Robustness Study of the Dynamic Inversion Based Indirect Adaptive Control of Flight Vehicles with Uncertain Model Data

Rama K. Yedavalli Praveen Shankar Professor Graduate Student Departemnt of Aerospace Engineering and Aviation The Ohio State University, 2036, Neil Avenue Columbus, OH-43201
Abstract The objective of this paper is to analyze the stability robustness of dynamic inversion based control laws being used for flight control with uncertainties in model data such as the aerodynamic stability and control derivatives. In particular, the paper is aimed at determining the robustness of an indirect adaptive control system developed by the researchers at the AFRL, which is built around a baseline dynamic inversion inner loop.. The controller, which is used to track the desired angular velocity commands, under perfect dynamic inversion, would result in a bank of integrators in the three angular velocity channels. While this control law is quite successful in tracking the desired angular velocity commands, no explicit parametric stability robustness margins are provided. A static approach to online system identification is used to estimate the vehicles changing control derivatives. Due to the existence of dynamic inversion errors resulting from uncertainties in these control derivatives, robustness of the designed control system needs to be guaranteed. In order to study the robustness of the controller, stability of the nominal system is guaranteed by the Input to State Stability criterion. The control derivatives are studied to determine the margins of perturbations under which the nonlinear stability margin between the input and the state of the rotational dynamics of the system are decreasing. Keywords: Feedback Linearization, Dynamic Inversion, Indirect Adaptive Control, Uncertainties, and Robustness.

David B. Doman Technical Lead Space Access and Hypersonic Vehicle Guidance and Control Team Air Force Research Laboratories
orbit on a reusable or expendable launch vehicle for onorbit missions, while reentering the atmosphere at hypersonic speeds and finally landing horizontally like an airplane. Such a vehicle would have a very large flight envelope that spans a very wide range of speeds and altitude. In the method proposed in [1], a nonlinear control law commands body-axis rotation rates. The magnitudes of the commanded body rates are determined by the magnitude of the rotation error. The commanded body rates form the input to a dynamic inversion-based adaptive/reconfigurable control law. The baseline control law is based on a full envelope design philosophy and eliminates trajectory dependent gain scheduling that is typically found on this type of vehicle. The indirect adaptive control portion of the control law uses on-line system identification to estimate the current control effectiveness matrix to update a control allocation module. Dynamic inversion control laws require the use of a control mixer or control surface allocation algorithm when the number of control surfaces exceeds the number of controlled variables. This is because a small number of desired moment or acceleration commands are calculated and a large number of control surfaces may be used to achieve the desired command. It is quite common that the desired commands can be achieved in many different ways and so control allocation approaches are used to provide consistent and unique solutions to such problems. The control allocation relies on accurate knowledge of the control derivatives. Under failure or damage conditions the control derivatives can be altered dramatically. Identifying control derivatives and supplying updated information to the control allocation block can improve the performance of the entire system. A fault-detection scheme is used to trigger an on-line system identification algorithm that is used the estimate control effectiveness when faults are detected. In reference [2], a Linear Quadratic Gaussian outer-loop controller is developed that improves the robustness of a Dynamic Inversion inner-loop controller in the presence of uncertainties. The selected dynamics are based on both performance and stability robustness requirements. The method however is based on the linearized equations of motion of the re-entry vehicle. An attempt is made in [3] to quantify the particular form of desired dynamics, which produce the best closed-loop performance and robustness in a Dynamic Inversion flight controller. Candidate forms of desired dynamics which invert the short period dynamics are evaluated. The controllers are synthesized for the prototype X-38 Crew Return Vehicle using a linear model at a selected point in the flight envelope. This paper

Nomenclature
BAE GB I Jx, Jy, Jz, Jxz L, M, N p, q, r P = = = = = = = = = Base Aerodynamics and Engine Control surface deflection vector Moment Vector Inertia Tensor Moments of Inertia Roll, pitch and yawing moments Roll, pitch and yaw velocities Parameter Vector Angular Velocities ([p; q; r])

