Journal of Control Science and Engineering

Journal of Control Science and Engineering / 2012 / Article
Special Issue

Adaptive Control Theory and Applications

View this Special Issue

Research Article | Open Access

Volume 2012 |Article ID 502149 |

Yu Liu, Luis G. Crespo, "Adaptive Control Allocation in the Presence of Actuator Failures", Journal of Control Science and Engineering, vol. 2012, Article ID 502149, 16 pages, 2012.

Adaptive Control Allocation in the Presence of Actuator Failures

Academic Editor: Yunjun Xu
Received03 Nov 2011
Revised17 Feb 2012
Accepted02 Mar 2012
Published14 May 2012


This paper proposes a control allocation framework where a feedback adaptive signal is designed for a group of redundant actuators and then it is adaptively allocated among all group members. In the adaptive control allocation structure, cooperative actuators are grouped and treated as an equivalent control effector. A state feedback adaptive control signal is designed for the equivalent effector and adaptively allocated to the member actuators. Two adaptive control allocation algorithms, guaranteeing closed-loop stability and asymptotic state tracking when partial and total loss of control effectiveness occur, are developed. Proper grouping of the actuators reduces the controller complexity without reducing their efficacy. The implementation and effectiveness of the strategies proposed is demonstrated in detail using several examples.

1. Introduction

Actuator redundancy is highly desirable for fault-tolerant control. This redundancy yields multiple ways to implement the forces and moments prescribed by the controller. However, this freedom creates the need for properly allocating the control inputs among all individual actuators. While multiple actuator configurations do generate the desired forces and moments, some of them may yield unintended outcomes, for example, the effect of some control surface deflections counteracts the effect of other ones. Redundancy management is the need for properly allocating the control inputs among functionally redundant actuators when some of them may not be fully functional.

The purpose of control allocation is to distribute the control signals to the available actuators such that the desired moments and forces are efficiently generated. Traditional control allocation methods include explicit ganging [1], daisy chaining [2], pseudinverse [3, 4], and error and control minimization [5–10]. Explicit ganging performs control allocation by finding a fixed relation between the desired control moments and forces and the designed control signals. Multiple actuators (e.g., two aileron surfaces) can be combined to generate the desired effects. Daisy chaining allocates inputs in a prioritized fashion. It utilizes the actuators in sequence to generate certain effect. If an actuator is unable to generate such an effect, say due to actuator saturation, the next actuator in the sequence will be used. The pseudoinverse approach, which accounts for input saturation and failure, performs control allocation by solving a linear optimization problem in real time. Error and control minimization is another common control allocation approach. This approach minimizes the error between the desired and generated control moments subject to control constraints. Several approaches can accommodate for actuator failure and saturation compensation, provided that the actuator failure or saturation has been identified. Thus one obvious drawback of these approaches is that they require a fault detection system. These systems, which are usually complex, require the characterization of several failure modes a priori. Furthermore, they may require solving optimization problems in real time; thus they can substantially increase the software and hardware demands of the flight control system.

Adaptive control, on the other hand, does not require knowing which controllers have failed nor the class or severity of the failure. This is due to its ability to change control parameters according to the existing flight condition. Due to parameter adaptation, it is also able to accommodate for parametric uncertainties in the vehicle dynamics. Substantial developments in adaptive control for actuator failures have been made in the last decade [11, 12]. Adaptive control’s ability to seamlessly compensate for actuator failures requires for the system’s built-in actuation redundancy to be sufficient. This is usually described as a rank condition on the input matrix 𝐡 [11, 13]. To take advantage of all redundancy available, one common approach in multivariable adaptive control is to generate a control signal for each control surface. Such an approach endows the controller the maximum degree of freedom to compensate for failure. When certain control surfaces are stuck or have reduced control effectiveness, the remaining control surfaces will adaptively cooperate until a new combination of inputs for the remaining control surfaces is found. This will occur automatically without knowing which surfaces have failed, or when such failures occur. Although adaptive control ensures closed-loop stability and tracking performance, it does not differentiate between control generation and control allocation, and the resulting actuation scheme may be unacceptable. For instance, separately designed control signals for multiple control surface segments may cancel each other’s effects, for example, it has been observed that the steady-state deflection of both rudder segments for a direct adaptive control had opposite angles, resulting in a wings leveled flight with increased drag.

The lack of control allocation in the current direct adaptive control framework motivates this research effort. In this study, we aim at separating control generation from control allocation. In the adaptive control allocation framework, a key step is to combine redundant control surfaces similar to explicit ganging. For each group of combined actuators, we design an adaptive control signal, which is then allocated among group members by an adaptive gain. If no failure occurs, a nonadaptive control allocation scheme set in advance is enforced. In the presence of uncertainty of actuator failure, the adaptive flight controller modifies the allocation of input accordingly.

The structure of the adaptive control allocation framework is illustrated in Figure 1. This is a simple aircraft control example with the elevator controlling the pitch motion. The aircraft longitudinal state, denoted as π‘₯(𝑑), should track the state of a reference system for a given reference input π‘Ÿ(𝑑). The elevator consists of four segments, namely, left outboard, left inboard, right outboard, and right inboard segments. For pitch control, they can be grouped together and considered as an equivalent elevator by adding the four columns of the input matrix. A β€œvirtual’’ elevator signal 𝑣0 is generated by the adaptive controller for the desired pitching maneuver. This elevator signal is then allocated by the adaptive allocation gains 𝛼𝑖(𝑑),𝑖=1,…,4. The resulting elevator signals 𝑣0𝑖(𝑑)=𝛼𝑖(𝑑)𝑣0(𝑑),𝑖=1,…,4 will be fed to the four elevator segments. The allocation gains can be updated on-line based on the knowledge of the nominal plant and 𝑣0 to mitigate the uncertainties of actuator failures. Conversely, in a fixed allocation scheme 𝛼𝑖 are constant, for example, 𝛼𝑖=1/4, 𝑖=1,…,4.

One advantage of this adaptive control allocation structure is the ease in solving problems such as the counteracting actuation. To remedy this problem, the signs of the adaptive allocation gains 𝛼𝑖 can be made the same, say, via the projection algorithm [14] so that the allocated control signals cannot go in opposite directions. Another advantage of this structure is the reduction of the controller complexity. For instance, let us consider a state feedback adaptive control design for an 𝑛-state system with π‘š controls. The number of controller parameters to be updated is π‘šΓ—π‘›. If the problem is solved by having all actuators in one group, there are π‘š+𝑛 adaptive parameters. For example, if we try to stabilize a 5-state system with 3 controls, the total adaptive parameters using a direct model reference adaptive controller would be 15. With a grouping of control inputs for which the system remains controllable (i.e., the system is controllable for the control input matrix associated with the grouping), the total number of adaptive parameters required is 8 (1Γ—5 gain vector and 3𝛼’s). The larger the family of redundant actuators, the bigger the benefit. Another advantage of this approach is that the virtual control signal can be designed using a nonadaptive (fixed) state feedback gain, and failure compensation is achieved by only adapting the control allocation gains 𝛼’s. This implies that the proposed structure could be added to a conventionally designed state feedback controllers without having to use fault detection and isolation.

In this paper, we develop the mathematical foundation of this adaptive control allocation structure for a single group of actuators. Two adaptive control allocation algorithms are presented for both loss of effectiveness and constant-magnitude actuator failures. Technical issues such as design conditions, adaptive law designs, and stability analysis are addressed. The proposed schemes are shown to guarantee stability and asymptotic state tracking in the presence of unknown failures. Simulation-based examples are used to demonstrate the strategy proposed.

The paper is organized as follows. Section 2 presents the adaptive control allocation algorithm for loss of control effectiveness. This is followed by Section 3, where a scheme for constant-magnitude actuator failure compensation is developed in Section 3. Finally, a few concluding remarks close the paper.

