Mario A. Santillo

ADAPTIVE CONTROL BASED ON RETROSPECTIVE COST OPTIMIZATION

U NLIKE robust control, which chooses control gains based on a prior, fixed level of modeling uncertainty, adaptive control algorithms tune the feedback gains in response to the true plant and exogenous signals: that is, commands and disturbances. Generally speaking, adaptive controllers require less prior modeling information than robust controllers and thus can be viewed as highly parameter-robust control laws. The price paid for the ability of adaptive control laws to operate with limited prior modeling information is the complexity of analyzing and quantifying the stability and performance of the closed-loop system, especially in light of the fact that adaptive control laws, even for linear plants, are nonlinear. Stability and performance analysis of adaptive control laws often entails assumptions on the dynamics of the plant. For example, a widely invoked assumption in adaptive control is passivity [1], which is restrictive and difficult to verify in practice. A related assumption is that the plant is minimum phase [2,3], which may entail the same difficulties. In fact, sampling may give rise to non-minimum-phase zeros whether or not the continuous-time system is minimum phase [4], which must ultimately be accounted for by any adaptive control algorithm implemented digitally in a sampled-data control system. Beyond these assumptions, adaptive control laws are known to be sensitive to unmodeled dynamics and sensor noise [5,6], which necessitates robust adaptive control laws [7]. In addition to these basic issues, adaptive control laws may entail unacceptable transients during adaptation, which may be exacerbated by actuator limitations [8–10]. In fact, adaptive control under extremely limited modeling information, such as uncertainty in the signof thehigh-frequencygain [11,12],mayyielda transient response that exceeds the practical limits of the plant. Therefore, the type and quality of the available modeling information as well as the speed of adaptation must be considered in the analysis and implementation of adaptive control laws. These issues are stressed in [13]. Adaptive control laws have been developed in both continuoustime and discrete-time settings. In the present paper, we consider discrete-time adaptive control laws, since these control laws can be implemented directly in embedded code for sampled-data control systems without requiring an intermediate discretization step that may entail loss of stability margins. References on discrete-time adaptive control include [2,3,14–24]. In [2], a discrete-time adaptivecontrol lawwith guaranteed stability is developed under a minimum-phase assumption. Extensions given in [3] based on internal model control [25] and Lyapunov analysis also invoke this assumption. To circumvent the minimum-phase assumption, the zero annihilation periodic control law [23] uses lifting to move all of the plant zeros to the origin. The drawback of lifting, however, is the need for open-loop operation during alternating data windows. An alternative approach, developed in [14,15,17,18], is to exploit knowledge of the non-minimum-phase zeros. In [14], knowledge of the non-minimum-phase zeros is used to allow matching of a desired closed-loop transfer function, recognizing that minimum-phase zeros can be canceled but not moved, whereas non-minimum-phase zeros can neither be canceled nor moved. In [15,18], knowledge of a diagonal matrix that contains the non-minimum-phase zeros is used within a multi-input/multioutput (MIMO)direct adaptivecontrol algorithm.Finally, knowledge of the unstable zeros of a rapidly sampled continuous-time singleinput/single-output (SISO) system with a real non-minimum-phase zero is used in [17]. Motivated by the adaptive control laws given in [3,24], the goal of the present paper is to develop a discrete-time adaptive control law that is effective for non-minimum-phase systems. In particular, we present an adaptive control algorithm that extends the retrospective cost optimization approach used in [24]. This extension is based on a retrospective cost that includes controlweighting aswell as a learning rate, which can be used to adjust the rate of controller convergence and thus the transient behavior of the closed-loop system. Unlike [24], which uses a gradient update, the present paper uses a Newtonlike update for the controller gains, as the closed-form solution to a quadratic optimization problem.Nooffline calculations are needed to implement the algorithm. A key aspect of this extension is the fact that the required modeling information is the relative degree, the first nonzero Markov parameter, and non-minimum-phase zeros, if any. Exceptwhen theplant hasnon-minimum-phase zeroswhoseabsolute value is less than the plant’s spectral radius, we show that the required zero information can be approximated by a sufficient number of Markov parameters from the control inputs to the performance variables. No matching conditions are required on either the plant uncertainty or disturbances. The goal of the present paper is to develop the retrospective correction filter (RCF) adaptive control algorithm and demonstrate its effectiveness for handling non-minimum-phase zeros. To this end, we consider a sequence of examples of increasing complexity, ranging fromSISOminimum-phase plants toMIMOnon-minimumphase plants, including stable and unstable cases. We then revisit these plants under offnominal conditions: that is, with uncertainty in the required plant modeling data, unknown latency, sensor noise, and Received 15 August 2009; revision received 7 November 2009; accepted for publication 16 November 2009. Copyright

Spacecraft tracking using sampled-data Kalman filters

The goal of this article is to illustrate and compare two algorithms for nonlinear sampled-data state estimation. Under idealized assumptions on the astrodynamics of bodies orbiting the Earth, we apply SDEKF and SDUKF for range-only as well as range and angle observations provided by a constellation of six LEO satellites in circular, equatorial orbits. We study the ability of the filters to acquire and track a target satellite in geosynchronous orbit as a function of the sample interval, initial uncertainty, and type of available measurements. For target acquisition, SDUKF yields more accurate position and velocity estimates than SDEKF. Moreover, the convergence of SDEKF is sensitive to the initialization of the error covariance; in fact, a nondiagonal initial covariance is found to be more effective than a diagonal initial covariance. Like SDUKF, by properly setting a nondiagonal initial error covariance, SDEKF also exhibits global convergence, that is, convergence is attained for all initial true-anomaly errors.

