AIAA 2015-0608

AIAA SciTech
5-9 January 2015, Kissimmee, Florida
AIAA Guidance, Navigation, and Control Conference

Application of a Novel Scalability Notion in Adaptive
Control to Various Adaptive Control Frameworks∗

Simon P. Schatz∗†, Tansel Yucelen†‡, Benjamin Gruenwald†§, and Florian Holzapfel‡¶

In adaptive control theory it is a well-known phenomena that nonidentical command
profiles entail nonidentical closed-loop responses of these adaptive systems. While adaptive
controllers provide a viable methodology to control uncertain dynamical systems, this lack
Downloaded by CARLETON UNIVERSITY LIBRARY on July 31, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.2015-0608

of predictability is a significant disadvantage, in particular in terms of certification of such
control methods. Consequently, achieving predictable closed-loop responses of adaptively-
controlled systems is of grand practical interest. For this purpose, we recently introduced a
method1 to scale the learning rates of the adaptive weight update laws in order to achieve
predictable closed-loop performances for nonidentical, but scalable command profiles. This
paper applies the proposed methodology to a model of the longitudinal motion of a Boeing
747 aircraft and simulations for diverse adaptive control schemes illustrate the efficiacy
of the proposed scalability notion, which may be a further step towards validation and
verification of these adaptive control frameworks.

I. Introduction
In this paper, direct model reference adaptive control (MRAC)2, 3 is considered. Adaptive controllers
require less modeling information in comparison to fixed-gain controllers as the controller gains are tuned
online driven by the tracking error between the system’s output (respectively, state) and the reference model’s
output (respectively, state). Hence, controllers employing adaptive control laws have the ability to deal with
uncertainties e.g. resulting of unknown nonlinearities or imprecisely modeled system parameters.
Although adaptive controllers show good results for well-tuned cases, it is well-known in the adaptive
control community that adaptive controllers are sensitive to the level of excitation, particularly in the tran-
sient phase when the adaptive control method “learns” the uncertainty. On the other hand, employing high
learning rates for the adaptive weight update laws in order to increase the adaptation’s speed in the transient
phase may result in unacceptable control input signals due to high-frequency content in the control channel.4
Nowadays, a lot of research is conducted towards these problems and modifications such as pseudo control
hedging,5 low-frequency learning adaptive control,6 and L1 adaptive control7 have been introduced in order
to employ high learning rates. Frameworks to achieve improvement of the transient performance of adaptive
controllers were further introduced, namely among others closed-loop reference models8, 9 and the command
governor adaptive control framework.10
In addition, high excitation of the regressor vectors in the adaptive weight update law have a similar effect
as high learning rates, which results of the nature of the adaptive weight update laws. As a consequence,
tuning the adaptive controller for good transient behavior over the whole envelope of permissible system states
is a challenging task. Although some adaptive control architectures show a certain level of predictability and
are approximately scalable, for example L1 adaptive control7 and the command governor framework,10 the
schemes introduced above are all related to improving the performance of adaptive controllers, particularly
∗ This research was supported in part by the University of Missouri Research Board.
† Graduate Research Assistant, Institute of Flight System Dynamics, Technische Universit¨ at M¨unchen, 85748 Garching,
Germany, simon.p.schatz@tum.de
‡ Assistant Professor, Missouri University of Science and Technology, Department of Mechanical and Aerospace Engineering,

Rolla, MO 65409, USA, tyucelen@mst.edu
§ Graduate Research Assistant, Missouri University of Science and Technology, Department of Mechanical and Aerospace

Engineering, Rolla, MO 65409, USA, bg8t2@mst.edu
¶ Professor, Institute of Flight System Dynamics, Technische Universit¨ at M¨unchen, 85748 Garching, Germany, flo-
rian.holzapfel@tum.de

1 of 17

American Institute of Aeronautics and Astronautics
Copyright © 2015 by Simon P. Schatz, Tansel Yucelen, Benjamin Gruenwald, Florian Holzapfel. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.

