Residual mode filters

Abstract

Methods and systems for controlling a physical system (plant) are disclosed. The plant is modeled as a linear, finite-dimensional system having a state vector, a control input vector, a plant output vector, and a disturbance vector comprising disturbances having known basis functions and unknown amplitudes. An adaptive control law is used with separate adaptive gains for an error vector associated with the plant output vector, and the disturbance vector, plus a fixed gain for a disturbance estimator. The adaptive control law is operable to adjust the control input vector so as to minimize the error vector. The plant includes modes which are not Almost Strictly Positive Real (ASPR).

Claims

1. A physical system having an adaptive controller system for controlling the physical system, the physical system further comprising sensors and actuators; the physical system being nonlinear with unmodeled dynamics, and the adaptive controller system being configured to receive one or more output from the physical system, including indications of one or more nonlinear disturbances to the physical system, to analyze the output, and to cancel the one or more nonlinear disturbances without needing knowledge of their amplitude, wherein the physical system is modeled as a linear, finite dimensional system having at least one mode which is not Almost Strictly Positive Real (ASPR), wherein said at least one not ASPR mode includes one or modes associated with: a wind turbine train, a blade flap, or periodic interactions between turbine blades and the wind turbine drive train; and the model of the physical system further comprises: a state vector, a control input vector and a plant output vector; a disturbance vector comprising disturbances having known basis functions and unknown amplitudes and not dependent upon parameters that define the plant system, wherein the disturbance vector affects said at least one mode which is not ASPR; a disturbance estimator, associated with the disturbance vector, for estimating at least one of a present value and a future value of the disturbance vector; and wherein the adaptive controller system being configured to analyze the output and to cancel the one or more nonlinear disturbances includes configuring the adaptive controller system to execute an adaptive control law comprising: (i) a first adaptive gain for an error vector associated with the plant output vector; (ii) a second adaptive gain for the disturbance vector; and (iii) a fixed gain for the disturbance estimator, wherein the adaptive control law adjusts the control input vector, to cause the error vector to asymptotically approach a 0 value as time increases; (iv) a residual mode filter that causes an amplitude associated with the at least one mode that is not Almost Strictly Positive Real to asymptotically approach a zero value as time increases; (v) a first frequency-locked loop, operable to lock a first filter output to at least one frequency manifested by the disturbance vector; and (vi) a second frequency-locked loop and phase-locked loop, operable to lock a second filter output onto at least one frequency and at least one phase manifested by the residual mode filter.

2. The physical system of claim 1, wherein said error vector comprises deviations of said output plant vector from a vector of set points.

3. The physical system of claim 1, wherein said error vector comprises deviations from a desired model behavior.

4. The physical system of claim 1, wherein said physical system comprises a wind turbine having at least one turbine blade.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) FIG. 1 shows a schematic diagram of an adaptive control system according to embodiments of the instant invention.

(2) FIG. 2 shows a schematic diagram of an adaptive control system including a frequency lock loop. (Reference numbers 200-210 correspond with 100-110 of FIG. 1).

(3) FIG. 3 shows a flow chart of a frequency lock loop suitable for use with the adaptive control system of FIG. 2.

(4) FIG. 4 shows the actual and estimated wind speed for a series of step changes in wind speeds between the cut-in and rated wind speeds for an adaptive control system for a three-bladed horizontal axis wind turbine.

(5) FIG. 5 shows the response of the controlled tip speed ratio to the same series of step changes in wind speed shown in FIG. 4.

(6) FIG. 6 shows the generator rotational speed for the adaptive gain controller and for a fixed-gain controller for the same series of step changes in wind speed shown in FIG. 4.

DETAILED DESCRIPTION

(7) It must be noted that as used herein and in the claims, the singular forms a, and the include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to a structure includes two or more structures, and so forth. Embodiments are described for speed control for horizontal axis wind turbines, although it is understood that the methods disclosed herein apply to control of any non-linear system with or without disturbances.

