Adaptive control in the presence of input constraints

Abstract

An adaptive control system is provided that scales both gain and commands to avoid input saturation. The input saturation occurs when a commanded input uc exceeds an achievable command limit of umax. To avoid input saturation, the commanded input uc is modified according to a factor μ.

Description

TECHNICAL FIELD

This invention relates generally to model reference adaptive control (MRAC) and more particularly to a direct adaptive control technology that adaptively changes both control gains and reference commands.

BACKGROUND

An adaptive model reference flight control loop is shown in FIG. 1. The flight control loop responds to a scalar reference input r provided by an external guidance system (not illustrated) or from guidance commands from a pilot. For example, an external guidance system may command a climb of a certain number of feet per second and scale input r accordingly. Aircraft 10 includes a set of actuators 15 that operate control surfaces on the aircraft in response to reference input r. In response to the actuation of the control surfaces, aircraft 10 will possess a certain state as measured by sensors 20, denoted by a state vector x. State vector x would include, for example, pitch, roll, and other standard guidance measurements. Similar measurements are provided by a reference model 50 for aircraft 10. Reference model 50 receives reference input r and generates a reference state x_ref for aircraft 10 that would be expected in response to the flight control loop receiving reference input r.

An adaptive control system receives an error signals representing the difference between state vector x and reference state x_ref and provides a linear feedback/feedforward input command u_lin. It may be shown that u_lin is a function of the sum of the product of a state gain k_x and the state vector x and the product of a reference gain k_r and the scalar reference input r. Responsive to the measurement of vector x and a scalar reference input r, a baseline or nominal controller 30 generates a nominal control signal 35, u_lin-nominal. Unlike the adaptive gains used to form u_lin, u_lin-nominal is the sum of the product of a static gain k_x0 and the state vector x and the product of a static gain k_r0 and the scalar reference input r. In that regard, a control loop topology could be constructed as entirely adaptive without any nominal control component such as baseline controller 30. However, the nominal control component will assist to “point in the right direction” such that the adaptive control may more quickly converge to a stable solution.

An input command u_c is formed from the summation of u_lin and u_lin-nominal. Responsive to the input command u_c, the appropriate control allocations amongst the various control surfaces are made in control allocation act 40 to provide commands to actuators 15. In turn, actuators 15 implement actual input command u. Under normal conditions, u and u_c should be very similar or identical. However, there are limits to what control surfaces can achieve. For example, an elevator or rudder may only be deflectable to a certain limit. These limits for the various control surfaces may be denoted by an input command saturation limit, u_max. Thus, u can not exceed u_max or be less than −u_max. If u_c exceeds u_max, u will be saturated at limit u_max.

Conventional linear control such as that shown in FIG. 1 is typically fairly robust to small modeling errors. However, linear control techniques are not intended to accommodate significant unanticipated errors such as those that would occur in the event of control failure and/or a change in the system configuration resulting from battle damage. A common characteristic of conventional adaptive control algorithms is that physical limitations are encountered such as actuator displacement and rate limits such that u becomes saturated at u_max or −u_max. This input saturation may lead to instability.

Accordingly, there is a need in the art for improved adaptive control techniques that explicitly accounts for and has the capability of completely avoiding input saturation.

SUMMARY

In accordance with an aspect of the invention, an adaptive control technique is provided in the presence of input constraints. For example, in an aircraft having actuators controlling control surfaces, the actuators may possess an input command saturation of u_max. Despite these limits, if the aircraft uses an adaptive or nominal control system, the control system may provide a linear feedback/feedforward commanded input of u_lin that may exceed u_max such that the actual command input realized by the actuators is saturated at u_max. The following acts avoid such input saturation: defining a positive input command limit u_max^δ equaling (u_max−δ), where 0<δ<u_max; defining a negative input command limit equaling −u_max^δ; if the absolute value of u_lin is less than or equal to u_max^δ, commanding the actuators with u_lin; if u_lin exceeds u_max^δ, commanding the actuators with a first command input that is a function of the sum of u_lin and a scaled version of u_max^δ; and if u_lin is less than −u_max^δ; commanding the actuators with a second command input that is a function of the difference of u_lin and a scaled version of u_max^δ.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram of a conventional model reference flight control loop.

FIG. 2 is a block diagram of an adaptive flight control loop having a baseline controller in accordance with an embodiment of the invention.