B) is controllable and the uncertainty is parameterized as h i ∆(x(t)) = Λ WxT WcT wκ ω(x(t). c(t)) ∈ <n+l+1 is a known regressor vector. but scalable command profiles for a class of direct model reference adaptive controllers with uncertainties parameterized using linear regressor vectors. given by   x(t) ω(x(t).2015-0608 II.10 This simulation study analyzes the numerical results in detail and shows the efficiacy of the proposed scalability concept.6 adaptive control architectures employing closed-loop reference models.12 low-frequency learning adaptive control. 1.8. Ad- ditionally. in the transient phase. Λ ∈ <m×m + is an unknown control effectiveness matrix. Although the formulation of the uncertainty in (2) represents a class of linear uncertainties. Finally. From the theoretical point of view. (2) where Wx ∈ <n×m denotes an uncertainty in the system matrix. in Ref. ∆(x(t)) : <n → <m is an uncertainty. and B ∈ <n×m is a known control input matrix.6 and the command governor adaptive control framework. The nomenclature used for this paper is fairly standard and according to the nomenclature used in previous publications. which shows scalable performance with respect to nonidentical. 12 adaptive control architectures with low-frequency learning. e. u(t) ∈ <m is the control input vector. emphasizing the approximatly scalable nature of this framework. and wκ ∈ <m denotes a constant disturbance. In comparison to previous work. specifically σ-modification and e-modification adaptive control architectures.aiaa.11. an adaptive control scheme using artificial basis functions13 is shown to be scalable in the sense presented in this paper. A. A ∈ <n×n is a known system matrix. the problem formulation utilized for classical MRAC schemes is presented and the scalability notion for MRAC is introduced.8. (4) 2 of 17 American Institute of Aeronautics and Astronautics . Wc .g.13 This method is also illustrated in simulations and the effect of the scalability notion is analyzed. This scalability concept relies on scaling the learning rates relating to the command coefficient of these command profiles. c(t)). Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. (3)   κ where c(t) ∈ <l is the uniformly continuous bounded command and κ ∈ < is a constant. this command governor example is evaluated in more detail. the application of the scalability notion on different MRAC schemes. The contribution of this paper is to execute an illustrative simulation study using the previous theoretical results1 and a model of the lon- gitudinal motion of a Boeing 747 aircraft. xr (0) = xr0 . Scalability for Model Reference Adaptive Control This section overviews our previous results on scalability in Ref. but do not directly address predictability of the system response over the envelope of permissible commands. This simulation study incorporates adaptive control architectures using reference model modifications. 2015 | http://arc. the scalability notion is further shown for an adaptive control framework employing artificial basis functions. We recently introduced a scalability notion in Ref. (1) where x(t) ∈ <n is the accessible state vector. x(0) = x0 .org | DOI: 10.2514/6. and wκ are considered time-invariant. It is assumed that the pair (A. ω(x(t). 1 and hence. 9 adaptive controllers with e-modification.10 is overviewed shortly in order to elucidate the previous results before using these adaptive control frameworks for illustrative simulations displayed in Section III. the overall system including the adaptive control scheme is inherently nonlinear. The uncertain dynamical system specified in equation (1) is desired to track the reference system given by x˙ r (t) = Ar xr (t) + Br c(t). Note that Wx . Wc ∈ <l×m denotes an uncertainty in the command input matrix. c(t)) =  c(t)  . MRAC Problem Formulation Consider the uncertain dynamical system given by x(t) ˙ = Ax(t) + BΛu(t) + B∆(x(t)). 9 and command governor- based adaptive controllers. Additionally. 10.

org | DOI: 10. c(t)). 3 B.2. Now. W ˆ (t) − W ∈ <(n+l+1)×m is the adaptive weight estimation error. Now. 2015 | http://arc. (5). Scalability In this section. Im − Λ−1 . (11) and P ∈ <n×n is the positive definite solution of the Lyapunov equation Q + AT r P + P Ar = 0. Stability of direct model reference adaptive controllers as introduced in this section can be shown according to the literature. This can be shown in a straight forward manner by transformation1 of the system dynamics (13). but scalable command profiles c(t). the uncertain dynamical system is given by = Ar x(t) + Br c(t) + BΛW T ω(x(t). c(t)). uad (t) = W (9) ˆ (t) ∈ <(n+l+1)×m is the adaptive weight matrix satisfying the adaptive weight update law where W ˆ˙ (t) W = Γω(x(t). Using (2). c(t)) − BΛuad (t). = Ar x(t) + Br c(t) − BΛW x(0) = x0 . (7) can now be given as x(t) ˙ ˜ T (t)ω(x(t). x(t) − xr (t). the following definitions are employed: 3 of 17 American Institute of Aeronautics and Astronautics . The reference system (4) is desired to be tracked by the uncertain dynamical system (1) using the control law u(t) = unom (t) − uad (t).aiaa. (12) where Q ∈ <n×n is a positive definite design matrix. Hence. the adaptive control law is chosen as   ˆ T (t)ω(x(t). where xr (t) ∈ <n is the reference model state vector. Ar ∈ <n×n is the desired Hurwitz system matrix. and Br ∈ <n×l is the command input matrix. (10) which is driven by the tracking error e(t) ∈ <n defined as e(t) . Hence.2015-0608 x(t) ˙ (7) where h iT W .2514/6. the uncertain dynamical system (1). the idea of the scalability notion is to scale the learning rate Γ corresponding to the specified command profile. (6) where Kx ∈ <m×n is the nominal feedback matrix and Kc ∈ <m×l is the nominal feedforward matrix chosen such that A − BKx = Ar and BKc = Br . Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. uad (t) ∈ <m is the adaptive control input and unom (t) ∈ <m denotes the nominal control input given by unom (t) = −Kx x(t) + Kc c(t). and (6) in (1). (5) where u(t) ∈ <m is the control input. ˆ (0) = W W ˆ 0. In order to execute this transformation. For the idea of scalability we assume that a control engineer has designed a positive definite matrix Q for the Lyapunov equation (12) and a learning rate Γ0 yielding appropriate performance of the adaptive control system for a specified command history c0 (t). WxT − Λ∗ Kx WcT + Λ∗ Kc wκ . The objective of the scalability notion is to achieve scalable performance for nonidentical. (13) where W ˜ (t) . (8) and Λ∗ . c(t))eT (t)P B. the reference system (4) and the weight update law (10). the concept of scalability1 is introduced. for any scaled command profiles c(t) = αc0 (t) with scalar scaling command coefficients α 6= 0 given a Lyapunov design matrix Q it is possible to achieve scaled system responses with respect to this well-tuned set-up by choosing Γ = Γ0 /α2 .