Internal Model Control in the Shift and Delta Domains

We construct multivariable internal model controllers in the shift and delta domains. To do so, we develop a linear algebraic approach to the multivariable command following and disturbance rejection problem that facilitates a unified treatment of the differential, shift, and delta domains.

A retrospective correction filter for discrete-time adaptive control of nonmiminum-phase systems

We present a discrete-time adaptive control law that is effective for systems that are unstable, MIMO, and/or nonminimum phase. The adaptive control algorithm provides guidelines concerning the modeling information needed for implementation. This information includes a sufficient number of Markov parameters to capture the sign of the high-frequency gain as well as the nonminimum-phase zeros. No additional information about the poles or zeros need be known.We present numerical examples to illustrate the algorithm¿s effectiveness in handling nonminimum-phase zeros.

Adaptive Control of the NASA Generic Transport Model Using Retrospective Cost Optimization

We provide a detailed description of retrospective-cost based adaptive control, which is a discrete-time adaptive control law for stabilization, command following, and disturbance rejection that is effective for systems that are unstable, MIMO, and/or nonminimum phase. The adaptive control algorithm includes guidelines concerning the modeling information needed for implementation. This information includes the sign of the high-frequency gain as well as the nonminimum-phase zeros. Except when the plant has nonminimum-phase zeros whose absolute value is less than the plant’s spectral radius, the required information can be approximated by a sufficient number of Markov parameters. No additional information about the poles or zeros need be known, and no matching conditions are required. We apply this adaptive control technique to NASA’s Generic Transport Model to illustrate disturbance rejection under unknown, reduced controller authority.

System identification using a retrospective correction filter for adaptive feedback model updating

In this paper we use a retrospective correction filter (RCF) to identify MIMO LTI systems. This method uses an adaptive controller in feedback with an initial model. The goal is to adapt the closed-loop response of the system to match the response of an unknown plant to a known input. We demonstrate this method on numerical examples of increasing complexity where the initial model is taken to be a one-step delay. Minimum-phase and nonminimum-phase SISO and MIMO examples are considered. The identification signals used include zero-mean Gaussian white noise as well as sums of sinusoids. Finally, we examine the robustness of this method by identifying these systems in the presence of actuator noise.

Inherent Robustness of Minimal Modeling Discrete-Time Adaptive Control to Flight Anomalies

We present a discrete-time adaptive control law that is efiective for systems that are unstable, MIMO, and/or nonminimum phase. The adaptive control algorithm provides guidelines concerning the modeling information needed for implementation. This information includes a su‐cient number of Markov parameters to capture the sign of the highfrequency gain as well as the nonminimum-phase zeros. No additional information about the poles or the zeros need be known. We present numerical examples to illustrate the robustness of the algorithm under conditions of uncertainty.

CFD-based adaptive flow control using ARMARKOV disturbance rejection

In this paper we demonstrate adaptive flow control for an incompressible viscous fluid through a two-dimensional channel without the use of an analytical model. An adaptive disturbance rejection algorithm is implemented within a CFD simulation to reduce the effects of an unknown velocity disturbance on the performance variable z, which is the transverse velocity component of the flow at a downstream location. The algorithm requires minimal knowledge of the system, specifically, the numerator coefficients of the transfer function from the control input to the performance variable. System identification, based on CFD simulations prior to the disturbance rejection simulations, is used to identify the required parameters

Adaptive control using retrospective cost optimization with RLS-based estimation for concurrent Markov-parameter updating

We present a discrete-time adaptive control law that is effective for systems that are MIMO and either minimum phase or nonminimum phase. The adaptive control algorithm provides guidelines concerning the modeling information needed for implementation. This information includes a sufficient number of Markov parameters to capture the sign of the high-frequency gain as well as the nonminimum-phase zeros. No additional information about the poles or zeros need be known. In this paper, recursive least-squares estimation is used for concurrent Markov parameter estimation. We present numerical examples to illustrate the algorithms effectiveness in handling nonminimum-phase zeros as plant changes occur.

CFD-Based Adaptive Flow Control for Steady Flow Field Modification

In this paper we demonstrate adaptive flow control for an incompressible viscous fluid through a two-dimensional channel without the use of an analytical model. An adaptive algorithm is implemented within a CFD simulation to control the performance variables, which are velocity components of the flow at downstream locations, to desired steady-state values. The algorithm requires minimal knowledge of the system, specifically, the numerator coefficients of the transfer functions from the control inputs to the performance variables. System identification, based on CFD simulations prior to the closed-loop simulations, is used to identify the required parameters. Steady flow field modification is achieved using the ARMARKOV disturbance rejection algorithm implemented within Fluent, a commercial CFD software package available from Fluent, Inc