2. Adaptive Control Allocation Design for Loss of Effectiveness Failures

2.1. Problem Formulation
2.1.1. Plant

Consider the linear time-invariant (LTI) systemΜ‡π‘₯(𝑑)=𝐴π‘₯(𝑑)+𝐡𝑒(𝑑),(1) where π‘₯βˆˆπ‘…π‘› and π‘’βˆˆπ‘…π‘š are the system state and the control input. The matrices π΄βˆˆπ‘…π‘›Γ—π‘› and π΅βˆˆπ‘…π‘›Γ—π‘š are constant and known. The matrix 𝐡 is the control gain matrix for the group of actuators.

The control signal 𝑒(𝑑) can be expressed as𝑒(𝑑)=Λ𝑣(𝑑),(2) where 𝑣(𝑑) is the allocated control signal (i.e., input to the actuator) and Ξ› is a piecewise constant uncertain diagonal control effectiveness matrix with Ξ›=diag{πœ†1,πœ†2,…,πœ†π‘š}, and 0<πœ†π‘–β‰€1, 𝑖=1,…,π‘š. The 𝑖th actuator is fully functional when πœ†π‘–=1 and has a loss of effectiveness failure when 0<πœ†π‘–<1. For the design in this section, we assume that πœ†π‘–β‰ 0, that is, no actuator outage occurs. For the reminder of this section we assume that Ξ› is always positive definite.

The control input 𝑣(𝑑)=[𝑣1(𝑑),𝑣2(𝑑),…,π‘£π‘š(𝑑)]⊀ is given by𝑣𝑗(𝑑)=𝛼𝑗(𝑑)𝑣0(𝑑),𝑗=1,2,…,π‘š,(3) where 𝑣0(𝑑) is a control signal designed for the group, and 𝛼𝑗(𝑑) is the adaptive allocation gain for 𝑗th actuator. We assume that a nominal allocation scheme has been prescribed. This scheme will be enforced under nominal operating conditions. The nominal allocation gain vector is denoted as π›Όβˆ—βˆˆπ‘…π‘š. We also define the equivalent control gain vector as 𝑏0β‰œπ΅π›Όβˆ—. The vector 𝑏0 can be seen as the equivalent control gain matrix for the equivalent control effector representing the actuator group. We further assume that the pair (𝐴,𝑏0) is stabilizable.

2.1.2. Reference Model

For the adaptive control, the desired closed-loop dynamics is given byΜ‡π‘₯π‘š(𝑑)=π΄π‘šπ‘₯π‘š(𝑑)+π΅π‘šπ‘Ÿ(𝑑),(4) where π΄π‘šβˆˆπ‘…π‘›Γ—π‘› is a Hurwitz matrix, and π΅π‘šβˆˆπ‘…π‘›. The signal π‘Ÿ(𝑑)βˆˆπ‘… is a bounded piecewise continuous reference input, and π‘₯π‘š(𝑑) is the desired state. For a given symmetric positive definite matrix π‘„βˆˆπ‘…π‘›Γ—π‘›, there exists a unique π‘ƒβˆˆπ‘…π‘›Γ—π‘› that satisfiesπ‘ƒπ΄π‘š+π΄βŠ€π‘šπ‘ƒ=βˆ’π‘„,𝑃=π‘ƒβŠ€>0.(5) For the adaptive control design, we need the following standard plant model matching condition.

Assumption 1. There exist constant πΎβˆ—1βˆˆπ‘…π‘›, πΎβˆ—2βˆˆπ‘… such that 𝐴+𝑏0𝐾1βˆ—βŠ€=π΄π‘š,𝑏0πΎβˆ—2=π΅π‘š.(6)

2.1.3. Control Objective

The control objective is to design the virtual control signal 𝑣0(𝑑) and adaptive allocation gains 𝛼𝑗,𝑗=1,…,π‘š, such that all the closed-loop signals are bounded and the system state π‘₯(𝑑) tracks the desired state π‘₯π‘š(𝑑) asymptotically in the presence of uncertain control effectiveness Ξ›.

2.2. Adaptive Control Allocation Design
2.2.1. Nominal Controller

The plant model matching condition in Assumption 1 indicates the existence of a nominal controller π‘£βˆ—(𝑑) for the system without failures and a nominal constant allocation gain vector π›Όβˆ— such that the closed-loop response is identical to that of the reference model when the responses of any unmatched initial conditions vanish exponentially. This signal takes the formπ‘£βˆ—0(𝑑)=𝐾1βˆ—βŠ€π‘₯(𝑑)+πΎβˆ—2π‘Ÿ(𝑑)β‰œπœƒβˆ—βŠ€π‘£πœ”(𝑑),βˆ—(𝑑)=π›Όβˆ—π‘£βˆ—0(𝑑),(7) where πœƒβˆ—β‰œ[𝐾1βˆ—βŠ€,πΎβˆ—2]βŠ€βˆˆπ‘…π‘›+1, and πœ”(𝑑)=[π‘₯⊀(𝑑),π‘Ÿ(𝑑)]⊀. The above state feedback control π‘£βˆ—0(𝑑) together with a prespecified distribution π›Όβˆ— ensures the state tracking error 𝑒(𝑑)=π‘₯(𝑑)βˆ’π‘₯π‘š(𝑑) approaches zero exponentially.

2.2.2. Adaptive Controller

When a failure occurs, the allocation gain will be adaptively adjusted to accommodate for failure, but π›Όβˆ— may no longer be effective. For this, we use the adaptive versions of control signal and allocation gain𝑣0(𝑑)=𝐾⊀1(𝑑)π‘₯(𝑑)+𝐾2(𝑑)π‘Ÿ(𝑑)=πœƒβŠ€(𝑑)πœ”(𝑑),𝑣(𝑑)=𝛼(𝑑)𝑣0(𝑑),(8) where 𝐾1(𝑑) and 𝐾2(𝑑) are estimates of πΎβˆ—1 and πΎβˆ—2, and πœƒ(𝑑)=[𝐾⊀1(𝑑),𝐾2(𝑑)]βŠ€βˆˆπ‘…π‘›+1. The updated 𝛼(𝑑)=[𝛼1(𝑑),…,π›Όπ‘š(𝑑)]⊀ is an estimate of π›Όβˆ—.

2.2.3. Error Dynamics

With the plant and control in (1) and (2), nominal controller in (7), and adaptive controller in (8), the closed-loop dynamics can be expressed as Μ‡π‘₯(𝑑)=𝐴π‘₯(𝑑)+𝐡Λ𝛼(𝑑)𝑣0(𝑑)=𝐴π‘₯(𝑑)+π΅π›Όβˆ—π‘£0(𝑑)+𝐡𝛼(𝑑)𝑣0(𝑑)=𝐴π‘₯(𝑑)+𝑏0𝑣0(𝑑)+𝐡𝛼(𝑑)𝑣0(𝑑)=𝐴π‘₯(𝑑)+𝑏0π‘£βˆ—0(𝑑)+𝑏0ΜƒπœƒβŠ€(𝑑)πœ”(𝑑)+𝐡𝛼(𝑑)𝑣0=ξ€·(𝑑)𝐴+𝑏0𝐾1βˆ—βŠ€ξ€Έπ‘₯(𝑑)+𝑏0πΎβˆ—2π‘Ÿ(𝑑)+𝑏0ΜƒπœƒβŠ€(𝑑)πœ”(𝑑)+𝐡𝛼(𝑑)𝑣0(𝑑)=π΄π‘šπ‘₯(𝑑)+π΅π‘šπ‘Ÿ(𝑑)+𝑏0ΜƒπœƒβŠ€(𝑑)πœ”(𝑑)+𝐡𝛼(𝑑)𝑣0(𝑑),(9) where 𝛼(𝑑)β‰œΞ›π›Ό(𝑑)βˆ’π›Όβˆ—(𝑑) and Μƒπœƒ(𝑑)=πœƒ(𝑑)βˆ’πœƒβˆ—.

