann and B.D. McNally, Flight Test Evaluation of the Stanford University and United Airlines Differential GPS Category III Auotmatic Landing System, NASA Technical Memorandum 110355, June 1995. [8] Y. Iiguni, H. Akiyoshi,, and N. Adachi, An Intelligent Landing System Based on Human Skill Model, IEEE Transactions on Aerospace and Electronic Systems, vol. 34, no. 3, pp. 877-882, 1998. [9] H. Izadi, M. Pakmehr, and N. Sadati, Optimal Neuro-Controller in Longitudinal Autolanding of a Commercial Jet Transport, Proc. IEEE International Conference on Control Applications, CD-000202, pp. 1-6, Istanbul, Turkey, 2003.. [10] C.C. Jorgensen and C. Schley, A Neural Network Baseline Problem for Control of Aircraft Flare and Touchdown, Neural Networks for Control, pp. 403-425, 1991. [11] J.G. Juang and K.C. Chin, Intelligent Landing Control Based on Neural-Fuzzy-GA Hybrid System, Proc. IEEE International Joint Conference on Neural Networks, vol. 3, pp. 1781-1786, 2004. [12] Z. Liu, Y. Zhang, and Y. Wang, A Type-2 Fuzzy Switching Control System for Biped Robots, IEEE Transactions on Systems, Man, and CyberneticsPart C, vol. 37, no. 6, pp. 1202-1213, 2007. [13] Q. Liang and J. Mendel, Interval Type-2 Fuzzy Logic Systems: Theory and Design, IEEE Transactions on Fuzzy Systems, vol. 8, no. 5, pp. 535550, 2000. [14] C.H. Wang, C.S. Cheng, and T.T. Lee, Dynamical Optimal Training for Interval Type-2 Fuzzy Neural Network, IEEE Transactions on Systems, Man, and CyberneticsPart C, vol. 34, no.3, pp. 1462-1477, 2004. [15] P. Berkmann and H.J. Pesch, Abort Landing in Windshear : Optimal Control Problem with Third-Order State Constraint and Varied Switching Structure, Journal of Optimization Theory and Applications, vol. 85, no. 1, pp. 21-57, April 1995. [16] J.S. Albus, A New Approach to Manipulator Control: the Cerebellar Model Articulation Control (CMAC), ASME Journal of Dynamic Systems, Measurement, and Control, vol.97, pp.220-227, Sep. 1975. [17] J.G. Juang, H.K. Chiou, and L.H. Chien, Analysis and Comparison of Aircraft Landing Control Using Recurrent Neural Networks and Genetic Algorithms Approaches, Neurocomputing, vol. 71, pp. 3224-3238, 2008

1185