(8) Where a range of values is provided, it is understood that each intervening value, to the tenth of the unit of the lower limit unless the context clearly dictates otherwise, between the upper and lower limit of that range, and any other stated or intervening value in that stated range, is encompassed within the invention. The upper and lower limits of these smaller ranges may independently be included in the smaller ranges, and are also encompassed within the invention, subject to any specifically excluded limit in the stated range. Where the stated range includes one or both of the limits, ranges excluding either or both of those included limits are also included in the invention. The terms about and approximately generally refers to 10% of a stated value.

Definitions

(9) As used herein, the term plant refers to any physical system modeled as a collection of interconnected components. Plants typically include integrated mechanical systems such as windmills, buildings, bridges, robotic machinery, aircraft, and ground transport vehicles.

(10) As used herein the term Almost Strictly Positive Real or ASPR refers to the behavior of a control system for a plant with a control system that can be described according to Eqn. (1), where CB is positive definite and the system transfer function from the plant output to the plant input is minimum phase; i.e.,

(11) $\begin{array}{cc}P\left(s\right)=\frac{\mathrm{CB}}{\mathrm{sI}-A}\mathrm{is}\mathrm{minimum}\mathrm{phase},& \left(9\right)\end{array}$
where s is the (complex) argument of the Laplace transform of the transfer function, and I is the identity matrix of dim(A).

(12) As used herein, the term frequency locked loop or FLL refers to a digital control loop in which the output frequency is locked to a reference frequency. An FLL is similar to a phase lock loop both computationally and functionally, but in an FLL, it is the frequency rather than the phase that is detected and locked onto.

(13) Control systems and methods for controlling physical systems are provided. The innovation comprises a control algorithm and method for synthesis of the controller. The control algorithm adapts its gains to the plant output, making it ideally suited to plants that have unknown modeling parameters and operate in uncertain environments. Knowledge of the plant matrices A, B, C, , , F is not required. Applications that require tracking a reference point or following a reference model can be easily solved with the control algorithm. The approach allows for rejection of persistent disturbances to the plant. Knowledge of the amplitude of the disturbance is unnecessary, only the waveform or a superposition of waveforms need be known. Also disclosed is a synthesis method to design the controller to restore minimum phase to the plant. The synthesized controller operates such that asymptotic tracking is achieved with bounded adaptive gains.

(14) The control algorithm described herein is able to perform model tracking or regulation to a set point with little knowledge of the plant dynamics or the operating environment. Many control problems can benefit from the adaptive control algorithm described here, which is well-suited to nonlinear applications that have unknown modeling parameters and poorly known operating conditions. Disturbance accommodation is a critical component of many systems. By using adaptive feedback control with disturbance accommodation, plant performance can be improved, robustness to parameter or unmodeled dynamic variations can be provided, and reliability can be increased. Often the form of a disturbance is known, but the amplitude is unknown. For instance, a motor operating on a structure used for accurate pointing would cause a sinusoidal disturbance of a known frequency content. The algorithm described here is able to accurately cancel these disturbances, without needing knowledge of their amplitude.

(15) Many applications require filters to remove modal frequencies that might interact with the control system or drive the system's resonant frequencies to instability. Filters such as the low-pass filter and the notch filter are common examples used in practice. These filters have limitations that include phase shift and time delay of signals. The algorithm synthesis method described here allows for the creation of a filter that has a more accurate model of the modal systems to be removed from the feedback loop to the controller without causing phase shift or time delays.

(16) The methods and systems disclosed herein facilitate control of the following example of nonlinear plants that are difficult and expensive to model: plants with unmodeled dynamics or unknown operating conditions or parameters that vary, plants needing filters to remove certain frequencies without affecting signal phase or adding time delay, plants needing disturbance accommodation for disturbances with known waveform but unknown amplitude, and plants needing disturbance tracking.

(17) In embodiments of the present invention, the control objective is to cause the output y.sub.p of a plant to asymptotically track zero while accommodating disturbances of the form given by the disturbance generator. Define an error vector e.sub.y as
e.sub.yy.sub.pO(10)
The desired control objective is e.sub.y tends asymptotically toward zero:

(18) $\begin{array}{cc}\underset{t->}{\lim }{e}_y->0& \left(11\right)\end{array}$

(19) Consider a plant described by Eqn. (1) with the disturbance generator given by Eqn. (3). Define an adaptive control law of the form
u=G.sub.ee.sub.y+G.sub.D.sub.D(12)
where G.sub.e and G.sub.D are matrices.