4 on most adaptive control frameworks as excessively large learning rates Γ. C. the application of the scalability concept to the σ-modification and e-modification adaptive control architectures11. xr0 /α ez (t) . consider the adaptive weight update law for the original case (10) and for the transformed system (18). W (19) z Consequently. namely undesirable oscillations. the transformed reference system dynamics and the adaptive weight update law are given by z(t) ˙ = ˜ T ωz (x(t). 1. Scalability of other MRAC schemes In this section.aiaa. 9 and command governor-based adaptive controllers10 are modified to achieve predictable performances as shown previously. Hence.2015-0608 1 Using (14) and (15) the transformed system dynamics. ˆ (0) = W W ˆ 0. xr (t)/α (14) zr0 . Finally. This deviates from a traditional adaptive control architecture since large regressor vectors ω(x(t).g. x(t)/α z0 . (18) Note that the equations (16). c(t))eT z (t)P B. and (18) hold for any α 6= 0. (15) and the scaled learning rate Γ = Γ0 /α2 . yielding a certain level of predictability. (17) Wˆ˙ (t) = Γ0 ωz (x(t). (19) shows that using Γ = Γ0 /α2 renders invariance of the adaptive weight update law with respect to the scaling parameter α. Using (14). z(t) . Ar z(t) + Br c0 (t) − BΛW z(0) = z0 . which is required for a further step towards validation and verification of adaptive controllers. c(t))eT (t)P B = Γ0 ωz (x(t).17) are scalable in the sense that state histories can be given by a nominal system response scaled by α. Further. c(t)) have the same negative effects. Thus. c(t)). note that the uncertain system (13.org | DOI: 10. 2015 | http://arc. c(t))eT (t)P B. 4 of 17 American Institute of Aeronautics and Astronautics . c(t)) = ω(x(t). and e modification architectures The σ-modification11 was introduced with the claim that it prevented the estimated adaptive weights from becoming unbounded. e(t)/α κ= α Hence. (15)   Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. for uncertain dynamical systems of the form (13) scalable performance can be achieved for any α 6= 0 and large commands c(t) do not cause high excitation of the adaptive weights. it can be stated ˆ˙ (t) = Γω(x(t).   z(t) ωz (x(t). In order to achieve this increased robustness with respect to e. (17). unmodeled dynamics. the adaptive weight responses are identical given any α. σ. 12 is shown. In particular.16) and the reference system (4. x0 /α zr (t) . 1. zr (0) = zr0 . adaptive control architectures with low-frequency learning6 . (16) z˙r (t) = Ar z(t) + Br c0 (t). adap- tive control architectures employing closed-loop reference models. In terms of brevity of this paper. these modifications are introduced in short and for more detailed information the reader may be referred to Ref. c(t))/α = c0 (t) . which is obvious from (10).2514/6. Furthermore.8. the application of the scalability notion to extensions of the MRAC framework is presented.

ˆ f (0) = W W ˆ 0. as discussed in the previous section. satisfying ˆ˙ f (t) = Γf [W W ˆ (t) − W ˆ f (t)]. (23) where Γf ∈ <(n+l+1)×(n+l+1) is a positive definite filter gain matrix such that λmax (Γf ) ≤ γf. the uncertain dynamical system (13) and the adaptive weight update law (10) remain unchanged and are scalable as presented in Section B. The adaptive weight update law is given by ˆ˙ (t) = Γω(x(t). control law and reference model.1 Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. remain unchanged.1 4. the adaptive weight update law was modified as ˆ˙ (t) = Γω(x(t). c(t))eT (t)P B − σ[W W ˆ (t) − W ˆ f (t)]. Reference Model Modification An approach to improve the transient performance of MRAC controllers by feeding back the tracking error into the reference model is the so-called reference model modification8. Incorporating the scaling factor into the learning rate as Γ = Γ0 /α2 yields once again invariance of the adapive weights with respect to the command coefficient α. Command Governor Adaptive Control Finally. (20) where σ > 0 is a damping coefficient used to ”pull” the estimated adaptive weights towards the origin. ˆ (0) = W W ˆ 0. 5 of 17 American Institute of Aeronautics and Astronautics . Hence the effect of the damping in the so-called e-modification12 was proportional to the tracking error of the system. It is claimed that the modification term filters high-frequency content out of the adaptive weight update law. invariance of the adaptive weight response with respect to the command coefficient α can be achieved. By scaling the learning rate Γ = Γ0 /α2 and the damping coefficient σe = σ0 /α. the reference model is given by x˙ r (t) = Ar xr (t) + Br c(t) + Le(t). 9 . Low-Frequency Learning The low-frequency learning adaptive control architecture employs a gradient based modification term and a low pass filter6 . it can be concluded that a system employing this adaptive control framework will have predictably scalable responses. Employing the relations (14) and scaling the command with the command coefficient c(t) = αc0 (t) scalability for adaptive control architectures with modified reference models is obtained. 2. As a consequence.max and γf. xr (0) = xr0 .1 The standard MRAC adaptive law was further modified by introducing a time-varying damping coefficient σe ke(t)k2 instead of the constant σ in (20). the uncertain dynamical system’s response is scalable as introduced in Section B. allowing for the controller to be tuned with high learning rates in order to enable robust and fast adaptation.2514/6. (24) where L ∈ <n×n is a positive definite matrix. However.1 Therefore. By introducing the scaling factor for the learning rate Γ = Γ0 /α2 as introduced in Section B. c(t))eT (t)P B − σ W W ˆ (t). Basi- cally. c(t))eT (t)P B − σe ke(t)k W W ˆ (t).org | DOI: 10. it can be shown that the adaptive weight response is invariant to the scaling factor α.2015-0608 Since the adaptive weight update laws for these two modifications are invariant to the scaling factor α and both. (22) ˆ f (t) ∈ <(n+l+1)×m is the low-pass filtered weight estimate of W where σ > 0 is a modification gain and W ˆ (t). resulting in the adaptive weight update law given by ˆ˙ (t) = Γω(x(t). the scalability notion is applied to the command governor framework for adaptive control10 . ˆ (0) = W W ˆ 0. an additional command was introduced in order to cancel uncertainties in the transient phase. (21) 2 where σe > 0. 3.max > 0 is a design parameter. 2015 | http://arc. ˆ (0) = W W ˆ 0.aiaa.

