You are on page 1of 6

Int. J. of Computers, Communications & Control, ISSN 1841-9836, E-ISSN 1841-9844 Vol. III (2008), Suppl.

issue: Proceedings of ICCCC 2008, pp. 337-342

Intelligent Autopilot Control Design for a 2-DOF Helicopter Model


Saeed Jafarzadeh, Rooholah Mirheidari, Mohammad Reza Jahed Motlagh, Mojtaba Barkhordari

Abstract: In this paper, we introduce a new intelligent control approach called Brain Emotional Learning Based Intelligent Controller (BELBIC). BELBIC is a controller based on emotional model of human brain which has been introduced not for so long. This controller has been applied to a nonlinear model of a helicopter. Feedback linearization method has also been applied to the system, and the performance of two controllers has been compared as an intelligent and a classical control method. An Input to State linearization method with some changes has been used to control the system. The performance of the controllers has been justied by the simulation. Keywords: BELBIC, Reward, Feedback Linearization, Input to State Linearization.

Introduction

Most of the existing results for helicopter control have been based on the linearization model or through several linearization techniques [1, 2]. The linearized models of the helicopter have some unmodeled dynamics that make the proposed controls unreliable when applied to original nonlinear models. Using nonlinear techniques gives better response and improves the performance of controlled system. A very thorough survey of linear techniques for helicopter control has been given by Garrad and Low [1]. Miniature helicopter control problems have also been discussed by Furuta et al. [3] and Kientz et al. [2]. The work of Pallett et al. [4] has served as the basis for our understanding of the helicopter model. Some other nonlinear control methods have been applied to the model which has been considered in this paper. Sliding mode Control [5] and robust control method [6] are some examples of these works. In this paper we apply two methods to the system and compare their performance. First method is a new intelligent approach called Brain Emotional Learning Based Intelligent Controller (BELBIC). This technique is based on physiological structure of brain in excited situations and can be used in control engineering problems. Recently, there is rising tend to intelligent controllers and BELBIC is not an exception [7, 8, 9]. This controller has a certain structure, but it can be changed to achieve the control objectives. There are both continuous and discrete time BELBIC controllers [9, 10]. We use the continuous one to compare with another continuous time control method (feedback linearization). To design the BEBIC controller, we should choose the appropriate reward function according to physical aspects of the control problem, and tune the training coefcients of the controller to achieve desired control objectives. Second method is feedback linearization which is a classical control method. An input to state linearization has been applied to the system for this method. There has been applied a feedback linearization method to a model of a helicopter [11], but it is different from the model which we use here. Therefore, we design a controller by feedback linearization method for the helicopter system. There is a complicated fth order nonlinear state space model for the system. The system has two inputs and two outputs. The control objective is to track the desired set points of the outputs. The sections of the paper are as follows. Section 2 describes the fth order model of the helicopter. Section 3 contains the theoretical results in BELBIC controller design and its tuning. Theoretical aspect and controller design for feedback linearization method has been brought in the section 4. Section 5 shows the simulation results and nally section 6 is conclusions.

Helicopter Model
Consider the nonlinear helicopter model as follows [6]:

Copyright 2006-2008 by CCC Publications - Agora University Ed. House. All rights reserved.

338

Saeed Jafarzadeh, Rooholah Mirheidari, Mohammad Reza Jahed Motlagh, Mojtaba Barkhordari

x1 x2 x3 x4 x5

= x2 2 2 = a0 + a1 x2 + a2 x2 + a3 + a4 x4 a5 + a6 x4 x3
2 = a7 + a8 x3 + (a9 sin x4 + a10 )x3 + u1 = x5 2 = a11 + a12 x4 + a13 x3 sin x4 + a14 x5 + u2

x = [h h ]T and u = [u1 u2 ]T = [k1 uth k2 u ]T In these equations h is the height of helicopter above ground and is measured in meter. is the rotational speed of the rotor blades and is measured in radian per second. is the collective pitch angle of rotor blades and is measured in radian. uth and u are the input to the throttle and the input to collective servomechanisms respectively.