(20) Adaptive gain laws can be written as

(21) 0$\begin{array}{cc}\left\{\begin{array}{c}{\stackrel{.}{G}}_e=-e_y{e}_y^T{}_e\\ {\stackrel{.}{G}}_D=-e_y{}_D^T{}_D\end{array}\right\}& \left(13\right)\end{array}$
where .sub.e and .sub.D are positive definite constant matrices. The adaptive controller is specified by (12) with adaptive gain laws (13). Frost et al. (Direct adaptive control of a utility-scale wind turbine for speed regulation, Int. J. Robust Nonlinear Control, 19, 59-71, 2009, incorporated herein by reference) proved that if a plant satisfies the ASPR condition, has bounded disturbances .sub.D, and can be stabilized with adaptive controller and gains given by (12) and (13) then the adaptive gains G.sub.e and G.sub.D remain bounded and asymptotic tracking occurs in accordance with (11).

(22) In some cases, the plant described by Eqn. (1) does not satisfy the requirement of Almost Strictly Positive Realness (ASPR). For example, an unmodeled modal subsystem of the full nonlinear plant can contribute to a non-ASPR condition of a plant operating with feedback control. There may be a modal subsystem comprising Q modes that inhibits ASPR. In embodiments of the present invention, the adaptive controller is modified with a Residual Mode Filter (RMF) to compensate for this modal subsystem. The overall process is shown schematically in FIG. 1. The plant 100 is partitioned into retained modes 102 that are to be controlled and stabilized and which are ASPR, and residual modes 104 which are not ASPR. An Adaptive Controller 106 includes a Disturbance Estimator 108 and a Residual Mode Filter 110.

(23) Assume that (1) can be partitioned into modal form:

(24) $\begin{array}{cc}\left\{\begin{array}{c}\left[\begin{array}{c}\stackrel{.}{x}\\ {\stackrel{.}{x}}_Q\end{array}\right]=\underset{\underset{A_p}{}}{\left[\begin{array}{cc}A& 0\\ 0& A_Q\end{array}\right]}\left[\begin{array}{c}x\\ x_Q\end{array}\right]+\underset{\underset{B_p}{}}{\left[\begin{array}{c}B\\ B_Q\end{array}\right]}u+\underset{\underset{{}_p}{}}{\left[\begin{array}{c}\\ {}_Q\end{array}\right]}{u}_D\\ y_p=\underset{\underset{C_p}{}}{\left[\begin{array}{cc}C& C_Q\end{array}\right]}\left[\begin{array}{c}x\\ x_Q\end{array}\right]=\underset{\underset{y}{}}{\mathrm{Cx}}+\underset{\underset{y_Q}{}}{C_Q{x}_Q}\end{array}\right\}& \left(14\right)\end{array}$
Now define

(25) $\begin{array}{cc}{x}_p\left[\begin{array}{c}x\\ x_Q\end{array}\right];A_p\left[\begin{array}{cc}A& 0\\ 0& A_Q\end{array}\right];B_p\left[\begin{array}{c}B\\ B_Q\end{array}\right];{}_p\left[\begin{array}{c}\\ {}_Q\end{array}\right];C_p\left[\begin{array}{cc}C& C_Q\end{array}\right],& \left(15\right)\end{array}$
A disturbance generator is defined as above in (2) and (3) to model the disturbances entering the plant. The control objective remains the same, i.e., error asymptotically approaching zero (Eqn. (11)).

(26) Now it is possible to assume only that the subsystem (A, B, C) is ASPR rather than the full unpartitioned plant (A.sub.p, B.sub.p, C.sub.p). The modal subsystem (A.sub.Q, B.sub.Q, C.sub.Q) is assumed known and stable. The modal subsystem may be directly affected by the disturbance input. (Recall that ASPR means the matrix product C B is positive definite and the system transfer function is minimum phase.) Suppose the actual plant has an ASPR subsystem plus a known modal subsystem that is stable but inhibits ASPR for the full plant. Then to achieve stable control, the non-ASPR modal subsystem can be compensated or filtered away.