From the closed-loop dynamics in (9) and reference model in (4), the error dynamics can be obtained aṡ𝑒(𝑑)=π΄π‘šπ‘’(𝑑)+𝑏0ΜƒπœƒβŠ€(𝑑)πœ”(𝑑)+𝐡𝛼(𝑑)𝑣0(𝑑).(10) It can be seen that the error dynamics in (10) is suitable for adaptive law design in that the latter half of its right-hand side is linear in Μƒπœƒ(𝑑) and 𝛼(𝑑).

2.2.4. Adaptive Laws

Based on the error dynamics in (10), we can design the adaptive laws for the control parameter πœƒ(𝑑) and allocation gain 𝛼(𝑑) asΜ‡πœƒ(𝑑)=βˆ’Ξ“πœƒπœ”(𝑑)π‘’βŠ€(𝑑)𝑃𝑏0,(11)̇𝛼(𝑑)=βˆ’Ξ“π›Όπ΅βŠ€π‘ƒπ‘’(𝑑)𝑣0(𝑑),(12) where Ξ“πœƒ and Γ𝛼 are symmetric positive definite matrices and 𝑃 is determined by (5). From the adaptive laws, we can see that the controller parameters of the virtual control signal for the actuator group are updated using the information of the equivalent control gain vector 𝑏0. The allocation gains are updated with the 𝐡 matrix since successful allocation of the virtual control signal to each actuator requires the knowledge of each column of 𝐡.

The properties of the adaptive control allocation scheme can be summarized in the following theorem whose proof is presented in the appendix.

Theorem 1. For the system in (1), the adaptive controller and allocation scheme in (8), and the adaptive laws in (11) and (12) guarantee that the all the closed-loop signals are bounded and limπ‘‘β†’βˆž[π‘₯(𝑑)βˆ’π‘₯π‘š(𝑑)]=0 in the presence of uncertain loss of effectiveness actuator failures in (2).

Remark 2. From the definition of 𝛼(𝑑) in (9), we can see that the adaptation of 𝛼(𝑑) is essential for compensating the uncertainties in Ξ›, while 𝐾1(𝑑) and 𝐾2(𝑑) can be fixed to their nominal values πΎβˆ—1 and πΎβˆ—2. In this case, with Μƒπœƒ(𝑑)=0 in (10) and (A.1), the closed-loop stability and asymptotic state tracking results still hold.

2.3. Examples
2.3.1. Linear Plants

Two case studies based on linearized aircraft models are presented next.

Consider the linearized lateral dynamic model of a large transport aircraft flying in a steady wings-level cruise condition with 𝑒=778ft/s [15]. The aircraft model is 𝑣̇π‘₯=𝐴π‘₯+𝐡𝑒,(13)π‘₯=𝑏,𝑝𝑏,π‘Ÿπ‘ξ€»,πœ™,πœ“βŠ€ξ€Ίπ›Ώ,𝑒=π‘Ž,π›Ώπ‘Ÿξ€»βŠ€.(14) The state includes the lateral velocity 𝑣𝑏 (ft/s), roll rate 𝑝𝑏 (rad/s), yaw rate π‘Ÿπ‘ (rad/s) (all in body-axis frame), roll angle πœ™ (rad), and yaw angle πœ“ (rad). The control inputs are aileron π›Ώπ‘Ž and rudder π›Ώπ‘Ÿ deflections (deg). The system matrices 𝐴 and 𝐡 are ⎑⎒⎒⎒⎒⎒⎒⎒⎒⎣⎀βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯⎦,⎑⎒⎒⎒⎒⎒⎒⎒⎒⎣⎀βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯⎦.𝐴=βˆ’0.12928.328βˆ’774.9232.1450βˆ’0.012βˆ’1.44190.9409000.004βˆ’0.0409βˆ’0.1757βˆ’0.00010010.037200001.000700𝐡=0.05420.46690.04430.02000.0025βˆ’0.03820000(15) Hereafter we will group the aileron and rudder into a single group. Note that these surfaces have related but different functionalities. If adaptive control signals are designed separately for the two control inputs with the state feedback structure, the total updated parameters would be (5+1)Γ—2=12. They include 5 state feedback parameters and 1 feed-forward parameter (for tracking the reference input) for each control surface. The adaptive control allocation scheme, on the other hand, requires 5+1+2=8 parameters since only one adaptive control signal is designed and allocated with two updated allocation gains.

Nominal Parameters
The nominal allocation gain vector is selected as π›Όβˆ—=[8.6103,βˆ’1]⊀, thus the equivalent control gain vector is 𝑏0=π΅π›Όβˆ—=[0,0.3614,0.0594,0,0]⊀. The nominal allocation gain is chosen so that the first element in 𝑏0, that is, the control gain for the lateral acceleration is zero. The purpose of this choice is to attain a coordinated turn.
The nominal controller is designed based on the LQR approach for (𝐴,𝑏0). The resulting gains are πΎβˆ—1=[]βˆ’0.8963,22.1655,βˆ’73.6645,28.8488,4.0825⊀,πΎβˆ—2=1.(16)

Reference Model
For this example, the reference model is chosen as the closed-loop dynamics with the above LQR controller, that is, π΄π‘š=𝐴+𝑏0𝐾1βˆ—βŠ€,π΅π‘š=𝑏0πΎβˆ—2.(17) The reference input to the reference model is chosen as π‘Ÿ(𝑑)=2.1376 for 𝑑β‰₯0, which leads to a desired state trajectory with steady-state values π‘₯π‘š[](∞)=0,0,0,0,0.5236⊀.(18) This reference command yields a right turn with a change in the yaw angle of 30 degrees.

Actuator Failure
We will consider the 80% loss in control effectiveness: 𝑒1(𝑑)=0.2𝑣1(𝑑),for𝑑β‰₯10seconds,(19) where 𝑒1(𝑑) is the output of the aileron actuator and 𝑣1(𝑑) is the designed control input to aileron.

Simulation Results
The following cases will be studied.(i)Case 1: adaptive allocation of the adaptive control designed in Section 2.2 ((11) and (12)).(ii)Case 2: adaptive allocation (as in (12)) of a nonadaptive control signal with fixed gains (as in (16)).

Case 1. The time history of the system state under failure and the desired state is shown in Figure 2. It can be seen that the asymptotic state tracking is achieved after failure.

Figure 3 shows the designed control signal 𝑣0(𝑑), allocated control signals 𝑣(𝑑) and actuator outputs 𝑒(𝑑). The designed control signal is allocated by the adaptive allocation gains. The actuator outputs are different from the allocated control signals due to the failure.

The adaptive parameters are shown in Figure 4. We can see that 𝐾2(𝑑) also plays an important role in the compensation of actuator failures. The adaptive allocation gains are shown in Figure 5. Both allocations gains adapt immediately after the failure occurs and settle to a new combination of steady-state values that, together with 𝐾2(𝑑) and 𝐾1(𝑑), guarantee state tracking after failure.

Case 2. The system state and desired state are shown in Figure 6. Despite the uncertain actuator failure, asymptotic state tracking is achieved. The designed control signal, allocated control signals and the actuator outputs are shown in Figure 7. Similar to the previous case, the allocated control signals and the actuator outputs are not identical due to the actuator failure. The allocation gains are shown in Figure 8. The allocation gains are updated autonomously for failure compensation.
In this simulation case, we have achieved similar results to Case 1 with a slightly degraded transient response (see Figures 2 and 6). A possible explanation for it is that the controller parameters are fixed in this case and cannot contribute to the failure compensation and trajectory tracking as they do in Case 1. However, closed-loop stability and asymptotic state tracking are achieved even though only the allocation gains are adapted. The successful demonstration of failure compensation in Case 2 implies that the proposed adaptive allocation unit can be added to a control loop having any state feedback controller. The added adaptive allocation to the nonadaptive controller ensures closed-loop stability and asymptotic state tracking despite uncertain actuator failures.