1. Introduction
The reentry vehicle considered in this study is representative of the Space Maneuvering Vehicle in that it has a similar shape with four control surfaces: right/left tails and right/left flaps. Such a vehicle would be carried to

proposes a method that utilizes the ISS property in terms of nonlinear stability margins to determine the performance of the closed loop system under dynamic inversion so that an appropriate perturbation study can be carried out on the control derivatives. The emphasis is to maintain the nonlinear nature of the indirect adaptive control law proposed in [1], and not rely on linearization and the resulting gain scheduling procedures, which is done in the current literature as mentioned above. The proposed method attempts to study the robustness of the control system across the entire flight envelope and not just at a selected point. The rest of the paper is organized in the following way: Section 2 addresses the concept of input to state stability and global asymptotic stability of nonlinear systems. Section 3 explains the concept of dynamic inversion and Section 4 gives a brief overview of the dynamic inversion control law as applied to the re-entry vehicle. The stability of the closed loop system is discussed in Section 5 and Section 6 addresses the concept of nonlinear stability margin applied for robustness analysis to our system. Finally Section 7 analyzes the results for a particular case solved using the simulation and further research paths are discussed in Section 8.

GAS, ISS can be defined in terms of an ISS Lyapunov function. A nonlinear system with non-zero inputs given by

x = f ( x, u )
(V (0 ) = 0 and V ( x ) > 0 for x 0 ) and there exist 2 Class K functions and such that V ( x ) f ( x, u ) ( u ) ( x ) x n m holds for all x and u .
definite

(3)

is said to be ISS if there exists a storage function V(x) which is continuously differentiable, proper and positive

(4)

Another approach to defining ISS is motivated by the classical concept of total stability and the generalization of the gain margin for linear systems. A nonlinear stability margin for the above system is defined as any function with the following property: for each admissible possibly nonlinear and/or time varying feedback law k bounded by , that is so that

2. Preliminaries 2.1 Global Asymptotic Stability (GAS)

Given a nonlinear system

k (t , x ) ( x )
for all (t,x), the closed loop system

(5)

x = f (x)

x = f ( x, k (t , x ))
(1)

(6)

The GAS property amounts to the requirements that the system is complete and the following property holds: A storage or energy function

is GAS. We take advantage of this definition of ISS to conduct robustness analysis on the dynamic inversion control law developed for the rotational dynamics of the reentry vehicle.

V : n 0
for

is one

3. Dynamic Inversion
Dynamic Inversion is a design technique used to synthesize flight controllers whereby the set of existing or undesirable dynamics are cancelled out and replaced by a designer selected set of desired dynamics. Consider a nonlinear system of the form

which is continuously differentiable, radially unbounded) and positive definite. i.e.

V (0 ) = 0 and V ( x ) > 0

x 0.

The nonlinear system is said to be GAS if there exists some function ClassK function ( > 0 and is strictly increasing) such that

V ( x ) f ( x ) ( x ) x n holds for all x

2.2 Input to State Stability (ISS)

x= f x +g xu (7) where x is the state and u is the control. Then, we can design a control law given by

( )

( )

(2)

u = g 1 (x )( x des f ( x )) where x des is the designer

(8) selected set of desired

The ISS property provides a natural framework in which to formulate notions of stability with respect to the input. In this section, we review the various equivalent definitions of ISS as expressed in [7]. The objective is to express the fact that states remain bounded for bounded controls. Similar to

dynamics. Dynamic inversion is similar to model following control, in that both methodologies invert dynamical equations of the plant. Whereas model-following control specifies the desired plant behavior with an internal model to be followed, dynamic inversion specifies the desired plant behavior explicitly by specifying the rate of the desired control variable, not the control variable itself. This cancellation and replacement is accomplished by careful algebraic selection of the feedback function. It is for this

reason that this methodology is also called feedback linearization. It applies to both SISO and MIMO systems, provided that the control effectiveness function (in the SISO case) or the control influence matrix (in the MIMO case) is invertible. The method works for both full-state feedback (input-state feedback linearization), and output feedback (input-output feedback linearization).A fundamental assumption in this methodology is that the plant dynamics are perfectly modeled, and can therefore be canceled exactly. In practice this assumption is not realistic, and therefore the new dynamics require some form of robust controller to suppress undesired behavior due to plant uncertainties. In the next section we will introduce the dynamic inversion control law developed in [1] and discuss the need for robustness guarantees for the system. The emphasis in this paper is on the uncertainties arising in the control derivatives rather than the uncertainty in the model used for dynamic inversion.

