Elder Moreira Hemerly

Adaptive control for mobile robot using wavelet networks

This work improves recent results concerning the adaptive control of mobile robots via neural and wavelet networks, in the sense that the stability proof, based on the second method of Lyapunov, encompasses (1) unmodeled dynamics and disturbances in the robot model; (2) adaptation of all parameters in the wavelet networks; and (3) a flexible procedure for automatically adjusting the wavelet architecture. Prior knowledge of dynamic of the mobile robot and network training is not necessary because the controller learns the dynamics online. The wavelet networks parameters and structure are also adapted online. Simulation results are presented by using parameters of the Magellan mobile robot from IS Robotics, Inc.

Optimal control for a class of noisy linear systems with markovian jumping parameters and quadratic cost

The stochastic optimal control problem is discussed for a class of noisy linear systems with markovian jumping parameters and quadratic cost. First a collection of preliminary results is derived which has an important bearing on the rigorous derivation of the optimal control policy from a dynamic programming standpoint. Then the stochastic control problem for the finite-horizon case is analysed. Finally, a result for the infinite-horizon case (long-run average cost) is presented when the jumping parameter is an irreducible continuous-time Markov chain.

Neural adaptive observer for general nonlinear systems

Addresses the problem of nonlinear adaptive observer design for unknown general multivariable nonlinear systems. Only mild assumptions on the system are imposed; the output equation is at least C/sub l/; and existence and uniqueness of solution for the state equation. The proposed observer uses linearly parameterized neural networks whose weights are adaptively adjusted, and Lyapunov theory is used in order to guarantee stability for state estimation and NN weight errors. No strictly positive real assumption on the output error equation is required for the construction of the proposed observer.

Adaptive observers for unknown general nonlinear systems

Several neural network (NN) models have been applied successfully for modeling complex nonlinear dynamical systems. However, the stable adaptive state estimation of an unknown general nonlinear system from its input and output measurements is an unresolved problem. This paper addresses the nonlinear adaptive observer design for unknown general nonlinear systems. Only mild assumptions on the system are imposed: output equation is at least C(1) and existence and uniqueness of solution for the state equation. The proposed observer uses linearly parameterized neural networks (LPNNs) whose weights are adaptively adjusted, and Lyapunov theory is used in order to guarantee stability for state estimation and NN weight errors. No strictly positive real (SPR) assumption on the output error equation is required for the construction of the proposed observer.

Adaptive Stochastic Filtering for Online Aircraft Flight Path Reconstruction

State and parameter estimation using flight test data is highly affected by process and measurement noises, especially with noises displaying time varying statistical properties. Hence, if an estimation problem is to be solved, an adaptive filtering approach is recommended. It is also desirable to obtain the estimates online, simultaneously with flight execution, aiming at a maneuver validation before concluding the flight. Indeed, it is more expensive to put the aircraft back in the air than to extend a little the flight and repeat a test point. Flight path reconstruction is a technique which produces a consistent flight test data set from noisy measurements as a preprocessing scheme to a parameter identification routine. Air data can also be calibrated simultaneously if the problem is formulated properly. This work proposes a methodology to deal with time varying noise statistical properties using a new approach for an adaptive extended Kalman filter. Besides the main filter, two other Kalman filters are proposed to run in parallel, to estimate the process and measurement noise statistics based on the main filter residuals. The proposed adaptive method is derived from the covariance matching technique, by employing filter residuals to adjust the noise statistical properties. Because the method has a low computational cost and is recursive, it is suitable for online applications. The method is validated in a flight path reconstruction application, with simultaneous air data calibration for angle of attack, angle of sideslip, and static pressure sensors. A 100 samples Monte Carlo simulation and real flight test data analysis are used for performance evaluation. Because the proposed approach adequately estimates the statistical noise properties, improved performance is obtained.

An NN-based approach for tuning servocontrollers

Neural networks (NN) are used in this paper to tune PI controllers for unknown plants, which may be nonlinear or open-loop unstable. A simple algorithm, which requires only knowledge of the plant output response direction, is used for training an NN controller, by employing the error between the reference and the plant output. Once this controller achieves good performance, its input-output behavior is approximated by a controller with PI structure, thereby enabling the computation of proportional and integral gains. These gains are familiar to process engineers and can be directly inserted into most existing softwares for process control in industry. Computer simulations on an unstable nonlinear plant and experimental results on a thermal plant are presented to illustrate the usefulness of the proposed approach.

Recursive order estimation of stochastic control systems

The predictive least squares criterion for order estimation is combined with an adaptive control strategy minimizing a quadratic cost and applied to multidimensional ARX systems. It is then shown that this combination enables us to estimate, recursively and in a strongly consistent way, both the order and the coefficients of the controlled system, while achieving asymptotically optimal cost.

Observação adaptativa neural com convergência assintótica na presença de parâmetros variantes no tempo e distúrbios

In this paper a scheme based on neural networks for adaptive observation of a class of uncertain continuous nonlinear systems in the presence of time-varying parameters and non-vanishing disturbances is proposed. Using standard Lyapunov procedures and an adaptive bounding technique, the state error convergence to zero is proved, even when approximation error and disturbances are present, while guaranteeing uniform ultimate boundedness of all others estimation errors (weight, parameter and bounding function). A simulation example to illustrate the application and performance of the proposed algorithm is provided.

PC-based packages for identification, optimization, and adaptive control

PC-based integrated environments are proposed for identification, optimization and adaptive control of industrial processes. The identification package uses the recursive least-squares method for parameter estimation and the predictive least-squares criterion for order selection. The optimization package concerns the optimal tuning of digital PID controllers, given the discrete model of the controlled process, which can be obtained via the identification procedure. The adaptive control package considers several certainty equivalence adaptive control strategies by combining various control strategies by combining various control strategies with several recursive identification schemes. One representative example, using a real process, is presented to illustrate the effectiveness of the techniques used and the usefulness of the integrated environments.<<ETX>>

Direct adaptive control using feedforward neural networks

This paper proposes a new scheme for direct neural adaptive control that works efficiently employing only one neural network, used for simultaneously identifying and controlling the plant. The idea behind this structure of adaptive control is to compensate the control input obtained by a conventional feedback controller. The neural network training process is carried out by using two different techniques: backpropagation and extended Kalman filter algorithm. Additionally, the convergence of the identification error is investigated by Lyapunovs second method. The performance of the proposed scheme is evaluated via simulations and a real time application.