2.3.2. NASA Generic Transport Model

In this section, we apply the control allocation strategy to the NASA Generic Transport Model (GTM). The NASA GTM is a high-fidelity model of the NASA AirSTAR UAV testbed [16, 17]. The purpose of this example is to show that this adaptive scheme, with the grouping of actuators having different physical functions, can be applied to the nonlinear plant to achieve closed-loop stability and asymptotic tracking.

LTI Model
For this simulation study, we trim and linearize the GTM at a wings-level flight for an aerodynamic speed of 92.09 knots. The same states and controls of (14) are used.

Flight Conditions
As in the previous example, the aircraft is commanded to turn right from the initial wings-level horizontal flight. The turn starts at 10 seconds and at steady state the heading angle will increase 60 degrees.

Actuator Failure
We let the left aileron lose 90% of its effectiveness at 12 seconds. The failure magnitude and its onset time instant are unknown to the controller. The control objective is for the aircraft to achieve an accurate turn in the presence of the failure.

Simulation Results
The simulation results are shown in Figures 9–14. Figure 9 shows the relevant states of the reference model and of the plant. The states track the reference trajectories asymptotically in spite of the disturbance caused by the failure. The yaw angle is shown to reach 60 degrees accurately. This accurate turning is also shown in Figure 11 by the ground track of the aircraft.
The longitudinal states are shown in Figure 10. There are fluctuations in the longitudinal states since they are not controlled by aileron and rudder.
Figures 12 and 13 show the time history of the virtual control signal, allocated control signals, and actual aileron and rudder deflections. The discontinuity in Figure 13 at the top is a consequence of failure.
Figure 14 shows the control allocation gains and controller parameters. The parameters adjust autonomously after the failure occurs. Note that the allocation gains start to update at the beginning of the turn at 𝑑=10 seconds, before the failure occurs. The reason for this phenomenon is that the allocation gains are improving the tracking performance beyond of what the adaptive gains can do alone.

3. Adaptive Control Allocation Design for Constant Failures

3.1. Problem Formulation

When constant failures occur, the control signal can be rewritten as [11]𝑒(𝑑)=𝑣(𝑑)+πœŽπ‘“ξ€·ξ€Έ,π‘’βˆ’π‘£(𝑑)(20) where 𝑒=[𝑒1,…,π‘’π‘š]⊀ is the failure vector whose elements are unknown constants, and πœŽπ‘“ represents the failure pattern, which is defined asπœŽπ‘“ξ€½πœŽ=diag1,𝜎2,…,πœŽπ‘šξ€Ύ,(21) with πœŽπ‘–=1 if the 𝑖th actuator has failed, that is, 𝑒𝑖=𝑒𝑖, and πœŽπ‘–=0 otherwise. The failures are assumed to occur instantaneously, that is, πœŽπ‘– are piecewise constant function of time. An example of such actuator failures is when a control surface (such as the rudder or an aileron) is stuck at some unknown fixed angle at an unknown time instant. This type of failures could be caused by failed hydraulic systems or mechanical linkages.

The plant dynamics can then be rewritten asξ€·Μ‡π‘₯(𝑑)=𝐴π‘₯(𝑑)+π΅πΌβˆ’πœŽπ‘“ξ€Έπ‘£(𝑑)+π΅πœŽπ‘“π‘’.(22) The constant failure 𝑒 introduces an uncertain disturbance that needs to be accommodated for. The control objective for this adaptive control allocation scheme is to design 𝑣(𝑑) to guarantee closed-loop stability and asymptotic state tracking when uncertain constant failures occur.

For adaptive control of constant actuator failures, sufficient built-in actuation redundancy is required. The redundancy condition is described in the following assumption.

Assumption 2. The rank of 𝐡 matrix satisfies that rank[𝐡]=1, and there is at least one operable actuator in the system.

The rank condition characterizes the redundancy of actuation which is necessary for a successful constant failure compensation. This rank condition suggests that the system remains controllable after a failure, and the effect of the constant failure can be properly matched and canceled by the allocated control signals through other columns of 𝐡. Based on this condition, there can be up to π‘šβˆ’1 constant actuator failures at any given time.

3.1.1. Nominal Controller

Consider the nominal controller structureπ‘£βˆ—0(𝑑)=𝐾1βˆ—βŠ€π‘₯(𝑑)+πΎβˆ—2π‘Ÿ(𝑑)+πΎβˆ—3β‰œπœƒβˆ—βŠ€π‘£πœ”(𝑑),βˆ—(𝑑)=π›Όβˆ—π‘£βˆ—0(𝑑),(23) where πœƒβˆ—=[𝐾1βˆ—βŠ€,πΎβˆ—2,πΎβˆ—3]⊀ and πœ”=[π‘₯⊀(𝑑),π‘Ÿ(𝑑),1]⊀. When there is no failure in the system, πΎβˆ—1, πΎβˆ—2, and π›Όβˆ— are chosen as in (6), and πΎβˆ—3 is set to be zero. In this way, the controller ensures the match between the reference model and the nominal plant without failures. Similar to the previous section, we also define π΅π›Όβˆ—=𝑏0 which is known in advance for controller design.

Next we will show that the controller in (23) attains model matching when failures occur. When there are failures in the system, πΎβˆ—3 cannot generally be zero, and a new set of allocation gains, denoted as π›Όβˆ—, may be needed. From (22) and (23) we obtainξ€·Μ‡π‘₯=𝐴π‘₯+π΅πΌβˆ’πœŽπ‘“ξ€Έπ›Όβˆ—ξ€·πΎ1βˆ—βŠ€π‘₯+πΎβˆ—2π‘Ÿ+πΎβˆ—3ξ€Έ+π΅πœŽπ‘“π‘’.(24) We assume that at most 𝑝 actuators can fail with π‘β‰€π‘šβˆ’1 so that there is at least one operable actuator left. Define an index set for failed actuators as β„±={𝑖1,…,𝑖𝑝} such that πœŽπ‘˜=1 for any π‘˜βˆˆβ„±. Then 𝐡(πΌβˆ’πœŽπ‘“)π›Όβˆ— can be expressed asπ΅ξ€·πΌβˆ’πœŽπ‘“ξ€Έπ›Όβˆ—=ξ“π‘—βˆ‰β„±π‘π‘—π›Όβˆ—π‘—,(25) where 𝑏𝑗 is the 𝑗th column of 𝐡 and π›Όβˆ—π‘— is the 𝑗th element of π›Όβˆ—.

Based on the rank condition in Assumption 2, we know that the columns of 𝐡 are linearly dependent, that is, for any two columns of 𝐡, 𝑏𝑖 and 𝑏𝑗, 𝑏𝑖=𝑐𝑖𝑗𝑏𝑗 where 𝑐𝑖𝑗 is a constant scalar. Thus we know that the vector 𝑏0=π΅π›Όβˆ—, which is the linear combination of all columns of 𝐡, is also parallel to any column in 𝐡. So for each column π‘π‘˜ in 𝐡, π‘˜=1,2,…,π‘š, we can find a scalar π‘π‘˜ satisfying π‘π‘˜=π‘π‘˜π‘0. Therefore (25) can be expressed asπ΅ξ€·πΌβˆ’πœŽπ‘“ξ€Έπ›Όβˆ—=ξ“π‘—βˆ‰β„±π‘π‘—π›Όβˆ—π‘—=ξ“π‘—βˆ‰β„±π‘π‘—π›Όβˆ—π‘—π‘0.(26) If   π›Όβˆ—π‘—, π‘—βˆ‰β„± are chosen such thatξ“π‘—βˆ‰β„±π‘π‘—π›Όβˆ—π‘—=1,(27) then we may obtainπ΅ξ€·πΌβˆ’πœŽπ‘“ξ€Έπ›Όβˆ—=𝑏0.(28) One possible choice for π›Όβˆ—π‘—, π‘—βˆ‰β„± isπ›Όβˆ—π‘—=1𝑐𝑗.(π‘šβˆ’π‘)(29) Equation (28) indicates that, under constant failures, the plant model condition in (6) can still be satisfied. Note that for the system without failuresπ΅π›Όβˆ—=π‘šξ“π‘—=1π›Όβˆ—π‘—π‘π‘—=π‘šξ“π‘—=1π‘π‘—π›Όβˆ—π‘—π‘0=𝑏0(30) which implies thatπ‘šξ“π‘—=1cπ‘—π›Όβˆ—π‘—=1.(31) Comparing (27) and (31), we can see that π›Όβˆ— is generally different from π›Όβˆ—.