1 D. c(t)) + W ˜ T (t)ωa (t) . Kc−1 B T B  B g(t) ∈ <m . c(t))ωaT (t)W   W ˆ a (t) . 14 for more details). c(t)) + W uad (t) = W ˆ aT (t)ωa (t).2514/6. (27). desired tracking command (the original c(t) from the sections above). ˆ a (0) = W W ˆ a0 . which is an artificial and thus. the error dynamics of the system do not change and therefore. 14 It is claimed that the transient performance is improved by introducing an artificial regressor vector ωa (t) ∈ <q . Then the artificial weight estimation error is given by W ˜ a (t) . g(t) ∈ <n is the command governor output. c(t))+ W ˜ aT (t)ωa (t) is approximately suppressed by gradient minimization. Application of Scalability Notion using a Framework with Artificial Basis Functions In this section. Using cD (t) = αc0 (t). Since the additional command is applied on both reference model and nominal controller. x(0) = x0 . (28) Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. cD (t) + cg (t). resulting in the artificial basis function update law given by ˆ aT (t)[B T B]−1 B T {e(t) ω˙ a (t) = ka W ˙ − Ar e(t)}. ξ(0) = 0. the system dynamics can be computed as   x(t) ˙ = Ar x(t) + Br c(t) − BΛ W˜ T (t)ω(x(t). ωa (0) = ωa0 . c(t)) = [B T B]−1 B T {Ar e(t) − e(t)}. (26). Inserting (32) into the uncertain system given by (7). W ˆ a (t) − Wa = ˆ Wa (t). ˙ = Ar e(t) − BΛW e(0) = x0 − xr0 . which is defined as ˙ ξ(t) = −λξ(t) + λe(t). The interested reader may be referred to previous work10 for additional information about the command governor. (25) where cD (t) ∈ <m is the bounded. 14 ˆ˙ (t) = Γ ω(x(t). the scalability notion is further extended to an adaptive control framework using artificial basis functions. ˙ x(0) = x0 . The adaptive weight update laws are given by13. (35) ˆ˙ a (t) W = Γa ωa (t)eT (t)P B. (34) where ka ∈ <+ is a design parameter and ωa0 6= 0.2015-0608 which can be written as ˜ T (t)ω(x(t). By applying integration of parts.10 It can be shown that λξ(t) − λe(t) − e(t) Note that although the reference model is modified. (32) where W ˆ a (t) ∈ <q×m is an estimate of the ideal artificial weight Wa ∈ <q×m . (33) a In Ref. det(Kc ) 6= 0 is based on a linear system. 13 it is claimed that the mismatch term W˜ T (t)ω(x(t). W ˆ (0) = W ˆ 0. and (29) onto the uncertain system dynamics (13). nonexisting weight Wa = 0. and λ > 0 is the command governor gain. the command is given by c(t) . we have −1 T = Ar x(t) + Br cD (t) + B B T B  x(t) ˙ B {λξ(t) − λe(t) − e(t)}.aiaa. we have e(t) ˜ T (t)ω(x(t). which is updated online and is used in an adaptive control law of the form ˆ T (t)ω(x(t).D (t) = Ar xr. In contrast to traditional adaptive control frameworks. (27) where ξ(t) ∈ <n denotes the command governor states. (30) ˙ = 0 for λ → ∞ and that the overall system is stable. (26) g(t) = λξ(t) + [Ar − λIn ] e(t). c(t)).D (t) + Br cD (t). ΛW ˙ (29) Applying (25). 2015 | http://arc. −1 T The additional command cg (t) .org | DOI: 10. (36) 6 of 17 American Institute of Aeronautics and Astronautics . c(t))eT (t)P B + µω(x(t). and Γ = Γ0 /α2 it can be shown that the command governor framework is also scalable in the sense introduced in this paper. xr. the closed loop uncertain system still tracks the desired reference model given by x˙ r.13. this equation can be implemented without requiring differentiation of the system error e(t) or knowledge of its time derivative e(t) ˙ (see Ref.D (0) = xr0 (31) as the last term of (30) is appoximately zero for large λ.

