Grey box model-based adaptive control

Grey box model-based adaptive control

128 Abstracts decoupling is discussed. An equivalent linear controller for the algorithm is given for robust stability analysis. Some rules for choo...

129KB Sizes 0 Downloads 99 Views

128

Abstracts

decoupling is discussed. An equivalent linear controller for the algorithm is given for robust stability analysis. Some rules for choosing the tuning knobs of GPC/MRM from the robust stability point of view are established. The introduction of an observer, as in GPC, is extended to MIMO GPC/MRM, and its effects on robust stability are studied. The method is applied to a flexible arm system; simulation results show the controller's decoupling efficiency and robustness.

100 Model Reduction for PID Design A~I. Isaksson, S.F. Graebe, pp 467-472 Generally, Internal Model Control (IMC) results in a controller of the same order as the model. Therefore, within the IMC paradigm, PID control of higher than second-order plants is associated with a model-reduction problem. Solutions to this problem have previously been formulated as criterion optimisation. This paper outlines a computationally convenient alternative, in which the model is calculated as the average of the model obtained by retaining the slowest pole(s), and that obtained by retaining the low-order coefficients. Advantages of this technique include simplicity, preservation of physical parameters, and a closed-loop performance that is comparable to, or better than, existing methods in most cases. 101 Identification of Continuous-Time Systems with PartiallyKnown State-Dependent Disturbances C. Canudas de Wit, B. Brogliato, C.R. Johnson, pp 473-476 This paper presents two continuous-time versions (with and without data normalization) of the exponentially weighted reclusive least-squares algorithm (EW-RLS). These algorithms are suitable for identifying systems with bounded partially known disturbances, since they explicitly account for these disturbances and ensure parameter boundedness. The paper presents the derivation of these algorithms and the associated convergence issues.

102 Grey Box Model-based Adaptive Control H. Brabrand, S. Bay Jergensen, pp 477-480 A process-knowledge-based procedure for the development of model structures for adaptive control is presented. The method utilizes additional measurements, suitably located within the process, in order to handle nonlinearities using linear models. The model structure is represented in an ARX representation. Application of the grey-box model-development procedure, and identification and control results, are demonstrated upon a simulated sequence of five continuous stirred tank reactors. The output coupling structure is more efficient than a transfer function structure, for both identification and adaptive control. The implementation time for adaptive control is significantly reduced when compared to conventional adaptive control design.

103 Parameter Constrained Adaptive Control W.D. Timmons, H J . Chizeck, V. Chankong, P.G. Katona, pp 481.484 Optimally constrained identification for adaptive control can dramatically improve transient controller performance, especially when combined with techniques that selectively adjust adaptation gain. In this paper, a real-time identifier that optimally imposes linear constraints is developed, linear constraints that encode common information are presented, and application guidelines are demonstrated.

104 A Designer Guide for Grey-Box Identification of Nonlinear Dynamic Systems with Random Disturbances T. Bohlin, pp 485-488 This contribution reviews an attempt to create and test a theorybased procedure for grey-box identification of possibly nonlinear industrial processes. The model designer interacts in a systematic way with a general identification tool kit to create and validate models, when the a priori information and data are incomplete and uncertain. Two applications to industrial processes demonstrate the feasibility.

105 Experiment Design for Grey-Box Models H. Melgaard, P. Sadegh, H. Madson, J. Hoist, pp 489-492 In this paper, design methods are presented, where informationrelated criteria for the optimality of the precision of the resulting parameter estimates are extended with prior information, corresponding to the available physical knowledge about the system, and with cost-related loss function dements, reflecting the importance of the parameter estimates obtained. Bayesian, as well as non-Bayesiau, techniques are discussed. An example is given of the grey-box approach for optimal design of a powerconstrained input sequence.

106 Grey Dynamics of Heat Exchanger Networks E.I. Varga, J. Bokor, K.M. Hangos, pp 493-498 A robust modeling method for heat exchanger networks (HENs) with structured parametric perturbations, resulting in an LFT form, is described in this paper. The LFT of the overall HEN system can be generated algcrithmically from the model of the heat exchangers, and from the network topology. Based on these modeling paradigms, the effect of uncertainties represented by specific A block structures have been analysed by frequency domain I.t analysis, in order to get an insight into the causes of potential large overshoot-type responses. The proposed method has been illustrated on the simple example of a countercurrent heat exchanger with bypass.

107 On the Use of Descriptor Systems for Failure Detection and Isolation A. Benvenlste, M. Bassevllle, R. Nikoukhah, A.S. Willsky, pp 499-502 This paper investigates the problem of failure detection and isolation (FDI) for noisy dynamic systems. A statistical approach is followed, so that performance criteria are related to false alarm rate and detection delay. Using a simple and powerful reformulation of maximum likelihood estimation in linear Gaussian systems, a new approach is introduced, based on a descriptor systems formulation. A reduction in computational cost results, as well as a significant improvement in robustness against uncertainty in the dynamics.

108 Fault Estimation in IAnear Dynamic Systems J. Chen, RJ. Patton, pp 503-506 Information about faults plays an important role in fault diagnosis and control reconfiguration. This paper proposes a faultestimation method which involves a two-step procedure. The first step is used to generated the residual from inputs and outputs of the system. In the second step, a fault estimator is used to estimate faults using the residual signal. The design problem of the fault estimator is solved using linear inverse system theory. A simple example is used to illustrate the method.

109 Generating Directional Residuals with Dynamic Parity Equations J.J. Gertler, R. Monajemy, pp 507-512 It is shown how diagnostic residuals, which exhibit directional properties at all times in response to an arbitrary mix of input and output faults, can be generated using dynamic parity equations. The residual generator is obtained by applying a transfer-functiontype transformation to the dynamic input-output model of the monitored plant. It is also shown that the residual generator is polynomial (moving average) if the specified fault responses contain the invariant polynomial of the fault system. 110 Detection of Parameter Variations by Continuous-Time Parity Equations T. HSfling, pp 513-518 Because many faults induce parameter variations, it is important for detection and diagnosis to observe their development in time. One possibility is to estimate the parameters directly by wellknown least-squares methods; however, this requires computational effort at each sampling step. The proposed method uses the easy-to-calculate parity space equations with special, parameter-sensitive features and state variable fillers to generate signal derivatives. Parity equations are used here in the continuous-time domain. A parameter-classification table can be