FIG. 3 is a block diagram of an entirely adaptive flight control loop in accordance with an embodiment of the invention.

FIG. 4 is a plot of achieved input u as a function of time for a prior art adaptive control loop and an adaptive control loop having a non-zero μ, factor in accordance with an embodiment of the invention.

FIGS. 5 a through 5d demonstrate tracking performance and input commands for various values of μ.

Embodiments of the present invention and their advantages are best understood by referring to the detailed description that follows. It should be appreciated that like reference numerals are used to identify like elements illustrated in one or more of the figures.

DETAILED DESCRIPTION

The present invention provides an adaptive control methodology that is stable in the sense of Lyapunov (theoretically proven stability), yet explicitly accounts for control constraints to completely avoid input saturation. This adaptive control methodology may be better understood with reference to the conventional flight control of FIG. 1. As discussed with respect to FIG. 1, aircraft 10 includes actuators 15 that respond to commanded inputs u_c with an actual or achieved input u as discussed previously. In response to actual input u, aircraft 10 achieves a state x as measured by sensors 20. Given the state x and actual input commands u, an equation for model system dynamics is as follows:{dot over (x)}(t)=Ax(t)+bλu(t),xεRⁿ,uεR where A is an unknown matrix, b is a known control direction, λ is an unknown positive constant, R is any real number, and Rⁿ is an n-dimensional vector.

Should there be no saturation of control surfaces, actual or achieved input commands u and the commanded input u_c are identical. However, a typical control surface can only achieve a certain amount of deflection. For example, a rudder or elevator may only be deflectable through a certain angle or limit, which may be denoted as u_max. Thus, should u_c exceed this limit, the actual input u will equal u_max. This relationship between u_c and u may be represented mathematically as:

$\begin{array}{c}u\left(t\right)=u_{\max }\mathrm{sat}\left(\frac{u_c}{u_{\max }}\right)\\ =\{\begin{array}{cc}{u}_c\left(t\right),& u_c\left(t\right)\le u_{\max }\\ u_{\max }\mathrm{sgn}\left(u_c\left(t\right)\right),& u_c\left(t\right)\ge u_{\max }\end{array}\end{array}$ where u_max is the saturation level. Based upon this relationship, the equation for the system dynamics may be rewritten as{dot over (x)}=Ax+bλ(u_c+Δu),Δu=u−u_c Even if u_c is limited to u_max to avoid input saturation, it will be appreciated that u may approach u_max too quickly such that undesired vibrations are incurred as u equals u_max. Accordingly, a new limit on actual command inputs is introduced as followsu_max^δ=u_max−δ, where: 0<δ<u_max A commanded control deficiency Δu_c between the commanded input u_c and the actual input may then be represented as

$\Delta u_c=u_{\max }^{\delta }\mathrm{sat}\left(\frac{u_c}{u_{\max }^{\delta }}\right)-u_c$ The present invention introduces a factor μ into the commanded input u_c as follows:

$u_c=\underset{\underset{u_{\mathrm{lin}}}{︸}}{k_x^{T}x+k_{r}r}+\mathrm{\mu \Delta }{u}_c$ where k_x and k_r are the gains for the actual state x and the reference state r, respectively. As discussed with respect to FIG. 1, u_lin may be entirely adaptive or possess a nominal component.Note that u_c is implicitly determined by the preceding two equations. It may be solved for explicitly as:

$\begin{array}{cc}\begin{array}{c}{u}_c=\frac{1}{1+\mu }\left(u_{\mathrm{lin}}+\mu u_{\max }^{\delta }\mathrm{sat}\left(\frac{u_{\mathrm{lin}}}{u_{\max }^{\delta }}\right)\right)\\ =\{\begin{array}{cc}{u}_{\mathrm{lin},}& u_{\mathrm{lin}}\le u_{\max }^{\delta }\\ \frac{1}{1+\mu }\left(u_{\mathrm{lin}}+\mu u_{\max }^{\delta }\right),& u_{\mathrm{lin}}>u_{\max }^{\delta }\\ \frac{1}{1+\mu }\left(u_{\mathrm{lin}}-\mu u_{\max }^{\delta }\right),& u_{\mathrm{lin}}<-u_{\max }^{\delta }\end{array}\end{array}& \mathrm{Eq}.\left(1\right)\end{array}$ It follows that u_c is continuous in time but not continuously differentiable.To assure Lyapunov stability, it is sufficient to choose the factor μ as follows:

$\mu >\frac{\left(\kappa +2\lambda \mathrm{Pb}\left(\Delta k_x^{\max }+k_x^{*}\right)\right)u_{\max }}{\mathrm{\kappa \delta }}+\frac{\left(\Delta k_r^{\max }+k_r^{*}\right)\kappa r_{\max }}{\mathrm{\kappa \delta }}-2$ where Δk_x^max,Δk_r^max are the maximum initial parameter errors, k_x*,k_r* are parameters that define the ideal control law for achieving the desired reference model for the given unknown system, and κ is a constant that depends upon the unknown system parameters,

Implementation of the factor μ within an adaptive flight control loop is shown in FIG. 2. A module 90 receives u_c that is formed from the addition of u_lin and u_lin-nominal as discussed with respect to FIG. 1. The factors u_max, μ and δ discussed with respect to Equation (1) may be provided by an external system or stored within memory (not illustrated). Given these factors, module 90 examines u_c and implements Equation (1) accordingly to provide a modified commanded input u_c. From factors u_max and δ, module 90 calculates u_max^δ so that u_c may be compared to u_max^δ. If the absolute value of u_c (or equivalently, u_lin) is less than or equal to u_max^δ, then module 90 provides u_c′ as being equal to u_c. If, however, u_c exceeds u_max^δ or is less than −u_max^δ, module 90 provides u_c′ as being equal to the corresponding value from Equation (1). It will be appreciated that module 90 may be implemented within hardware, software, or a combination of hardware and software. Moreover, as seen in FIG. 3, module 90 may be implemented within an entirely adaptive control loop that does not possess a baseline controller 30.

A graphical illustration of the effect of module 90 with respect to the achieved command u and the saturation limits u_max and −u_max is illustrated in FIG. 4. Consider the case if the factor μ equals zero. Examination of Equation (1) and FIGS. 2 and 3 shows that for such a value of μ, the modified input command u_c′ will simply equal u_c. As discussed in the background section, the achieved input command u will thus saturate at u_max as u_c exceeds u_max. Such an input saturation may cause dangerous instability. Conversely, if μ equals the minimum value required for Lyapunov stability as discussed above, the achieved input command u does not saturate, thereby eliminating input saturation effects. However, it will be appreciated from examination of Equation (1) that the factor μ may not simply be set to an arbitrarily high value much greater than 1. In such a case, the achieved control u is so damped with respect to the commanded input u_c that flight will also become unstable.

The tradeoffs with respect to various values of the factor μ may be demonstrated by the following simulation example. Suppose an unstable open loop system has the following system dynamics:

$\stackrel{.}{x}=\mathrm{ax}+{\mathrm{bu}}_{\max }\mathrm{sat}\left(\frac{u_c}{u_{\max }}\right),\mathrm{where}\text{:}a=0.5,b=2,u_{\max }=0.47$ $\mathrm{If}$ $\delta =0.2u_{\max }⟶\begin{array}{c}{u}_{\max }^{\delta }=u_{\max }-\delta \\ =0.8u_{\max }\end{array}$ ${\stackrel{.}{x}}_m=-6\left(x_m-r\left(t\right)\right)$ $r\left(t\right)=0.7\left(\sin \left(2t\right)+\sin \left(0.4t\right)\right)$ The resulting simulation data may be seen in FIGS. 5a through 5d. In FIG. 5a, the μ factor is set to zero such that u_c′ may exceed u_max. With μ equaling zero, it may be seen that modified commanded input u_c′ equals the conventional commanded input u_c discussed with respect to FIG. 1. In FIG. 5b, μ equals 1. This is still less than the minimum amount set by the Lyapunov conditions discussed earlier. Thus, u_c′ may again exceed u_max. In FIG. 5c, μ equals 10, which exceeds the minimally-required amount for Lyapunov stability so that u_c′ does not exceed u_max. Thus, the tracking performance is significantly improved. However, μ cannot simply be increased indefinitely. For example, as seen in FIG. 5d for a value of μ equals 100, the tracking performance has degraded significantly in that the modified commanded input u_c′ has become too damped in its response to changes in external conditions.

Those of ordinary skill in the art will appreciate that many modifications may be made to the embodiments described herein. Accordingly, although the invention has been described with respect to particular embodiments, this description is only an example of the invention's application and should not be taken as a limitation. Consequently, the scope of the invention is set forth in the following claims.