3 Feedback Linearization
Feedback linearization has been used successfully to address some practical control problems. These include the control of helicopter, high performance aircraft, industrial robots, and biomedical devices [12]. The central idea of the approach is to algebraically transform a nonlinear system dynamics into a (fully or partly) linear one, so that linear control techniques can be applied. This differs entirely from conventional linearization in that feedback linearization is achieved by exact state transformation and feedback, rather than by linear approximations of the dynamics [12]. The idea of feedback linearization, i.e., of canceling the nonlinearities and imposing a desired linear dynamics, can be simply applied to a class of nonlinear systems described by the so-called companion form, or controllability canonical form. A system is said to be in companion form if its dynamics is represented by xn = f (x) + b(x)u where u is the scalar control input, x is the scalar output of interest, X = [x, x, ..., x(n1) ]T is the state vector, and f (x),b(x) are nonlinear functions of the states. This form is unique in the fact that, although derivatives of appear in this equation, no derivative of the input u is present. Note that, in the state space representation, the above equation can be written x1 x2 d ... ... (1) xn1 = xn dt xn f (x) + b(x)u From systems which can be expressed in the controllability canonical form, using the control input (assuming b to be non-zero) 1 u = [v f ] b We can cancel the nonlinearities and obtain the simple input-output relation (multiple integrator form) x(n) = v. Thus, the control law v = k0 x k1 x ...kn1 x(n1) with the chosen so that the polynomial pn +kn1 pn1 +...+ k0 has all its roots strictly in the left half complex plane, leads to the exponentially stable dynamics xn + kn1 xn1 + ... + k0 x = 0 which implies that x(t) 0. For tasks involving the tracking of a desired output xd (t), the control (n) law v = xd k0 e k1 e ... kn1 e(n1) (where e(t) = x(t) xd (t) is the tracking error) leads to exponentially convergent tracking. For the case of helicopter, we should nd a state transformation to rewrite the equations in the form of (1). We use the input to state linearization method [12]. There is a input to state linearization method for SISO systems in [12]. We apply this method with some changes to the helicopter model which is a MIMO system. Suppose that the vector z(x) = [z1 (x) ... zn (x)]T is the new state vector, called linearized state vector. To obtain this vector, we should make the following equations z1 ad if g = 0 i = 0, ..., n 2 and z1 ad n1 g = 0 f (2)

Intelligent Autopilot Control Design for a 2-DOF Helicopter Model

339

where is the Lie Bracket of f and g , and is a third vector eld dened by ad f g = g f f g . According to these equations, we can nd z1 (x). The linearized vector z(t) is z(x) = z1 L f z1 . . . Ln1 z1 f
T

(3)

where L f z1 is the Lie Derivative of f and z1 , and dened by L f z1 = z1 f . T The outputs of the system are x1 , x4 . We separate the states of the system into two sections ( x1 x2 x3 |x4 x5 ), and use the rst part to control the rst output by the rst control input, and the same for the second part. We start to nd z1 (x) for the rst set of states of the helicopter model. In this part of the model n = 3, so we have the following equations

z1 =0 zi

f or

i = 2, 3

z1 =0 zi

From these equations, we dene z1 = x1 . We use this method to linearize half of the system. It means that we linearized rst three equations of the model by the rst control input, and the last two equations by the second one. Thus, we obtain the states according to the (3) z2 Z3 = L f z1 = X2

2 2 = L2 z1 = a0 + a1 x2 + a2 x2 + (a3 + a4 x4 a5 + a6 x4 )x3 f

According to [12], we have u1 = (x) + (x)v,

(x) =

Ln z1 f Lg Ln1 z1 f

, (x) =

1 Lg Ln1 z1 f

For the rst three states, we have n = 3 . The rst control input has been chosen so that the characteristic equation of the rst output becomes (s + 30)(S + 3)2 = 0. For the last two states, we dene the second control input to cancel the nonlinearity of the fth equation. The characteristic equation of the second output is (s+1)2 = 0.