2 c1 = (J y J z )J z J xz

c2

[ = [(J

Jy + Jz

] )J ]
xz

c3 = J z c 4 = J xz c5 = ( J z J x ) J y c6 = J xz J y c7 = 1 J y
2 c8 = (J x J y )J x + J xz

(12)

c9 = J x where
2 = J x J z J xz

4. Brief Overview of AFRLs Dynamic Inversion Control Law

A quaternion-based outer-loop control system designed for the descent phase of the re-entry vehicle generated bodyaxis angular velocity commands

4.2 On-line System Identification

A static approach to online system identification is used to estimate the vehicles changing control derivatives. The newly updated control derivatives are then used by the dynamic inversion control law to track the angular velocity guidance commands. Prior information about the system such as the prior knowledge of the effectiveness of the control surfaces can be included in the identification algorithm in the form of stochastic constraints for the better estimates of these values. Since the online estimates of the control derivatives are not exact, robustness margins for these parameters need to be studied. The rotational dynamics of the re-entry vehicle is given by

p des , q des , rdes that were

aligned with the error Euler-axis [1]. The inner-loop dynamic inversion control law was designed so that the vehicle tracked these body-rate commands. During the descent/landing flight phase, the reentry vehicle tracks the body rate commands generated by the guidance system.

4.1 Rotational Dynamics

p = c1 rq + c 2 pq + c3 L + c 4 N q = c5 pr c 6 p 2 r 2 + c7 M r = c8 pq c 2 rq + c 4 L + c9 N L = Ldelta + LBAE M = M delta + M BAE N = N delta + N BAE

Ldelta = L 1 1 + L 2 2 + L 3 3 + L 4 4 M delta = M 1 1 + M 2 2 + M 3 3 + M 4 4 N delta = N 1 1 + N 2 2 + N 3 3 + N 4 4
where (11) (10)

= f ( , P ) + g (P, )

(13)

(9)

where ( = [p q r]) and P denotes measurable or estimable quantities that influence the body- rate states. The parameter vector P includes variables such as Mach number, angle of attack, and sideslip angle and vehicle mass properties such as moments of inertia. Equation (13) expresses the body-axis rotational accelerations as a sum that includes control dependent accelerations g(P,) and accelerations that are due only to the base engine and aerodynamics.

= [ 1 2 3 4 ] =

Control

Surface

Deflections and

Li , M i , N i i = 1..4 are the Control Derivatives.

The angular accelerations of the vehicle are given by the following equation

= I 1 (G B I )

the base engine and aerodynamic moments. In this paper we intend to determine the robustness margins for parameter variations in the term

(14)

G (P ) .

where

G B = G BAE ( , P ) + G (P, )
L L M = + M N BAE N
(15)

Once the control law is developed we need to study the stability of the closed loop system before calculating the stability margins. The next section discusses the Lyapunov approach to determine the stability our closed loop system.

5. Stability of the Closed Loop System

The stability of the closed loop system is determined by considering a Lyapunov function is a proper and positive definite storage function. From the definition of global asymptotic stability, we require

GBAE(, P) is the moment generated by the base engineaerodynamic system and G is the sum of the moments produced by the control surfaces. Thus

V ( ) = 0.5( p 2 + q 2 + r 2 ) which

f ( , P ) = I 1 (G BAE ( , P ) I )

(16) (17)

and

g (P, ) = I 1G (P, )

V ( ) = pp + qq + rr < 0 3 holds for all .

6. Robustness Study for perturbations in the Control Derivatives

(22)

Dynamic inversion based linear control allocation methods require that the control dependent portion of the model be affine in the controls. We therefore develop a linear approximation of the control dependent part such that:

G (P, ) G (P )

(18)

The model used for the design of the dynamic inversion control law becomes

= f ( , P ) + I 1G (P )
and the control law provides direct control over

Having established a control law to stabilize the closed loop systems for some `nominal parameter values, it is of interest to determine the robustness of this control scheme under parameter perturbations. The most critical parameters in the above dynamic inversion based indirect adaptive control law are the control moment derivatives with respect to the control surface deflections. The nominal system is found to be globally asymptotically stable in the sense that V is found to be negative for all time. It is of interest to determine a function

(19)

( )

so that

= des i.e. des = f ( , P ) + I 1G (P )

Therefore, the inverse control law satisfies

(20)

des f ( , P ) = I 1G (P )

(21)

The aerodynamic database for the re-entry vehicle under consideration provides moment coefficient data that is taken at a moment reference point (MRP) that is located at the center of gravity of the empty vehicle (i.e. no fuel/oxidizer). Control derivative information is extracted from the tables in the database for Mach numbers, angles of attack and sideslip angles that were to be encountered on the trajectory. Since there are more control surfaces than controlled variables, a control allocation algorithm must be used to obtain a unique solution. There are four control surfaces that may be used on descent: Right and Left Tails and Right and Left Flaps. Equation (21) states that the control surfaces are to be used to correct for the difference between the desired accelerations and the accelerations due only to