Claims

1. A system, comprising: actuators for controlling control surfaces, the actuators being responsive to an input command u, the actuators being constrained by an actuatable limit umax of the control surfaces such that if u exceeds umax or is less than −umax, input saturation occurs;an adaptive control loop responsive to a state x of the system and a reference input r to provide an input command ulin, wherein the system is configured to modify ulin to provide a modified input command uc′ to the actuators, the modification including the acts of:defining a positive input command limit umaxδ equaling (umax−δ), where 0<δ<umax;defining a negative input command limit equaling −umaxδ;if the absolute value of ulin is less than or equal to umaxδ, commanding the actuators with ulin;if ulin exceeds umaxδ, commanding the actuators with a first command input that is a function of the sum of ulin and a scaled version of umaxδ; andif ulin is less than −umaxδ; commanding the actuators with a second command input that is a function of the difference of ulin and a scaled version of umaxδ.

2. The system of claim 1, wherein ulin is a function of both an adaptive command input and a nominal command input.

3. The system of claim 2, wherein ulin is only a function of an adaptive command input.

4. The system of claim 1, wherein the modification further comprises: providing a scale factor μ, wherein the first input command equals (1/(1+μ))*(ulin+μumaxδ) and wherein the second input command equals ((1/(1+μ))*(ulin−μumaxδ).

5. The system of claim 4, wherein μ is greater than 1.

6. The system of claim 4, wherein μ is greater than 5.

7. The system of claim 4, wherein μ is greater than 100.

8. The system of claim 1, wherein the system is an aircraft.

9. The system of claim 8, wherein the aircraft is an unmanned aircraft.

10. The system of claim 8, further including an external guidance system configured to provide the reference input r.

11. The system of claim 1, wherein the system is a spacecraft.

12. An adaptive control system for an aircraft, the system comprising: control surfaces;actuators that operate the control surfaces, the actuators having an input command saturation of umax;a control loop including a module and providing a linear feedback/feedforward commanded input of ulin to the module, the module performing the acts of:defining a positive input command limit umaxδ to further limit the input command saturation umax, umaxδ equaling (umax−δ), where 0<δ<umax so that umaxδ is strictly less than umax;defining a negative input command limit −umaxδ equaling (−umax+δ);comparing the linear feedback/feedforward commanded input ulin to the positive input command limit umaxδ and the negative input command limit −umaxδ; andbased on the comparison: if ulin is greater than or equal to −umaxδ and is less than or equal to umaxδ, commanding the actuators with a first command input that is equal in value to the linear feedback/feedforward commanded input ulin;if ulin exceeds umaxδ, commanding the actuators with a second command input being equal to (1*ulin+μ*umaxδ)/(1+μ) where a scale factor μ is a positive number; andif ulin is less than −umaxδ; commanding the actuators with a third command input being equal to (1*ulin−μ*umaxδ)/(1+μ).

13. The adaptive control system of claim 12, wherein ulin is a function of both: an adaptive command input, wherein: an adaptive control subsystem receives an error signal, e, representing a difference between a state vector x describing the system and receives a reference state xref describing a reference model, and provides the adaptive command input; anda nominal command input ulin-nominal, wherein: a baseline controller receives the state vector x and a reference input r(t) and forms a sum of the product of a static gain kx0 and the state vector x and the product of a static gain kr0 and the reference input r(t) to provide the nominal command input, ulin-nominal; andthe linear feedback/feedforward commanded input ulin is the sum of the adaptive command input and the nominal command input, ulin-nominal.

14. The adaptive control system of claim 12, wherein ulin is only a function of an adaptive command input, wherein: an adaptive control subsystem receives an error signal, e, representing a difference between a state vector x describing the system and receives a reference state xref describing a reference model, and provides the adaptive command input; andthe linear feedback/feedforward commanded input ulin comprises the adaptive command input.

15. The adaptive control system of claim 12, further comprising: providing the scale factor μ, wherein μ is chosen sufficiently large to assure stability of the control system.

16. The adaptive control system of claim 12, wherein ulin is a function only of an adaptive command input.

17. The adaptive control system of claim 12, wherein the aircraft is an unmanned aircraft.

18. The adaptive control system of claim 12, wherein the aircraft is configured for space travel.

19. The adaptive control system of claim 12, wherein the aircraft is a manned aircraft.

20. The adaptive control system of claim 13, further including an external guidance system configured to provide the reference input r.