aiaa. low-frequency learning. (43) where   −0. x(0) = 0. Note furthermore. x3 (t) represents the y-body-axis component of the angular velocity of the aircraft (pitch rate) with respect to the reference axes (in rad/s). III.065 −0. c(t))ω T (t)W   W ˆ a (t) = (40) a   = Γ0 ωz (x(t). ωaz0 6= 0. (41) Additionally.003 0. which can now be given as ˆ aT (t)[B T B]−1 B T {e˙ z (t) − Ar ez (t)}. (37) ωaz0 . x2 (t) represents the z-body-axis component of the velocity of the aircraft center of mass with respect to the reference axes (in ft/s). c(t))eT (t)P B + µω(x(t). e-modification. the scalability notion is illustrated using the controlled longitudinal motion of a Boeing 747 aircraft for numerical examples with reference model modification. ˆ a (0) = W W ˆ a0 . Then. ˆ a0 6= 0.429 0  0 0 1 0 h iT B= 0. x4 (t) represents the 7 of 17 American Institute of Aeronautics and Astronautics . A.322 −0.180 −1. consider the uncertain dynamical system representing the controlled longitudinal motion of a Boeing 747 airplane linearized at an altitude of 40 kft and a velocity of 774 ft/s given by x(t) ˙ = Ax(t) + Bu(t) + BW T ω(x(t).039 0 −0.2514/6. As a conclusion. ωa (t)/α. artificial basis function based.2015-0608 Conclusively. which is also consistent with the update law for artificial basis functions. by scaling the learning rate Γa = Γa0 /α2 (36) can be written as ˆ˙ a (t) = Γa0 ωaz (t)eT W z (t)P B. (33) can now be given as   ˙ = Ar z(t) + Br c0 (t) − BΛ W z(t) ˜ T (t)ωz (x(t). z(0) = z0 . ω˙ az (t) = ka W ωaz (0) = ωaz0 .74 0    A=  . 13 can be scaled in the sense introduced in this paper.01 −0. (42) which renders scalability of the system equation. 2015 | http://arc. c(t))eT T ˆ z (t)P B + µωz (x(t). (39) Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. that the adaptive weight update law (35) is invariant to the command coefficient when scaling the learning rate Γ = Γ0 /α2 ˆ˙ (t) = Γ ω(x(t). where Γa ∈ <q×q is a positive definite learning rate.org | DOI: 10. µ ∈ <+ is a design parameter and W For scaling of the system consider (14) and ωaz (t) .020 −1.36) with 1/α2 . (44)  0. ωa0 /α.39) are scalable in the sense described in this paper. c(t)) + W ˜ aT (t)ωaz (t) . Illustrative Examples In this section. (45) where x1 (t) represents the x-body-axis component of the velocity of the aircraft center of mass with respect to the reference axes (in ft/s). and command governor based adaptive control architectures as introduced in Sections C and D. Simulation Model Specifically.010 −0. scaling the learning rates of the adaptive weight update laws (35. (38) yielding invariance of the artificial weights update law with respect to the command coefficient α.319 7. (34. adaptive controllers using artificial basis functions according to Ref.16 0 . c(t)). c(t))ωaz (t)Wa (t) .