For π΅πœŽπ‘“π‘’, we can also getπ΅πœŽπ‘“ξ“π‘’=π‘˜βˆˆβ„±π‘π‘˜π‘’π‘˜=ξ“π‘˜βˆˆβ„±π‘π‘˜π‘’π‘˜π‘0β‰œπ‘‘βˆ—π‘0.(32)

With (28) and (32), (24) can be expressed asΜ‡π‘₯(𝑑)=𝐴π‘₯(𝑑)+𝑏0𝐾1βˆ—βŠ€π‘₯(𝑑)+𝑏0πΎβˆ—2π‘Ÿ(𝑑)+𝑏0πΎβˆ—3+π‘‘βˆ—π‘0.(33) By choosing πΎβˆ—3=βˆ’π‘‘βˆ—, we have 𝑏0πΎβˆ—3=βˆ’π΅πœŽπ‘“π‘’, and (33) can be reduced toΜ‡π‘₯(𝑑)=𝐴π‘₯(𝑑)+𝑏0𝐾1βˆ—βŠ€π‘₯(𝑑)+𝑏0πΎβˆ—2π‘Ÿ=ξ€·(𝑑)𝐴+𝑏0𝐾1βˆ—βŠ€ξ€Έπ‘₯(𝑑)+𝑏0πΎβˆ—2π‘Ÿ(𝑑).(34) Therefore when failures are present, a nominal controller can always be found, with πΎβˆ—1 and πΎβˆ—2 specified in (6), π›Όβˆ— characterized in (27), and πΎβˆ—3=βˆ’π‘‘βˆ— ensures that the closed-loop system is stable under constant failures, and the state converges to the desired state π‘₯π‘š in (4) exponentially.

3.2. Adaptive Control Allocation Design
3.2.1. Adaptive Controller

Due to the uncertain nature of the failures, the controller parameters must be adjusted. Consider the adaptive control allocation scheme𝑣0(𝑑)=𝐾⊀1(𝑑)π‘₯(𝑑)+𝐾2(𝑑)π‘Ÿ(𝑑)+𝐾3(𝑑)β‰œπœƒβŠ€(𝑑)πœ”(𝑑),𝑣(𝑑)=𝛼(𝑑)𝑣0(𝑑),(35) where 𝐾1(𝑑), 𝐾2(𝑑), and 𝐾3(𝑑) are the estimates of πΎβˆ—1, πΎβˆ—2, and πΎβˆ—3 in (23). The signal πœ”(𝑑)=[π‘₯⊀(𝑑),π‘Ÿ(𝑑),1]⊀. With the adaptive controller in (35) and the plant dynamics in (22), the closed-loop system isξ€·Μ‡π‘₯(𝑑)=𝐴π‘₯(𝑑)+π΅πΌβˆ’πœŽπ‘“ξ€Έπ›ΌΓ—ξ€ΊπΎ(𝑑)⊀1(𝑑)π‘₯(𝑑)+𝐾2(𝑑)π‘Ÿ(𝑑)+𝐾3ξ€»(𝑑)+π΅πœŽπ‘“π‘’.(36)

3.2.2. Error Dynamics

For this adaptive control scheme design, we define 𝛼(𝑑)β‰œπ›Ό(𝑑)βˆ’π›Όβˆ—, and (36) becomesξ€·Μ‡π‘₯(𝑑)=𝐴π‘₯(𝑑)+π΅πΌβˆ’πœŽπ‘“ξ€Έπ›Ό(𝑑)𝑣0(𝑑)+π΅πœŽπ‘“π‘’ξ€·=𝐴π‘₯(𝑑)+π΅πΌβˆ’πœŽπ‘“ξ€Έξ‚π›Ό(𝑑)𝑣0ξ€·(𝑑)+π΅πΌβˆ’πœŽπ‘“ξ€Έπ›Όβˆ—π‘£0(𝑑)+π΅πœŽπ‘“π‘’ξ€·=𝐴π‘₯(𝑑)+π΅πΌβˆ’πœŽπ‘“ξ€Έξ‚π›Ό(𝑑)𝑣0ξ€·(𝑑)+π΅πΌβˆ’πœŽπ‘“ξ€Έπ›Όβˆ—π‘£βˆ—0ξ€·(𝑑)+π΅πΌβˆ’πœŽπ‘“ξ€Έπ›Όβˆ—Μƒπ‘£0(𝑑)+π΅πœŽπ‘“π‘’,(37) where ̃𝑣0(𝑑)=(πœƒ(𝑑)βˆ’πœƒβˆ—)βŠ€Μƒπœƒπœ”(𝑑)β‰œβŠ€(𝑑)πœ”(𝑑).

With (28), 𝐡(πΌβˆ’πœŽπ‘“)π›Όβˆ—π‘£βˆ—0(𝑑) in (37) becomesπ΅ξ€·πΌβˆ’πœŽπ‘“ξ€Έπ›Όβˆ—π‘£βˆ—0(𝑑)=𝑏0π‘£βˆ—0(𝑑)=𝑏0𝐾1βˆ—βŠ€π‘₯(𝑑)+𝑏0πΎβˆ—2π‘Ÿ(𝑑)+𝑏0πΎβˆ—3.(38) With (38), (37), and the plant model matching condition in (6), we haveΜ‡π‘₯(𝑑)=π΄π‘šπ‘₯(𝑑)+π΅π‘šξ€·π‘Ÿ(𝑑)+π΅πΌβˆ’πœŽπ‘“ξ€Έξ‚π›Ό(𝑑)𝑣0(𝑑)+𝑏0ΜƒπœƒβŠ€(𝑑)πœ”(𝑑).(39) From the closed-loop dynamics in (39) and the reference model in (4), we obtain the error dynamicṡ𝑒(𝑑)=π΄π‘šξ€·π‘’(𝑑)+π΅πΌβˆ’πœŽπ‘“ξ€Έξ‚π›Ό(𝑑)𝑣0(𝑑)+𝑏0ΜƒπœƒβŠ€(𝑑)πœ”(𝑑).(40)

3.2.3. Adaptive Laws

From the error dynamics in (40), the following adaptive laws are derived:Μ‡β€Œπœƒ(𝑑)=βˆ’Ξ“πœƒπœ”(𝑑)π‘’βŠ€(𝑑)𝑃𝑏0,(41)̇𝛼𝑗(𝑑)=βˆ’π›Ύπ‘—π‘’βŠ€(𝑑)𝑃𝑏𝑗𝑣0(𝑑),𝑗=1,2,…,π‘š,(42) where 𝑏𝑗 is the 𝑗th column of 𝐡, 𝛾𝑗>0 and Ξ“πœƒ=Ξ“βŠ€πœƒ>0 are adaptive gains.

The following theorem summarizes the properties of the adaptive control allocation scheme and the proof is provided in the Appendix:

Theorem 3. The adaptive control allocation scheme in (35) with the adaptive laws in (41) and (42) applied to the plant in (22) in the presence of constant failures guarantees that all closed-loop signals are bounded and limπ‘‘β†’βˆž(π‘₯(𝑑)βˆ’π‘₯π‘š(𝑑))=0.