(27) The Residual Mode Filter (RMF) can be defined by

(28) $\begin{array}{cc}\left\{\begin{array}{c}{\stackrel{\stackrel{.}{^}}{x}}_Q=A_Q{\hat{x}}_Q+B_{Q}u+{}_Q{\hat{u}}_D\\ {\hat{y}}_Q=C_Q{\hat{x}}_Q\end{array}\right\}& \left(16\right)\end{array}$
where the disturbance estimator is given by

(29) $\begin{array}{cc}\left\{\begin{array}{c}{\hat{u}}_D={\hat{z}}_D\\ {\stackrel{\stackrel{.}{^}}{z}}_D=F{\hat{z}}_D+K_D{\tilde{e}}_y\end{array}\right\}& \left(17\right)\end{array}$
The compensated tracking error is given by
{tilde over (e)}.sub.yy.sub.p.sub.Q(18)
Note that the Disturbance Estimator .sub.D depends only on the disturbance waveform ( and ) and is independent of the plant (A, B, C).

(30) Now define
e.sub.Q{circumflex over (x)}.sub.Qx.sub.Q; e.sub.D{circumflex over (z)}.sub.Dz.sub.D(19)
The time derivative of (19) can be written using (2) and (14)-(18) as

(31) $\begin{array}{cc}\left\{\begin{array}{c}{\stackrel{.}{e}}_Q=A_Q{e}_Q+{}_Q{e}_D\\ {\stackrel{.}{e}}_D=K_D\left(\mathrm{Cx}-C_Q{e}_Q\right)+{\mathrm{Fe}}_D\end{array}\right\}& \left(20\right)\end{array}$
The compensated tracking error {tilde over (e)}.sub.y can be rewritten in terms of e.sub.Q:
{tilde over (e)}.sub.y=CxC.sub.Qe.sub.Q(21)

(32) Now define the ideal trajectories:

(33) $\begin{array}{cc}\left\{\begin{array}{c}{\stackrel{.}{x}}_{*}={\mathrm{Ax}}_{*}+{\mathrm{Bu}}_{*}+u_D\\ {\stackrel{.}{x}}_Q^{*}=A_Q{x}_Q^{*}+B_Q{u}_{*}+{}_Q{u}_D\\ |y_{*}={\mathrm{Cx}}_{*}=0\\ y_Q^{*}=C_Q{x}_Q^{*}\end{array}\right\}& \left(22\right)\end{array}$
where

(34) $\begin{array}{cc}\left\{\begin{array}{c}{x}_{*}{S}_1{z}_D\\ x_Q^{*}{S}_Q{z}_D\\ u_{*}{S}_2{z}_D\end{array}\right\}& \left(23\right)\end{array}$
This is equivalent to the matching conditions

(35) $\begin{array}{cc}\left\{\begin{array}{c}{\stackrel{\_}{S}}_{1}F=A_p{\stackrel{\_}{S}}_1+B_p{S}_2+{}_p\\ {\mathrm{CS}}_1=0\end{array}\right\}& \left(24\right)\end{array}$
where

(36) $\begin{array}{cc}{\stackrel{\_}{S}}_1\left[\begin{array}{c}{S}_1\\ S_Q\end{array}\right];\left(A_p,B_p,{}_p\right)\mathrm{from}\left(14\right)\mathrm{and}\left(15\right)& \left(25\right)\end{array}$
These matching conditions are known to be uniquely solvable when C.sub.pB.sub.p=C B is nonsingular.