pitch Euler angle of the aircraft body axes with respect to the reference axes (in rad). However.03 is assumed. where it is shown that scaling the first system with α = 1. The system responses with different scaling command coefficients can be seen to be qualitatively identical. the responses using the same learning rate and different command do not display a scaled response. the regressor vector is given by ω(x(t). c(t)) and the learning rate Γ.444 0 0. resulting in a total uncertainty of h i W T = 0. the reference model modification introduced in Section II.B when the scaling factor is used. For the simulations. This is illustrated by Figure 5 with large scaled errors in both the system states and input.2514/6. 8 of 17 American Institute of Aeronautics and Astronautics . Reference Model Modification As a first example. However. Additionally.340 −1.444 0 (46) is an unknown ideal weight representing uncertainty due to modeling error in the pitch rate dynamics15 . Finally. 2015 | http://arc.03 .5◦ /s 0 (52) for the system with α = 1.2015-0608 Kx = −0. Simulations For all examples the command profile for the pitch rate is given by c(t) = αcα (t).9655 and h i Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. we choose the design matrix Q = I4 for the Lyapunov Equation (12) and the nominal control gains Kc = −8. as expected. a constant elevator bias of 0. Furthermore.5. (47)  T Furthermore.5 and plotting it with the second system yields an identical response. Figure 6 displays that the adaptive weights are not invariant to the command coefficient α when the scaling factor is not used and hence. By scaling the learning rates predictable performance with respect to a well tuned nominal system response is achieved. Figures 1 and 2 show the response of the system with α = 1 and α = 1. 12s ≤ t ≤ 16s. which is displayed in Figure 4.3546 .B.6154 0.068 −0.5◦ /s 0 . For the second system. (50) The initial conditions are given by  T x0 = 0 0 3. 1. subtracting the scaled first system from the second system shows a difference of numerical magnitude. (49) ◦ cα (t) = −4 /s. where cα (t) = 4◦ /s. if not used the system response is not predictable for nonidentical commands.3 is used for the illustration of the scalability notion with Γ0 = 0.068 −0.5 and the initial conditions are scaled as introduced in Section II.aiaa. (51)  T xr0 = 0 0 −1.0045 −0.org | DOI: 10. 0s ≤ t < 4s. Figure 7 shows that the adaptive weight response is invariant to the command coefficient as discussed in Section II.5I4 . α = 1. 8s ≤ t < 12s. respectively.5I5 and L = 4. (48) B. This numerical example highlights the key advantage of introducing the proposed scaling factor.C. c(t)) = xT (t) 1 .340 −1. which is consistent with the scaled performance as shown in the figures above. Here.0759 9. This is illustrated further in Figure 3. 4s ≤ t < 8s. and u(t) represents the elevator input (in rad). the qualitative observation from Figures 1 to 3 is also numerically verified. the scaled errors can be explained by the sensitivity of the adaptive weights response to the overall level of excitation as a combination of the excitation of the regressor vector ω(x(t). Additionally. Note that just the first three components of the adaptive weights are used in the figures of all examples in order to increase readability. Thus. h i WxT = 0. if the standard reference model modification architecture is used without the scaling factor.

2I5 and σ = 12. when using the original adaptive framework with artificial basis functions without scaling the learning rates Γ.1 and ωa0 = 0. Hence the adaptive weight update law is invariant to the scaling factor α and consequently. Figure 20 displays that the adaptive weights and the artificial weights are not invariant to the scaling factor. when the scaling factor is used.C. Low-Frequency Learning Now.2 is used for an example utilizing Γ0 = 12I5 . Γa0 = 0. Furthermore. employing artificial basis functions for adaptive control yields approximate scalability of the uncertain system’s responses by adapting the artificial basis functions and the artificial weights accordingly. As a conclusion. Figure 15 shows this difference between the adaptive weights. which is comparatively small in particular for the filtered adaptive weights Wˆf. the plot shows poor tracking performance due to the chosen set of gains. invariant to the command coefficient. Figure 17 shows the response of the system with α = 1 and α = 1. respectively. However. that the artificial basis functions are now scaled. e-modification is used with Γ0 = I5 and σe = 50. This particular set of gains is chosen on purpose to show scalability for oscillatory responses. respectively. e Modification As an example of the robustness modifications presented in Section II. However. Figure 8 shows the response of the system with α = 1 and α = 1. 4. Figure 12 shows the response of the system with α = 1 and α = 1. Now. this fact is further emphasized as displayed in Figure 9. Figure 18 illustrates the approximate scaling behavior of the adaptive control framework using artificial basis functions when the first system is scaled with the command coefficient α and plotted with the second system. the artificial and adaptive weights are invariant to the command coefficient α as shown in Figure 23.D is illustrated. showing a small scaled difference. when the scaling factor is not used. artificial weights update law and adaptive weights update law.1. As a conclusion.aiaa. it is obvious that the error between the scaled system stays within numerical magnitude. Furthermore. when the scaling factor is used. the scaling factor is applied on both. 2. Figure 21 illustrates that the artificial basis functions show significant scaling errors. Figure 11 shows that the responses of the adaptive weights are identical and thus.2514/6. Note that the adaptive framework using artificial basis functions13 can be scaled as shown in Section II. the difference introduced by the different command coefficient α remains small in comparison to the adaptive weight response of the reference model modification. Γf = 0. Although some more oscillations can be identified for the larger command coefficient. which shows that the scaling notion works for the e-modification. Figure 16 emphasizes that the scaled difference between the adaptive weight responses stays within errors of numerical magnitude when the scaling factor is utilized.1. low-frequency learning adaptive controllers can be scaled in the sense introduced in this paper.D. the overall tendency of the adaptive weights shows qualitatively improved performance with respect to the reference model modification displayed in Figure 6. the system responses show to be of qualitatively similar shape. for a system to be scalable.5. the low-frequency learning adaptive control framework introduced in Section II. the scalability of the adaptive control framework employing artificial basis functions as shown in Section II. and W ˆ a0 = 0. The simulation is executed for Γ0 = 0. Figures 13 and 14 show the response of the adaptive weights W ˆ and the filtered adaptive weights Wˆf.15.5. As mentioned. 9 of 17 American Institute of Aeronautics and Astronautics . Furthermore. Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. respectively.1. introducing the scaling factor further yields exact scalability in the sense introduced in this paper.5. Figure 22 displays the fact. 2015 | http://arc.1I5 . adaptive control architectures employing e-modification can be scaled in the sense introduced in this paper.org | DOI: 10. When plotting the scaled comparison of the two systems. However.C. Adaptive Control with Artificial Basis Functions In this section. and ka = 100. Although the behavior of this class of controllers is already approximately scalable by design. Building the scaled difference as shown in Figure 10. respectively. As mentioned before. As a consequence. µ = 1. where the scaled errors stay within numerical magnitude when utilizing the scalability notion. The approximate scaling property is further emphasized in Figure 19. respectively. The initial conditions are chosen ωa0 = 0. it is required that the adaptive weights are invariant to the scaling factor.2015-0608 3. the overall system’s response is scaled as illustrated in Figure 24.