3.3. Example
3.3.1. Linear Plant

Consider the longitudinal LTI model of the NASA GTM given by𝑉̇π‘₯=𝐴π‘₯+𝐡𝑣,π‘₯=𝑇,π›Όπ‘Žξ€»,π‘ž,πœƒβŠ€,𝛿𝑣=elob,𝛿elib,𝛿erob,𝛿eribξ€»βŠ€,(43) where the state includes the true airspeed 𝑉𝑇 (ft/s), angle of attack π›Όπ‘Ž (rad), pitch rate π‘ž (rad/s), and pitch angle πœƒ (rad). The control inputs are the deflections of the four elevator segments: left outboard elevator 𝛿elob, left inboard elevator 𝛿elib, right outboard elevator 𝛿erob, and right inboard elevator 𝛿erib (deg). The system matrices 𝐴 and 𝐡 are⎑⎒⎒⎒⎒⎒⎒⎣⎀βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯⎦,⎑⎒⎒⎒⎒⎒⎒⎣⎀βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯⎦.𝐴=βˆ’0.0450βˆ’8.96320.0349βˆ’32.1740βˆ’0.0035βˆ’2.74290.95140βˆ’0.0056βˆ’42.6233βˆ’3.561600010𝐡=βˆ’0.0110βˆ’0.0110βˆ’0.0110βˆ’0.0110βˆ’0.0012βˆ’0.0012βˆ’0.0012βˆ’0.0012βˆ’0.1962βˆ’0.1962βˆ’0.1962βˆ’0.19620000(44) Obviously, rank[𝐡]=1. For the simulation study, we will include the four elevator surfaces into one group, for which an elevator control signal will be designed and allocated.

Nominal Parameters
The nominal allocation gain π›Όβˆ— is chosen as π›Όβˆ—=[0.25,0.25,0.25,0.25]⊀ and 𝑏0=π΅π›Όβˆ—=[βˆ’0.011,βˆ’0.0012,βˆ’0.1962,0]⊀ so the deflection of the four segments will be the same if no failure occurs. The nominal controller is designed using the LQR approach for (𝐴,𝑏0). The resulting gains are πΎβˆ—1=[]0.1494,25.0761,βˆ’3.2841,βˆ’27.1293⊀,πΎβˆ—2=1.(45)

Reference Model
Similar to the simulation study in Section 2, the reference model is chosen as the closed-loop dynamics of the LQR controller, that is, π΄π‘š=𝐴+𝑏0𝐾1βˆ—βŠ€,π΅π‘š=𝑏0πΎβˆ—2.(46) The reference input to the reference model is chosen as π‘Ÿ(𝑑)=4.1841 for 𝑑β‰₯0, which leads to a reference trajectory with steady-state values π‘₯π‘š[](∞)=10,βˆ’0.0139,0,βˆ’0.0110⊀,(47) whose physical meaning is that the aircraft speed is increased by 10ft/s; its angle of attack is reduced by 0.0139 rad; and its pitch angle is reduced by 0.0110 rad.

Actuator Failure
The left outboard elevator is stuck at βˆ’5 degrees after 1 second, that is, 𝑒1(𝑑)=βˆ’5deg,for𝑑β‰₯𝑑𝑓second,(48) where 𝑑𝑓=1 second. The signal 𝑒1(𝑑) is the output of the left outboard elevator. For 𝑑>𝑑𝑓 the elevator does not respond to the elevator input 𝑣1(𝑑).

Simulation Results
The results are shown in Figures 15 and 16. From Figure 15, we can see that the plant state tracks the reference state after failure. Figure 16 shows the designed elevator signal, elevator inputs, and elevator outputs. The failure occurs at 𝑑𝑓=1 second and the other elevators can be seen to accommodate the failure with the adaptation.

3.3.2. NASA Generic Transport Model

Here we apply the adaptive control scheme to the nonlinear NASA GTM.

LTI Model for Controller Design
For adaptive control design, we consider a LTI model for a wings levelled flight having a 3 deg angle of attack at trim. Similar to the linear simulation study, we study the compensation for constant elevator failures and will consider multiple elevator failures in which both the lateral and longitudinal are active. The state and control vectors are given by [],𝛿π‘₯=𝑒,π‘ž,πœƒ,𝑝,π‘Ÿ,πœ™,πœ“π‘’=elob,𝛿elib,𝛿erob,𝛿erib,𝛿tl,𝛿tr,𝛿al,𝛿ar,𝛿ru,𝛿rlξ€»,(49) where 𝛿elob, 𝛿elib, 𝛿erob, and 𝛿erib are the four elevator segments, 𝛿tl and 𝛿tr are the two engine throttles, 𝛿al and 𝛿ar are left and right aileron segments, and 𝛿ru and 𝛿rl are upper and lower rudder segments. The actuators will be grouped into elevator, engine, aileron, and rudder groups. A virtual control signal is designed for each group and allocated to its members adaptively. The design is based on that in Section 3.2 and is similar to that in the linear simulation study. We include lateral states and lateral actuators to regulate the disturbance of the elevator failures propagated to the lateral dynamics.

Flight Conditions
A set of commands that aim to make the aircraft climb at 4-degree pitch angle is applied at 5 seconds. The nominal parameters and reference inputs are obtained as in the linear simulation example.

Actuator Failures
The right outboard elevator is locked at 10 degrees at 10 seconds and the right inboard elevator at 5 degrees at 20 seconds. The control objective is to maintain the climbing flight in the presence of these failures.
The simulation results are shown in Figures 17–23. Figure 17 shows the time history of the longitudinal states and reference model states. It can be seen that the tracking of the desired longitudinal attitude is accurately achieved.
The lateral states are shown in Figure 18. Some disturbances can be observed after the occurrence of asymmetric elevator failures. These disturbances are regulated by the ailerons and asymptotic tracking of the lateral states can be also achieved.
Figure 19 shows the flight trajectory, ground track, and altitude of the aircraft. From the ground track plot, we can see the effect of asymmetric failure on the lateral dynamics. The deviation is corrected by the lateral actuators. Figure 20 shows the lumped elevator signal, allocated elevator signals, and the actual elevator deflections. The elevator failures appear as constant (step signals) on the lower plot, and the designed and allocated elevator signals are shown to accommodate for the failures after their occurrence. The input signals and outputs of other actuators are shown in Figure 21. Note the actuation of the aileron and rudder required to compensate for the asymmetric failure. The time history of the control allocation gains and some selected adaptive parameters are shown in Figures 22 and 23. It can be seen that the allocation gains and controller parameters are updated autonomously to ensure the desired flight.

4. Conclusions

A novel adaptive control allocation framework is proposed herein. The adaptive allocation scheme includes an adaptive control signal and a control allocation unit with adaptively updated allocation gains. Two adaptive control allocation algorithms have been proposed for the compensation of uncertain failures. The proposed algorithms have been shown to guarantee closed-loop stability and asymptotic state tracking. It has also been shown that the proposed adaptive control allocation framework reduces the controller complexity with proper grouping of the actuators. In this framework, the control signal to be adaptively allocated can be actuated by nonadaptive controllers. The simulation results demonstrate the performance of the proposed algorithms and their applicability to aircraft flight control. Some future research topics in this direction include the extension of the adaptive control allocation framework to systems with multiple groups of actuators and the strict enforcement of a pure allocation, that is, exactly enforcing the designed control signal with βˆ‘π›Όπ‘–=1.