(37) It is not necessary to know the actual solutions to the matching conditions in the adaptive control approach. Let

(38) 0$\begin{array}{cc}\left\{\begin{array}{c}xx-x;x_Q{x}_Q-x_Q^{*}\\ uu-u_x;yy-y_{*}\\ y_Q{y}_Q-y_Q^{*};y_p{y}_p-y_{*}\end{array}\right\}& \left(26\right)\end{array}$
Using (14), (16), (18), (19), and (22) one obtains

(39) $\begin{array}{cc}\left\{\begin{array}{c}y=Cx\\ y_Q=C_Q{x}_Q\\ y_p=Cx+y_Q\\ {\tilde{e}}_y=Cx-C_Q{e}_Q\end{array}\right\}& \left(27\right)\end{array}$
(Note that, per (22), y.sub.*=x.sub.*=0, and x=x.) Now let

(40) $\begin{array}{cc}e\left[\begin{array}{c}{e}_Q\\ e_D\end{array}\right];{\stackrel{\_}{C}}_Q\left[\begin{array}{cc}{C}_Q& 0\end{array}\right];{\stackrel{\_}{K}}_D\left[\begin{array}{c}0\\ K_D\end{array}\right];{\stackrel{\_}{A}}_e\left[\begin{array}{cc}{A}_Q& {}_Q\\ -K_D{C}_Q& F\end{array}\right]& \left(28\right)\end{array}$
Then with (20) one obtains

(41) $\begin{array}{cc}\left\{\begin{array}{c}\stackrel{.}{x}=Ax+Bu\\ {\stackrel{.}{x}}_Q=A_Q{x}_Q+B_{Q}u\\ \stackrel{.}{e}={\stackrel{\_}{A}}_{e}e+{\stackrel{\_}{K}}_{D}Cx\\ {\tilde{e}}_y=Cx-{\stackrel{\_}{C}}_{Q}e\end{array}\right\}.& \left(29\right)\end{array}$

(42) Now, let

(43) $\begin{array}{cc}=\left[\begin{array}{c}x\\ x_Q\\ e\end{array}\right]& \left(30\right)\end{array}$
Then (29) can be rewritten as

(44) $\begin{array}{cc}\left\{\begin{array}{c}\stackrel{.}{}=\underset{\underset{\stackrel{\_}{A}}{}}{\left[\begin{array}{ccc}A& 0& 0\\ 0& A_Q& 0\\ {\stackrel{\_}{K}}_{D}C& 0& {\stackrel{\_}{A}}_e\end{array}\right]}+\underset{\underset{\stackrel{\_}{B}}{}}{\left[\begin{array}{c}B\\ B_Q\\ 0\end{array}\right]}u\\ {\stackrel{.Math.}{e}}_y=\underset{\underset{\stackrel{\_}{C}}{}}{\left[\begin{array}{ccc}C& 0& -{\stackrel{\_}{C}}_Q\end{array}\right]}\end{array}\right\}& \left(31\right)\end{array}$

(45) (, B, C) is ASPR if and only if the matrix product C B is positive definite and H(s) P(s) is minimum phase where

(46) $\begin{array}{cc}\left\{\begin{array}{c}P\left(s\right)\frac{\mathrm{CB}}{\mathrm{sI}-A}\\ H\left(s\right)I-\frac{{\stackrel{\_}{C}}_Q{\stackrel{\_}{K}}_D}{\mathrm{sI}-{\stackrel{\_}{A}}_e}\end{array}\right\}& \left(32\right)\end{array}$
The proof of this lemma can be found in Balas and Frost (Adaptive Control of Linear Modal Systems using Residual Mode Filters and a Simple Disturbance Estimator, Proc. Amer. Control Conf. 2338-43, 2011 which is incorporated herein by reference). From this lemma, there exists a G*.sub.e such that (.sub.C, B, C) is Strictly Positive Real (SPR) where .sub.C+BG*.sub.eC. Then, from the well-known Kalman-Yacubovic Theorem, there exists P, Q such that

(47) $\begin{array}{cc}\left\{\begin{array}{c}{\stackrel{\_}{A}}_C^{T}P+\stackrel{\_}{P}{\stackrel{\_}{A}}_C=-\stackrel{\_}{Q}\\ \stackrel{\_}{P}\stackrel{\_}{B}={\stackrel{\_}{C}}^T\end{array}\right\}& \left(33\right)\end{array}$

(48) The complete adaptive control law with a residual mode filter and a disturbance estimator is given by