References 1 S. E.2514/6. Hovakimyan and C. vol. J.1I5 . Control of Aircraft and Spacecraft. “Limited Authority Adaptive Flight Control for Reusable Launch Vehicles.Wittenmark. Schatz and T.” IEEE Transactions on Automatic Control. Navigation. 13 T. “A new adaptive law for robust adaptation without persistent excitation. and Control Conference. A. pp. Johnson and A. NJ: Prentice-Hall. 2013. robustness. E.” arXiv:1409. The command governor gain λ = 15 is used and the learning rate is chosen Γ0 = 0. Command Governor As a final example.1695. N. N.” Automatica. and open problems. NY: Dover Publications. 11 P. Calise. Figure 26 illustrates the approximate scalability in more detail. 2014 2 K. Note further that introducing a metric to evaluate approximate scalability may be an important step towards validation and verification of such systems. Future work will include extensions of the scalability notion to nonlinear regressor vectors. the uncertain system’s response is approximately scalable for different command coefficients. 5 E. but scalable command profiles. Cao. Kokotovic. command governor based adaptive controllers are scalable in the sense introduced in this paper. Ioannou and P. Yucelen. Gruenwald. 2003. and Dynamics. Mineola. T. NJ: Princeton University Press.” and research on metrics evaluating scalability for existing adaptive control architectures. “Reference dynamics modification in adaptive controllers for improved transient performance. “Scalability Concept for Predictable Closed-Loop Response of Adaptive Controllers. J. pp. E. 2015 | http://arc. S. “Instability analysis and improvement of robustness of adaptive control. Narendra and A. P. and transient performance. 14 B. IV. vol.” IEEE Transactions on Automatic Control. Lavretsky. giving the opportunity to extend the performance of a well-tuned case to diverse nonidentical. the command governor framework as presented in Section II. 2006. “Artificial Basis Functions in Adaptive Control for Transient Performance Improvement. Gibson. 134–145. “A new command governor architecture for transient response shaping. Upper Saddle River. Stable Adaptive Systems. 1984. S. and E. Control. 9 T. 1993. Finally. 583–594. Yucelen and E.” IEEE Transactions on Automatic Control (submitted). Johnson. Annaswamy. Conclusion This paper presented the scalability notion for adaptive control design using a model of the longitudinal motion of a Boeing 747 aircraft. M. M.C. PA: Society for Industrial and Applied Mathematics. 1987. 2013. Yucelen and W. Philadelphia. 8 E. 26.2015-0608 differences stay within numerical magnitude when utilizing the scalability notion. 20. Albattat. A. 5. Lavretsky. Hovakimyan. Adaptive Control. This could be due to the fact that more differential equations impose their nonlinear effects on the system’s performance in case of the artificial basis functions. 6 T. 2011.” American Control Conference. showing a small scaled difference between the two uncertain systems when using the original command governor framework. Annaswamy. 1994. Bryson.” AIAA Guidance. “Adaptive control of flight: theory. Note that although the command governor shows approximately scalable responses. Note that these small errors imply better scalability of the command governor with respect to the adaptive control framework using artificial basis functions as shown in the section above. Haddad.” AIAA Guidance. The system responses appear to be scaled. Note that the command governor is designed for shaping the transient performance10 and hence.4 is considered. 2010. Figure 25 shows the response of the system with α = 1 and α = 1. applications. 2014 15 A. the adaptive weights are not invariant to the scaling factor as displayed in Figure 27. Narendra and A. 7 N.” Journal of Guidance. M. “Performance Optimization of Adaptive Control Architectures.” IEEE Conference on Decision and Control. Princeton. Astrom and B. and Control Conference. Yucelen and E.aiaa. “Low-frequency learning and fast adaptation in model reference adaptive control. 4 K. Scaling the learning rates Γ was shown to render scaled system responses for systems with linear regressor vectors. M. and N.5. respectively. achieving “aproximate scalability. 12 K. Johnson. when using the original command governor framework without scaling the learning rates Γ. Consequently. Annaswamy. “Adaptive systems with closed-loop reference models: Stability. 32. 10 of 17 American Institute of Aeronautics and Astronautics . Wise. 2005. 2013. vol. Yucelen and A. Figure 28 displays that the scaled Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. Analysis employing a variety of simulations illustrated the positive effect on predictability of the closed-loop system responses of various adaptive control architectures. Lavretsky. 10 T. 3 K. N. Navigation. L1 adaptive control theory : Guaranteed robustness with fast adaptation.” International Journal of Adaptive Control and Signal Processing.org | DOI: 10.