Proof of Theorem 1. When a failure occurs, the control effectiveness matrix Ξ› has a discontinuity. It would lead to a finite jump in 𝛼(𝑑) as well, which could result in a a finite jump in the error dynamics. Let us assume that there are 𝑙 failures occurring in the system, and the actuator failures occur at time instants π‘‘π‘˜, with π‘‘π‘˜<π‘‘π‘˜+1, π‘˜=1,2,…,𝑁. For the closed-loop stability and state tracking analysis, we choose the following Lyapunov-like function: 1𝑉(𝑑)=2π‘’βŠ€1(𝑑)𝑃𝑒(𝑑)+2ξ‚π›ΌβŠ€(𝑑)Ξ“π›Όβˆ’1Ξ›βˆ’11𝛼(𝑑)+2ΜƒπœƒβŠ€(𝑑)Ξ“πœƒβˆ’1Μƒπœƒ(𝑑),(A.1) for each time interval (π‘‘π‘˜,π‘‘π‘˜+1), π‘˜=0,1,…,𝑁 with 𝑑0=0 and 𝑑𝑁+1=∞. 𝑉(𝑑) is thus discontinuous with finite jumps at π‘‘π‘˜, π‘˜=1,…,𝑁. Taking the time derivative of 𝑉(𝑑) and substituting the adaptive laws in (11) and (12) into the result for each (π‘‘π‘˜,π‘‘π‘˜+1), we obtain  ̇𝑉(𝑑)=π‘’βŠ€(𝑑)𝑃̇𝑒(𝑑)+ξ‚π›ΌβŠ€(𝑑)Ξ“π›Όβˆ’1ΜƒπœƒΜ‡π›Ό(𝑑)+⊀(𝑑)Ξ“πœƒβˆ’1Μ‡πœƒ(𝑑)=π‘’βŠ€ξ‚ƒπ΄(𝑑)π‘ƒπ‘šπ‘’(𝑑)+𝑏0ΜƒπœƒβŠ€(𝑑)πœ”(𝑑)+𝐡𝛼(𝑑)𝑣0ξ‚„(𝑑)βˆ’ξ‚π›ΌβŠ€(𝑑)π΅βŠ€π‘ƒπ‘’(𝑑)𝑣0Μƒπœƒ(𝑑)βˆ’βŠ€(𝑑)πœ”(𝑑)π‘’βŠ€(𝑑)𝑃𝑏01=βˆ’2π‘’βŠ€π‘„π‘’β‰€0.(A.2) Thus we can conclude that for each (π‘‘π‘˜,π‘‘π‘˜+1), π‘˜=0,1,…,𝑁, 𝑉(𝑑) is bounded. Since 𝑉(𝑑) only has finite jumps at π‘‘π‘˜, π‘˜=1,…,𝑁, we can conclude that 𝑉(𝑑) is bounded for π‘‘βˆˆ[0,∞), and 𝑒(𝑑)∈𝐿∞, 𝛼(𝑑)∈𝐿∞, Μƒπœƒ(𝑑)∈𝐿∞, 𝛼(t)∈𝐿∞, πœƒ(𝑑)∈𝐿∞, π‘₯(𝑑)∈𝐿∞, and πœ”(𝑑)∈𝐿∞. Integrating both sides of (A.2), we can obtain 𝑉𝑑+π‘˜ξ€Έξ€·π‘‘βˆ’π‘‰βˆ’π‘˜+1ξ€Έ=12βˆ«π‘‘π‘˜+1π‘‘π‘˜π‘’βŠ€(𝜏)𝑄(𝜏)𝑒(𝜏)π‘‘πœ.(A.3) For 𝑁+1 intervals: [0,𝑑1), (𝑑1,𝑑2), …, (π‘‘π‘βˆ’1,𝑑𝑁), and (𝑑𝑁,∞), (A.3) holds. Summing both sides of (A.3) for π‘˜=0,1,…,𝑁, we obtain 12ξ€œβˆž0π‘’βŠ€ξ€·π‘‘(𝜏)𝑄(𝜏)𝑒(𝜏)π‘‘πœ=𝑉(0)βˆ’π‘‰βˆ’1𝑑+𝑉+1ξ€Έξ€·π‘‘βˆ’π‘‰βˆ’2𝑑+𝑉+2ξ€Έξ€·π‘‘βˆ’β‹―βˆ’π‘‰βˆ’π‘˜ξ€Έξ€·π‘‘+𝑉+π‘˜ξ€Έξ€·π‘‘βˆ’β‹―βˆ’π‘‰βˆ’π‘ξ€Έξ€·π‘‘+𝑉+π‘ξ€Έβˆ’π‘‰(∞)=𝑉(0)+𝑁𝑖=1𝑉𝑑+π‘–ξ€Έξ€·π‘‘βˆ’π‘‰βˆ’π‘–ξ€Έξ€»βˆ’π‘‰(∞)<∞,(A.4) because the jumps 𝑉(𝑑+𝑖)βˆ’π‘‰(π‘‘βˆ’π‘–) are finite and the number 𝑁 of jumps is also finite. Thus we have 𝑒(𝑑)∈𝐿2. We can also conclude from (10) that ̇𝑒(𝑑)∈𝐿∞. So from 𝑒(𝑑)∈𝐿2∩𝐿∞, and ̇𝑒(𝑑)∈𝐿∞, we have lim0β†’βˆžπ‘’(𝑑)=0.

Proof of Theorem 3. In Section 3.1, we have assumed that there are at most π‘β‰€π‘šβˆ’1 constant actuator failures and defined an index set for failed actuators as β„±={𝑖1,…,𝑖𝑝} such that πœŽπ‘˜=1 for all π‘˜βˆˆβ„±. Here we further assume that the failures occur at instants π‘‘π‘˜, with π‘‘π‘˜<π‘‘π‘˜+1, π‘˜=1,2,…,𝑁 with 1≀𝑁≀𝑝. The number of failure instants may be smaller than the total number of failures since multiple failures may happen at the same time. For the stability proof, we choose the following Lyapunov-like function 1𝑉(𝑑)=2π‘’βŠ€1(𝑑)𝑃𝑒(𝑑)+2ξ“π‘–βˆ‰β„±π›Ύπ‘–βˆ’1𝛼2𝑖1(𝑑)+2ΜƒπœƒβŠ€(𝑑)Ξ“πœƒβˆ’1Μƒπœƒ(𝑑),(A.5) for each time interval (π‘‘π‘˜,π‘‘π‘˜+1), π‘˜=0,1,…,𝑁, with 𝑑0=0 and 𝑑𝑁+1=∞. The time derivative in each time interval (π‘‘π‘˜,π‘‘π‘˜+1) is ̇𝑉(𝑑)=π‘’βŠ€ξ“(𝑑)𝑃̇𝑒(𝑑)+π‘–βˆ‰β„±π›Ύπ‘–βˆ’1𝛼𝑖(𝑑)Μ‡π›Όπ‘–Μƒπœƒ(𝑑)+⊀(𝑑)Ξ“πœƒβˆ’1Μ‡β€Œπœƒ(𝑑)=𝑒(𝑑)βŠ€π‘ƒπ΄π‘šπ‘’(𝑑)+π‘’βŠ€ξ€·(𝑑)π‘ƒπ΅πΌβˆ’πœŽπ‘“ξ€Έξ‚π›Ό(𝑑)𝑣0(𝑑)+π‘’βŠ€(𝑑)𝑃𝑏0ΜƒπœƒβŠ€(𝑑)πœ”(𝑑)βˆ’π‘’βŠ€(𝑑)𝑃𝑣0(𝑑)π‘–βˆ‰β„±ξ‚π›Όπ‘–(𝑑)π‘π‘–βˆ’ΜƒπœƒβŠ€(𝑑)πœ”(𝑑)π‘’βŠ€(𝑑)𝑃𝑏0=π‘’βŠ€(𝑑)π‘ƒπ΄π‘š1𝑒(𝑑)=βˆ’2π‘’βŠ€(𝑑)𝑄𝑒(𝑑)≀0(A.6) with the fact that 𝐡(πΌβˆ’πœŽπ‘“βˆ‘)𝛼(𝑑)=π‘–βˆ‰β„±ξ‚π›Όπ‘–(𝑑)𝑏𝑖. Following a similar approach to the stability analysis in Section 2, we can conclude that for any π‘‘βˆˆ[0,∞), 𝑉(𝑑)∈𝐿∞, 𝑒(𝑑)∈𝐿∞, π‘₯(𝑑)∈𝐿∞, πœ”(𝑑)∈𝐿∞, πœƒ(𝑑)∈𝐿∞, 𝑣0(𝑑)∈𝐿∞, 𝑒(𝑑)∈𝐿2. Since 𝑉(𝑑) only includes 𝛼𝑖(𝑑) with π‘–βˆ‰β„±, we can only conclude the boundedness of 𝛼𝑖(𝑑) and 𝛼𝑖(𝑑) for π‘–βˆ‰β„±. To show the boundedness of 𝛼𝑗(𝑑), π‘—βˆˆβ„±, we note that for any π‘—βˆˆβ„±π›Όπ‘—(𝑑)=𝛼𝑗(ξ€œ0)βˆ’π‘‘0π›Ύπ‘—π‘’βŠ€(𝜏)𝑃𝑣0(𝜏)π‘π‘—π‘‘πœ,(A.7) based on the adaptive law in (42). Note that all the columns of 𝐡 are linearly dependent based on the rank condition in Assumption 2. So we can have 𝑏𝑗=π‘βˆ—π‘—π‘π‘˜,βˆ€π‘—βˆˆβ„±,(A.8) where π‘π‘˜ is the column of 𝐡 that corresponds to an arbitrary healthy actuator, that is, π‘˜βˆ‰β„±, and π‘βˆ—π‘— is a nonzero constant. Equation (A.7) can thus be expressed as 𝛼𝑗(𝑑)=𝛼𝑗(ξ€œ0)βˆ’π‘‘0π›Ύπ‘—π‘’βŠ€(𝑑)𝑃𝑣0(𝑑)π‘βˆ—π‘—π‘π‘˜π‘‘π‘‘=𝛼𝑗𝛾(0)βˆ’π‘—π›Ύπ‘˜π‘βˆ—π‘—ξ€œπ‘‘0π›Ύπ‘˜π‘’βŠ€(𝑑)𝑃𝑣0(𝑑)π‘π‘˜π‘‘π‘‘=𝛼𝑗𝛾(0)+π‘—π›Ύπ‘˜π‘βˆ—π‘—ξ€œπ‘‘0ξ€·βˆ’π›Ύπ‘˜π‘’βŠ€(𝑑)𝑃𝑣0(𝑑)π‘π‘˜ξ€Έπ‘‘π‘‘=𝛼𝑗𝛾(0)+π‘—π›Ύπ‘˜π‘βˆ—π‘—ξ€œπ‘‘0Μ‡π›Όπ‘˜(𝑑)𝑑𝑑=𝛼𝑗𝛾(0)+π‘—π›Ύπ‘˜π‘βˆ—π‘—ξ€Ίπ›Όπ‘˜(𝑑)βˆ’π›Όπ‘˜ξ€»(0),βˆ€π‘—βˆˆβ„±.(A.9) Since π›Όπ‘˜(𝑑), π‘˜βˆ‰β„± has been proved to be bounded, we have 𝛼𝑗(𝑑)∈𝐿∞, for π‘—βˆˆβ„±.
We can further obtain that ̇𝑒(𝑑)∈𝐿∞ from (40). With 𝑒(𝑑)∈𝐿∞∩𝐿2 and ̇𝑒(𝑑)∈𝐿∞, we can have limπ‘‘β†’βˆžπ‘’(𝑑)=0.