(49) $\begin{array}{cc}\left\{\begin{array}{c}u{G}_e{\stackrel{\_}{e}}_y+G_D{}_D\\ {\tilde{e}}_y{y}_p-{\hat{y}}_Q\\ \stackrel{\stackrel{.}{^}}{x}=A_Q{\hat{x}}_Q+B_{Q}u+{}_Q{\hat{u}}_D\\ {\hat{y}}_Q=C_Q{\hat{x}}_Q\\ {\hat{u}}_D={\hat{z}}_D\\ {\stackrel{\stackrel{.}{^}}{z}}_D=F{\hat{Z}}_D+K_D{\tilde{e}}_y\end{array}\right\}& \left(34\right)\end{array}$
with the adaptive gains:

(50) $\begin{array}{cc}\left\{\begin{array}{c}{\stackrel{.}{G}}_e=-{\tilde{e}}_y{\tilde{e}}_y^T{}_e;{}_e>0\\ {\tilde{G}}_D=-{\tilde{e}}_y{}_D^T{}_D;{}_D>0\end{array}\right\}& \left(35\right)\end{array}$

(51) The following theorem shows that stable control is possible. In (14), assume

(52) a) (A.sub.Q, B.sub.Q, C.sub.Q) is stable and (F, ) is known

(53) b) C B is positive definite; H(s) P(s) is minimum phase where

(54) 0$\begin{array}{cc}\left\{\begin{array}{c}P\left(s\right)\frac{\mathrm{CB}}{\mathrm{sI}-A}\\ H\left(s\right)I-\frac{{\stackrel{\_}{C}}_Q{\stackrel{\_}{K}}_D}{\mathrm{sI}-{\stackrel{\_}{A}}_c}\\ {\stackrel{\_}{C}}_Q\left[\begin{array}{cc}{C}_Q& 0\end{array}\right];{\stackrel{\_}{K}}_D\left[\begin{array}{c}0\\ K_D\end{array}\right];{\stackrel{\_}{A}}_e\left[\begin{array}{cc}{A}_Q& {}_Q\\ -K_D{C}_Q& F\end{array}\right]\end{array}\right\}& \left(36\right)\end{array}$

(55) c) .sub.D is bounded.

(56) Then, the adaptive controller with RMF and disturbance estimator defined in (34) and (35) gives

(57) $\begin{array}{cc}\underset{t->}{\lim }->0& \left(37\right)\end{array}$
where

(58) $\begin{array}{cc}\left[\begin{array}{c}x\\ x_Q\\ e\end{array}\right]& \left(38\right)\end{array}$
with bounded adaptive gains (G.sub.e, G.sub.D). The proof of this theorem is given in the Appendix of Balas and Frost (op cit).

(59) From the result of the above theorem,

(60) $\begin{array}{cc}\left\{\begin{array}{c}\underset{t->}{\lim }{\tilde{e}}_y=\underset{t->}{\lim }\stackrel{\_}{C}->0\\ \underset{t->}{\lim }y=\underset{t->}{\lim }Cx->0\\ \underset{t->}{\lim }e->0\end{array}\right\}& \left(39\right)\end{array}$

(61) On the other hand, considering the entire plant y.sub.p, one obtains

(62) $\begin{array}{cc}\begin{array}{c}{y}_p=y_p-y_{*}\\ =y+y_Q\\ =Cx+y_Q\\ =Cx+C_Q{x}_Q+C_Q{x}_Q^{*}\end{array}\underset{t->}{\lim }{y}_p->y_Q^{*};\left(y_Q^{*}{C}_Q{x}_Q^{*}\right)& \left(40\right)\end{array}$
Note that y*.sub.Q is not necessarily zero. However, the adaptive control algorithm described herein can be satisfactory for many plants, depending on the properties of y*.sub.Q. If it is desired that

(63) $\underset{t->}{\lim }{y}_p->0,$
then the constraint y*.sub.Q=0 must be added to (22), resulting in C.sub.QS.sub.Q=0 being added to the matching conditions (24).