Response of reference system and the uncertain system using reference model modification when α = 1. 2015 | http://arc.aiaa.2015-0608 Figure 2.5 (solid line) and α = 1 (dotted line) when the scaling factor is utilized. Response of reference system and the uncertain system using reference model modification when α = 1 (where the dotted lines indicate the system response). Figure 3. Scaled comparison of the system response using reference model modification with α = 1. Figure 4. 11 of 17 American Institute of Aeronautics and Astronautics .2514/6.5 when the scaling factor is utilized (where the dotted lines indicate the system response). Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. Figure 1.org | DOI: 10. Scaled difference in system responses using reference model modification when the scaling factor is utilized.

5 (right). Figure 8. Scaled difference in system responses using reference model modification when the scaling factor is not utilized. Figure 5. Adaptive weights as a function of time using reference model modification when α = 1 (left) and α = 1. Figure 7. and the scaling factor is utilized.2015-0608 Figure 6.org | DOI: 10. Response of reference system (solid line) and the uncertain system (dotted line) using e-modification when α = 1 (left) and α = 1.aiaa. Adaptive weights as a function of time using reference model modification when α = 1 (left) and α = 1. 2015 | http://arc. 12 of 17 American Institute of Aeronautics and Astronautics . Downloaded by CARLETON UNIVERSITY LIBRARY on July 31.5 (right) when the scaling factor is utilized.2514/6.5 (right) when the scaling factor is not utilized.

and the scaling factor is utilized.2015-0608 Figure 10.5 (solid line) and α = 1 (dotted line) when the scaling factor is utilized. Adaptive weights using e-modification as a function of time when α = 1 (left) and α = 1. Scaled difference in system responses using e-modification when the scaling factor is utilized. Response of reference system (solid line) and the uncertain system (dotted line) using low frequency learning when α = 1 (left) and α = 1.2514/6.org | DOI: 10.5 (right).aiaa. Figure 11.5 (right). Scaled comparison of the system response using e-modification with α = 1. Figure 12. Figure 9. and the scaling factor is utilized. 13 of 17 American Institute of Aeronautics and Astronautics . 2015 | http://arc. Downloaded by CARLETON UNIVERSITY LIBRARY on July 31.

Response of filtered adaptive weights using low frequency learning when α = 1 (left) and α = 1. Downloaded by CARLETON UNIVERSITY LIBRARY on July 31.org | DOI: 10.5 (right) and the scaling factor is not utilized.2514/6.5 (right) and the scaling factor is not utilized. Figure 13. Difference of filtered adaptive weights (left) and adaptive weights (right) using low frequency learning when the scaling factor is not utilized.2015-0608 Figure 14. Difference of filtered adaptive weights (left) and adaptive weights (right) using low frequency learning when the scaling factor is utilized. 2015 | http://arc. 14 of 17 American Institute of Aeronautics and Astronautics . Response of adaptive weights using low frequency learning when α = 1 (left) and α = 1. Figure 15.aiaa. Figure 16.

Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. Figure 19. Scaled difference in system responses using artificial basis functions when the scaling factor is not utilized. and the scaling factor is not utilized.2015-0608 Figure 18. Difference in adaptive weights (right) and artificial weights (left) as a function of time when the scaling factor is not utilized.org | DOI: 10. Scaled comparison of the system response with α = 1. Figure 17.2514/6. 2015 | http://arc. Response of reference system (solid line) and the uncertain system (dotted line) using artificial basis functions when α = 1 (left) and α = 1.aiaa.5 (right).5 (solid line) and α = 1 (dotted line) using artificial basis functions when the scaling factor is not utilized. Figure 20. 15 of 17 American Institute of Aeronautics and Astronautics .

Scaled difference in system responses using artificial basis functions when the scaling factor is utilized. 2015 | http://arc. Difference (left) and scaling of artificial basis functions (right) as a function of time when the scaling factor is not utilized (where the dotted line represents the scaled first system). Figure 21.2015-0608 Figure 22. Figure 24.aiaa.org | DOI: 10. Difference in adaptive weights (right) and artificial weights (left) as a function of time when the scaling factor is utilized.2514/6. Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. Difference (left) and scaling of artificial basis functions (right) as a function of time when the scaling factor is utilized (where the dotted line represents the scaled first system). Figure 23. 16 of 17 American Institute of Aeronautics and Astronautics .

17 of 17 American Institute of Aeronautics and Astronautics . Figure 28.5 (right). Scaled difference in system responses using original command governor. Scaled difference in system responses using command governor when the scaling factor is utilized.aiaa. Figure 25.2015-0608 Figure 26.5 (right). Figure 27. Response of reference system (solid line) and the uncertain system (dotted line) using original command governor when α = 1 (left) and α = 1.org | DOI: 10. Downloaded by CARLETON UNIVERSITY LIBRARY on July 31. Adaptive weights using original command governor as a function of time when α = 1 (left) and α = 1.2514/6. 2015 | http://arc.