This work was supported by the NRA NNX08AC62A of the IRAC project of NASA. The authors would like to thank Drs. Suresh M. Joshi and Sean P. Kenny at the NASA Langley Research Center for their valuable comments.


  1. M. W. Oppenheimer, D. B. Doman, and M. A. Bolender, β€œControl allocation for over-actuated systems,” in Proceedings of the 14th Mediterranean Conference on Control and Automation (MED '06), pp. 1–6, June 2006. View at: Publisher Site | Google Scholar
  2. J. M. Buffington and D. F. Enns, β€œLyapunov stability analysis of daisy chain control allocation,” Journal of Guidance, Control, and Dynamics, vol. 19, no. 6, pp. 1226–1230, 1996. View at: Google Scholar
  3. J. B. Davidson, F. J. Lallman, and W. T. Bundick, β€œReal-time adaptive control allocation applied to a high performance aircraft,” in Proceedings of the 5th SIAM Conference on Control and Its Applications, 2001. View at: Google Scholar
  4. J. D. BoΕ‘koviΔ‡ and R. K. Mehra, β€œControl allocation in overactuated aircraft under position and rate limiting,” in Proceedings of the American Control Conference, pp. 791–796, Anchorage, Alaska, USA, May 2002. View at: Google Scholar
  5. Y. Luo, A. Serrani, S. Yurkovich, M. W. Oppenheimer, and D. B. Doman, β€œModel-predictive dynamic control allocation scheme for reentry vehicles,” Journal of Guidance, Control, and Dynamics, vol. 30, no. 1, pp. 100–113, 2007. View at: Publisher Site | Google Scholar
  6. R. Venkataraman, M. Oppenheimer, and D. Doman, β€œA new control allocation method that accounts for effector dynamics,” in Proceedings of the IEEE Aerospace Conference Proceedings, pp. 2710–2715, March 2004. View at: Google Scholar
  7. F. Fahroo and D. Doman, β€œA direct method for approach and landing trajectory reshaping with failure effect estimation,” in Proceedings of AIAA Guidance, Navigation, and Control Conference (AIAA '04), pp. 200–209, August 2004. View at: Google Scholar
  8. M. A. Holender and D. B. Doman, β€œNonlinear control allocation using piecewise linear functions,” Journal of Guidance, Control, and Dynamics, vol. 27, no. 6, pp. 1017–1027, 2004. View at: Google Scholar
  9. D. B. Doman, B. J. Gamble, and A. D. Ngo, β€œQuantized control allocation of reaction control jets and aerodynamic control surfaces,” Journal of Guidance, Control, and Dynamics, vol. 32, no. 1, pp. 13–24, 2009. View at: Publisher Site | Google Scholar
  10. M. Marwaha and J. Valasek, β€œFault-tolerant control allocation for Mars entry vehicle using adaptive control,” International Journal of Adaptive Control and Signal Processing, vol. 25, no. 2, pp. 95–113, 2011. View at: Publisher Site | Google Scholar
  11. G. Tao, S. Chen, X. Tang, and S. M. Joshi, Adaptive Control of Systems with Actuator Failures, Springer, London, UK, 2004.
  12. L. G. Crespo, M. Matsutani, and A. M. Annaswamy, β€œDesign of a model reference adaptive controller for a remotely operated air vehicle,” AIAA Journal of Guidance, Control and Dynamics, vol. 35, no. 2, pp. 406–422, 2012. View at: Google Scholar
  13. Y. Liu, X. Tang, G. Tao, and S. M. Joshi, β€œAdaptive compensation of aircraft actuation failures using an engine differential model,” IEEE Transactions on Control Systems Technology, vol. 16, no. 5, pp. 971–982, 2008. View at: Publisher Site | Google Scholar
  14. G. Tao, Adaptive Control Design and Analysis, John Wiley & Sons, 2003.
  15. X. Tang, Y. Liu, and G. Tao, β€œA study of adaptation of multiple actuating signals for LTI systems,” in Proceedings of the American Control Conference, pp. 5996–6001, June 2006. View at: Google Scholar
  16. T. Jodan, W. Langford, C. Belcastro et al., β€œDevelopment of a dynamically scaled generic transport model testbed for flight research experiments,” in Proceedings of the AUVSIs Unmanned Systems North America Symposium and Exhibition, August 2004. View at: Google Scholar
  17. T. L. Jordan, W. M. Langford, and J. S. Hill, β€œAirborne-subscale transport aircraft research testbedβ€”aircraft model development,” in Proceedings of the AIAA Guidance, Navigation, and Control Conference, pp. 5767–5778, San Francisco, Calif, USA, August 2005. View at: Google Scholar

Copyright Β© 2012 Yu Liu and Luis G. Crespo. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

More related articles

 PDF Download Citation Citation
 Download other formatsMore
 Order printed copiesOrder

Related articles

Article of the Year Award: Outstanding research contributions of 2020, as selected by our Chief Editors. Read the winning articles.