Damian Kowalów

Sequential design for model calibration in iterative learning control of DC motor

The positioning problem for repeated DC motor runs based on the iterative learning control technique enhanced with model calibration is discussed. In order to increase the quality of control and reduce the model uncertainty, the conventional iterative control approach is enhanced with parameter estimation of the mathematical model. This is achieved through proper adaptation of the iterative experimental design technique properly incorporated into general iterative control scheme. The setting examined here correspond to situation where from among all the measurements gathered in repeated trials of the process the most informative observations are selected in order to provide an update of the parameter estimates. In such a way, in each iteration loop both the quality of control and model of the process can be significantly improved. A proposed approach is verified on the application example of DC servo motor system.

Decentralized Scheduling of Sensor Networks for Parameter Estimation of Spatio-Temporal Processes

The activation scheduling problem for a scanning sensor network monitoring a spatio-temporal process is considered. The configuration of an activation schedule for network nodes measuring the system state is formulated in a sense of a suitable criterion quantifying an estimation accuracy for system parameters. Then, a decomposition of the scheduling problem is provided and a proper distribution of total computational effort and consensus between the network nodes is achieved via information flooding based on a pairwise communication scheme. As a result, a simple exchange algorithm is outlined to solve the design problem in a decentralized fashion. The proposed approach is illustrated on an example of sensor network configuration for monitoring an atmospheric pollution transport process.

Optimal sensor selection for model identification in iterative learning control of spatio-temporal systems

An approach to sensor location problem for parameter estimation of a distributed system controlled under repetitive regime is presented. In order to reduce the uncertainty of the model used for the control design, thus increasing the system performance, the iterative learning control scheme is extended with parameter estimation of mathematical model with the use of the sequential experimental design. The related sensor location problem corresponds to situation where from among all potential sites where the sensors can be placed we have to select a subset which provide the most informative measurements in order to update the system parameter estimates. Thus, in each process trial, both the control performance and process model can be substantially improved. As an illustration of the proposed approach the application to nontrivial chemical process of fuel combustion is given.

Robust sensor scheduling via iterative design for parameter estimation of distributed systems.

The problem of selection of measurement data provided by sensor array so as to maximize the accuracy of parameter estimation of a distributed system defined in a given multidimensional domain is discussed. Usually, when designing an identification experiment for nonlinear models, the uncertainty of nominal parameters has to be taken into account. In particular, an iterative scheme for parameter estimation is proposed enhanced with sequential experimental design techniques where there is no particular information about the parameter distribution. The setting examined here correspond to situation where from among the observations provided by nodes of given sensor array the most informative measurements have to be selected in order to provide an update of the parameter estimates. Finally, a proposed approach is verified by a computer simulation regarding heat transfer problem.

Distributed Configuration of Sensor Network for Fault Detection in Spatio-Temporal Systems

The problem of fault detection in spatio-temporal systems is formulated as that of maximizing the power of a parametric hypothesis test verifying the nominal state of the process under consideration. Then, adopting a pairwise communication schemes, a computational procedure is developed for the spatial configuration of the observation locations for sensor network which monitor changes in the underlying parameters of a distributed parameter system. As a result, the problem of planning the percentage of experimental effort spent at given sensor locations can be solved in a fully decentralized fashion. The approach is verified on a numerical example involving sensor selection for a convective diffusion process.

Optimum training design for neural network in synthesis of robust model predictive control

The paper deals with determining the neural network model uncertainty for the purpose of robust controller design. The approach presented in the paper is based on the application of optimum experimental design for the choice of sequences providing the most informative data during the training of neural network. As a criterion quantifying the quality of training process a measure operating on the Fisher Information Matrix related to the estimates of network parameters is used. Then, it is possible to analyze the variance of the predicted network response and estimate how possible variations of parameter values influence the changes observed in the predicted model output. This allows to construct an appropriate cost function for the control system taking into account the model uncertainty and incorporate it into model predictive control scheme.

Distributed scheduling of measurements in a sensor network for parameter estimation of spatio-temporal systems

Abstract The main aim of the paper is to develop a distributed algorithm for optimal node activation in a sensor network whose measurements are used for parameter estimation of the underlying distributed parameter system. Given a fixed partition of the observation horizon into a finite number of consecutive intervals, the problem under consideration is to optimize the percentage of the total number of observations spent at given sensor nodes in such a way as to maximize the accuracy of system parameter estimates. To achieve this, the determinant of the Fisher information matrix related to the covariance matrix of the parameter estimates is used as the qualitative design criterion (the so-called D-optimality). The proposed approach converts the measurement scheduling problem to a convex optimization one, in which the sensor locations are given a priori and the aim is to determine the associated weights, which quantify the contributions of individual gaged sites to the total measurement plan. Then, adopting a pairwise communication scheme, a fully distributed procedure for calculating the percentage of observations spent at given sensor locations is developed, which is a major novelty here. Another significant contribution of this work consists in derivation of necessary and sufficient conditions for the optimality of solutions. As a result, a simple and effective computational scheme is obtained which can be implemented without resorting to sophisticated numerical software. The delineated approach is illustrated by simulation examples of a sensor network design for a two-dimensional convective diffusion process.

Distributed design of sensor network for abnormal state detection in distributed parameter systems

The problem of measurement effort distribution for detection of the abnormal state of distributed parameter system monitored with sensor network is considered. The measurement strategy is formulated in terms of maximizing the power of parametric hypothesis test related to the nominal system state. Then, using communication schemes based on the class of so-called gossip algorithms a computational procedure for optimizing the measurement effort over the sensor network is proposed. Finally, the presented fault detection approach is verified on the example of convective-diffusion process.

Robust sensor location for parameter estimation in iterative learning control of spatio-temporal systems

The problem of spatial sensor location under parametric uncertainty of the repetitive distributed-parameter process is discussed. The idea is to reduce the uncertainty of the model used for the design of the iterative learning control, thus increasing the system performance. Particularly, an iterative scheme for estimation of the system parameter distributions is proposed based on the sequential experimental design techniques. From among the measurements provided by the nodes of given sensor array the most informative data have to be selected in order to provide an update of the system parameters. Therefore, in each process trial, both the control performance and parameter estimates can be significantly improved leading to robust experiment design with respect to unknown parameter distribution. The approach is illustrated with computer simulation of repetitive deceleration process using magnetic brake system.

Vision-Based Convoy Forming for Mobile Robots

An approach is proposed to form a convoy of vehicles autonomously following a given path. Particularly, the part of the complex mobile robot guidance system related to leader-follower control process is presented in detail. The ultimate objective is to use the local vision system of mobile robotic platform to follow the moving goal as accurate as possible simultaneously keeping the constant distance from the leading robot. Then, a proper controller design is proposed together with the implementation for the Amigobot mobile robotic platform from Adept Mobile Robots Inc. A verification of the guidance system performance via navigation experiments is also presented.