BELBIC Controller

Figure 1: Computational model of emotional learning in the Amygdala BELBIC is an abbreviation for Brain Emotional Learning Based Intelligent Controller. Motivated by the success in functional modeling of emotions in control engineering applications, a structural model based on the limbic system of mammalian brain, for decision making and control engineering applications has been developed. The computational model of emotional learning in the amygdala, based on Moren and Balkenius model, is depicted in Figure 2 [13, 14]. The main parts that are responsible for performing the learning algorithms are orbitofrontal cortex and amygdala. BELBIC controller has some sensory inputs. One of the designers tasks is to determine the sensory inputs. This controller has two states for each sensory input. One of these two is amygdalas output and another is the output of orbitofrontal cortex. Therefore, the number of sensory inputs has a key role in BELBIC controller. Usually the sensory inputs are rich signals [10]. Consider the i-th sensory input as si . Then we have amygdala and orbitofrontal cortex outputs Ai = si v i

340

Saeed Jafarzadeh, Rooholah Mirheidari, Mohammad Reza Jahed Motlagh, Mojtaba Barkhordari

Oi = si wi v, w are two states for the related sensory input. These states will be updated by the following equations vi = .si . max(0, rew Ai ), wi = .si .(rew Ai Oi max(Si ))

where and are training coefcients. We have a function named Reward. This function has a great role in BELBIC controllers. Reward is like its name. The controller strives to increase this reward. Therefore, the designer must dene a reward function that has its maximum values in the most desired regions. This reward function could be frequency domain function or a normal mathematic function. Amygdala acts as an actuator and orbitofrontal cortex acts as a preventer. Therefore the control effort of BELBIC controller is u = Ai Oi BELBIC is a controller that has only one output. Therefore for systems with more than one control inputs we must use one BELBIC controller for each control input. As it can be seen there are several tuning parameters for each sensory input. The general algorithm for tuning these parameters is trial and error. We use continuous form of BELBIC in this paper. In continuous form, the BELBIC states are updated with not a discrete relation but a continuous one. These continuous relations are vi = .si .(rew Ai ), wi = .si (rew + si + Oi Ai ) To control the helicopter in take-off and landing problem, we have two references to be tracked. One of them is desired height and the other one is the desired value of collective pitch angle of rotor blades. We must design a BELBIC controller for each control input of helicopter. For throttle we have one sensory input and it is heights error. s1 = hd h The reward function for this BELBIC controller is as gure 2.

Figure 2: Reward Function The reward functions parameters are positive real numbers. As it could be seen, the BELBIC controller receives the maximum reward when the sensory input is zero. According to the fact that the sensory input is an error signal, the BELBIC controller tries to vanish the sensory input and it means tracking. With this reward function we have reward and punishment together for BELBIC controller. There are some areas that reward takes negative values. It could help the controllers to sense whole domain of sensory inputs. The training coefcients and reward functions parameters for the rst BELBIC controller are as follows:

= 5, = 34, k1 = 1000, k2 = 100


The BELBIC controller for servo mechanical control has a sensory input and it is the error of collective pitch angle of rotor blades. si = d The BELBIC controller for the collective servomechanism control input is like the one for throttle control and with the same reward function but with different parameters. The training coefcients for the rst BELBIC controller are as follows

Intelligent Autopilot Control Design for a 2-DOF Helicopter Model

341

= 3, = 1.8, k1 = 1200, k2 = 80

Simulation Results

To see the performance of the two mentioned controllers, we have simulated the controlled system in Simulink. The height of the helicopter should track a desired path. The second output has a set point too. A sinusoidal path has been considered for the height, and second output kept constant. It can be seen from these simulations that the tracking performance of BELBIC controller for the height is better than Feedback linearization controller. The simulation results of controller system by BELBIC controller and feedback linearization controller have been demonstrated in the gures 3, and gure 4 demonstrates the control inputs of the system.