(64) In some embodiments, an exact representation of the frequency of the non-ASPR modes is not known. An approximate representation of the non-ASPR mode can be determined with an adjustable frequency parameter. The frequency can then be determined while the system is being controlled. A frequency-locked loop (FLL) 212 is provided such that the Residual Mode Filter (RMF) can acquire and lock onto the frequency of the non-ASPR mode as shown in FIG. 2. The input to the FLL is the bandpass filtered plant output y.sub.BP. The output of the plant is bandpass filtered so that the signal content near the modeled frequency is retained, while the remainder (especially the high frequency content) is discarded.

(65) A schematic diagram of the FLL is shown in FIG. 3. The first part of the FFL is the frequency detector. It uses a well established method for generating sinusoidal comparison signals (often referred to as the Synchronized Oscillator 302) at the current FLL frequency estimate, .sub.FLL. The initial frequency estimate is the modeled frequency .sub.s. Next, the input signal y.sub.BP is multiplied (304) by the generated sine and cosine signals. The results are passed through low pass filters 306 to derive two remainder terms, r.sub.1 and r.sub.2. The inverse tangent 308 of the quotient of the two remainder terms yields a signal that is proportional to the frequency error. This signal then goes to the loop filter 310 where it is multiplied by a loop gain, G(s)/s, then summed with the initial frequency estimate to obtain an updated frequency estimate .sub.FLL, which is the output of the FLL. This output is also used for the next cycle of the FLL in the Synchronized Oscillator 302 and to update the bandpass filter used to filter the plant output. The frequency estimate .sub.FLL is used to update the frequency of the non-ASPR mode modeled in the RMF 210. An FLL can also be used for a disturbance estimator that has basis functions at varying frequencies. The residual mode filter (RMF), provided here as part of an adaptive control law, is applied to at least one non-ASPR mode so that this mode asymptotically approaches a zero value as time increases. Use of an FLL, without more, may not provide the synchronization of frequency and phase required to achieve the desired asymptotic behavior. It is preferable here to use a combination of an FLL and a PLL to analyze a signal associated with the RMF, including identification of, or locking onto, at least one frequency and at least one phase manifested by the RMF.

Advantages and Applications

(66) Embodiments of the present invention provide methods for control synthesis that are relatively simple, allowing for easier implementation and verification. The algorithms can handle nonlinear systems with unmodeled dynamics, and are robust to uncertain operating environments. The methods provide disturbance accommodation that can accurately reject disturbances of known waveform, but unknown amplitude. Methods for synthesizing a bank of residual mode filters are provided that can exactly remove the component of a plant output that causes the system to become minimum phase, or put energy into resonant modes that would become excited, leading to system instability.

(67) This technology has many potential applications. Any system having compliance and nonlinearities could benefit. In particular, the control of systems with compliant components or flexible structures is ideally suited to the approach described here. With the use of a residual mode filter, it is possible to track a model with fewer dimensions than the actual plant. Other adaptive control methods generally require the plant and tracking model to have the same dimension. RMFs can be designed independently from the controller. For example, it is possible to modify and add a bank of RMFs without any controller redesign. Conversely, it is also possible to modify a base controller while retaining the function of a tested RMF.

(68) Control systems for wind turbines are useful applications of adaptive controllers with RMFs. The main nonlinearities in a wind turbine model come from the nonlinear aerodynamic loads on the turbine. Creating an accurate model of the dynamic characteristics of a wind turbine is expensive and extremely difficult, if not impossible. Additionally, wind turbines operate in highly turbulent and unpredictable conditions. These complex aspects of wind turbines make them attractive candidates for the application of adaptive control methods. It has been demonstrated that the uniform wind disturbance, without shear, across the rotor disk of a turbine can be accurately accounted for when modeled as a step disturbance of unknown amplitude. Rejection of step disturbances can be implemented in the controller to improve performance and to reduce destructive loads.

(69) The auto industry can use the algorithms in many of their component systems and in the drive system. Other applications include heavy machinery, precise systems, such as surgical or video tracking systems, control of large cranes such as those used to load and unload container ships, control of support structures for space vehicle solar panel arrays, and even financial modeling and control. Yet other applications include automated assembly lines for manufacturing, automated handling and distribution (e.g., on-line retailer with automated order filling and shipping system), as well as other robotic applications.