K that would define the nonlinear stability margin of the system. The dynamic inversion control law depends on the online calculation of the control derivatives in order to compute the control variables (Control Moments L, M, and N). Since it is difficult to define an explicit function for the control in terms of the state of the system, we introduce a function that is defined as the ratio of the norm of the control to the norm of the state. We state that the control law is robust for all those values of the control derivatives such that the closed loop system is globally asymptotically stable and the function is decreasing. The robustness results developed in this paper are not around the individual values of the control derivatives. Instead perturbations on the norm of roll moment, pitch moment and yaw moment control derivatives were carried out i.e. the bounds around the norms of the following vectors were determined:
L 1 M 1 N1 L M 2 M = 2 N = N 2 L = L 3 M 3 N 3 L 4 M 4 N 4

7. Results
The simulations were carried out for different values of the desired rotational/angular velocities. The initial conditions were also varied during the simulation. Here we present the results obtained for an initial condition starting from the origin of the system. The desired angular velocities are given below

pdes = 0.03, qdes = 0.05, rdes = 0.01

Case 1 considers the nominal values of control derivatives that stabilize the system. In Case 2, the control derivatives considered outside the bounds of the robust design and hence we notice that the system is unstable. Figures 2 and 5 are the trajectories of and Figures 3 and 6 are the deflections of the control surfaces. Figures 4 and 7 show the variation of gain with respect to the norm of the state . Table 1 gives the bounds on the norms of the roll, pitch and yaw control derivatives for the above particular case. Case 1: Trajectory for Nominal Control Derivative Values (Stable) Case 2: Trajectory for Perturbed Control Derivatives (Unstable)

research is directed towards including all the translational and rotational motion state variables into the model to better analyze the robustness of the control system.

References
[1] David B. Doman and Anhtuan D. Ngo Dynamic Inversion Based Adaptive/Reconfigurable Control of the X33 on Ascent Journal of Guidance Control and Dynamics, 2002 [2] D. Ito, D. Ward and J. Valasek Robust Dynamic Inversion Controller Design and Analysis for the X- 38 AIAA paper 2001 4380, AIAA Guidance, Navigation, and Control Conference & Exhibit, 6-9 August 2001 [3] J. Georgie and J. Valasek Selection of Longitudinal Desired Dynamics for Dynamic Inversion Controlled ReEntry Vehicles AIAA paper 2001-4382, AIAA Guidance, Navigation, and Control Conference & Exhibit, 6-9 August 2001 [4] Zhihua Qu Robust Control of Nonlinear Uncertain Systems John Wiley & Sons, Inc.[1998] [5] D. Ito, D. Ward, J. Georgie and J. Valasek Reentry Vehicle Flight Controls Design Guidelines: Dynamic Inversion Final Technical Report, NAG9-1085 [6] Bernard Etkin Dynamics of atmospheric flight John Wiley & Sons, Inc. [1972] [7] Sontag E.D Remarks on Stabilization and Input to State Stability Proceedings of IEEE Conference on Decision and Control, Tampa, 1989 [8] Khalil H.K. Nonlinear Systems Prentice Hall 1996 [9] Bernard Etkin, Lloyd Duff Reid Dynamics of Flight Stability and Control, John Wiley and Sons, 3rd Ed. [10] Aircraft Control and Simulation Brian L.Stevens and Frank L.Lewis, John Wiley and Sons

The system under dynamic inversion was found to be stable for the range of values of the control derivatives given below. Table 1: Bounds on Control Derivatives Control Derivative Nominal Value 0.0038 0.0243 0.081 Minimum Value -5% -2% -12% Maximum Value +10% +20% +8%

L
M N

8. Conclusion
In this paper, a method is presented to determine the robustness of a baseline dynamic inversion control system developed for a representative space-maneuvering vehicle. The stability of the nominal system is determined by using the Lyapunov function approach for nonlinear systems. In order to study the robustness of the controller, the input to state stability criterion with respect to nonlinear gain margin is used in the analysis of the closed loop system under dynamic inversion so that appropriate perturbation study can be carried out. Further study is being carried out to determine the domain of attraction for the system. Future