Figure 3: : Height (left) and Collective rotor blade angle (right) of helicopter (Solid: set point, Dashed: feedback linearization, Dotted: BELBIC)

Figure 4: First (left) and second (right) control input of helicopter (Solid: feedback linearization, Dotted: BELBIC)

Summary and Conclusions

In this paper a BELBIC controller and a feedback linearization technique applied to a nonlinear model of a helicopter to attain path tracking of a given path for the height of the helicopter. The system has ve states and two control inputs. A continuous time BELBIC controller has been applied to the system, and tuned for the best performance. An input to state linearization has been used to control the system too. In this method the states of the system have been separated into two parts, and each part has been controlled by one of the control inputs. The transient response of the BELBIC controller is better than the case with feedback linearization controller, but in the sense of steady state the performance of both controllers is good. However, there is an important disadvantage for this controller that is stability guarantee. The BELBIC is a very suitable method to solve control engineering problems, and it can be applied to a large variety of the linear and nonlinear systems.

342

Saeed Jafarzadeh, Rooholah Mirheidari, Mohammad Reza Jahed Motlagh, Mojtaba Barkhordari

References
[1] W. Garrad, And E. Low, Eigenspace design of helicopter ight control systems, Technical report, Dept. of Aerospace Engineering and Mechanics, University of Minnesota, Nov. 1990. [2] K. Kienitz, Q. Wu, And M. Mansour, Robust stabilization of a helicopter model. Proceedings of the 9th CDC, pp. 2607-2612, 1990. [3] K. Furuta, Y. Ohyama, And Yamanao, Dynamic of RC helicopter and control, in Mathematics and computers simulation XXVI(North Holland, Amsterdam, 1984), pp. 148-159. [4] Tj. PalleIt, B.J. Wolfert And S. Ahmad, Real time helicopter ight control test bed. Technical Report TR-EE 91-28, School of Electrical Engineering, Purdue University, 1991. [5] H. Sira-Ramirez, M. Zribi, S. Ahmad, Dynamical sliding mode control approach for vertical ight regulation in helicopters, IEE Proc.-Control Theory Appl., Vol. 141, No. I, January 1994. [6] J. Kaloust, C. Ham, Z. Qu, Nonlinear autopilot control design for a 2-DOF helicopter model, IEE Proc.Control Theory Appl., Vol. 144, No. 6, November 1997. [7] Sharba A. Maziar, Lucas Caro, Mohammadinejad Aida, Yaghobi Mostafa, Designing a Football Team of Robots from Beginning to End, International journal of Information Technology, Vol 3, No. 2, 2006. [8] C. Lucas, S. Moghimi, Applying BELBIC (Brain Emotion Learning Based Intelligent Controller) to an Auto landing System, Conference: WSEAS AIKED03, 2003. [9] Tutunchi Ali Ghasem, Optimal BELBIC control with application to Autopilot control, Masters Thesis, University of Tehran, 2006. [10] C. Lucas, D. Shahmirzadi, N. Sheikholeslami, Introducing BELBIC: Brain Emotional Learning Based Intelligent Controller, International Journal of Intelligent Automation and Soft Computing, Vol. 10, No. 1, pp. 11-22, 2004. [11] G. Meyer, R.L. Hunt And R. Su, Design of a helicopter autopilot by means of linearizing transformations, Guidance and Control Panel, 35th Symposium, AGARD CP321, Paper 4, 1983. [12] Jean-Jacques E. 12, Weiping Li, Applied Nonlinear Control, Prentice Hall, 1991. [13] J. Moren, C. Balkenius, A Computational Model of Emotional Conditioning in the Brain, in Proc. workshop on Grounding Emotions in Adaptive Systems, Zurich, 1998. [14] J. Moren, C. Balkenius, A Computational Model of Emotional Learning in The Amygdala: From animals to animals, in Proc. 6th International conference on the simulation of adaptive behavior, Cambridge, Mass., The MIT Press, 2000. Saeed Jafarzadeh, Rooholah Mirheidari, Mojtaba Barkhordari Iran University of Science and Technology Department of Electrical Engineering Address: Daneshgah Street, Hengam Street, Resalat Sq., Tehran, Iran E-mail: sjafarzadeh, rmirheidari@ee.iust.ac.ir, mbarkhordary@iust.ac.ir Mohammad Reza Jahed Motlagh Iran University of Science and Technology Department of Computer Engineering Address: Daneshgah Street, Hengam Street, Resalat Sq., Tehran, Iran E-mail: jahedmr@iust.ac.ir