EXAMPLES

Example 1: Low-Wind-Speed Operation of a Three-Bladed Horizontal-Axis Wind Turbine

(70) The adaptive control with Residual Mode Filter was implemented for a 5 MW offshore wind turbine model. The wind turbine had a rotor radius of 63 m and 5 MW of rated power. The cut-in, rated and cut-out wind speeds were 3 m/s, 11.4 m/s and 25 m/s respectively. Cut-in and rated rotor speeds are 6.9 m/s and 12.1 m/s respectively. Rated tip speed was 80 m/s with a tip speed ratio (ratio of blade tip speed to wind speed) of 7.55.

(71) The model included eleven switches for various degrees of freedom which could be switched on and off to add complexity. When all degrees of freedom were switched on, the model had 31 states. To design the wind speed and partial state estimation of the wind turbine, the turbine was linearized at a constant wind speed of 8 m/s with blade pitch held at 0. During the linearization, the drive train and generator degrees of freedom switches were turned on giving four states with two states due to the generator and two states due to the drive train. The first generator state (generator azimuth position) was removed to get a three state model of the wind turbine. This three state model was then augmented with a wind disturbance model (the disturbance generator) given by Eqn. (2). The adaptive gain controller (with Disturbance Estimator) of Eqn. (34) and (35) was implemented with G.sub.D=15,000 and G.sub.e=0.0001.

(72) FIG. 4 shows the actual and estimated wind speed for a series of step changes in wind speeds between the cut-in and rated wind speeds (i.e., operational wind speeds below the rated speed, an operation regime generally referred to as Region II.) The Disturbance Estimator is seen to estimate the wind speed with good accuracy.

(73) Maximum power output in Region II is achieved for a constant tip speed ratio.

(74) FIG. 5 shows the response of the controlled tip speed ratio to the same series of step changes in wind speed shown in FIG. 4. Results are shown for a conventional fixed-gain controller and for the adaptive controller. As can be seen in FIG. 5, the adaptive controller was significantly more effective at adapting to the wind speed than the fixed-gain controller.

(75) The generator rotational speed is shown in FIG. 6 for the adaptive controller and for a fixed-gain controller for the same series of step changes in wind speed shown in FIG. 4. At the lower wind speeds, there is behavior difference between the two controllers, because in the fixed-gain controller, the generator torque is directly a function of generator speed (=K.sup.2, where K is some constant depending on the desired tip speed ratio, and is the generator rotational speed). But in the adaptive disturbance-tracking control, there is no direct relation between the generator rotational speed and torque; instead, the generator torque is adjusted so that the generator rotational speed is proportional to the wind speed.

Example 2: Robotic Surgical Systems

(76) Robotic surgical systems are used to operate on a patient's body, and can provide very fine control over movements or access that would be difficult or impossible to practice by hand. However, as the patient is alive during the surgery, the body is generating rhythmic movements due to breathing or heart beat that can generate disturbances that could propagate to the surgical system. By utilizing the control systems and methods described herein, the surgical system can be made safer and more reliable.

Example 3: Vehicle Control Systems

(77) There are a variety of control systems implemented in cars, trucks, trains, and the like to control speed, braking, suspension, steering, and other operating features. There are also various sources of persistent or periodic disturbances such as tire irregularities, road surface structure (e.g., regularly spaced seams in concrete road surface or bridge roadways), engine vibrations, and the like that can prevent a simple controller from functioning adequately. Improved control system performance can be achieved by using the adaptive disturbance accommodating controllers described herein to filter out the disturbances.

(78) It will be understood that the descriptions of one or more embodiments of the present invention do not limit the various alternative, modified and equivalent embodiments which may be included within the spirit and scope of the present invention as defined by the appended claims. Furthermore, in the detailed description above, numerous specific details are set forth to provide an understanding of various embodiments of the present invention. However, one or more embodiments of the present invention may be practiced without these specific details. In other instances, well known methods, procedures, and components have not been described in detail so as not to unnecessarily obscure aspects of the present embodiments.