مجله کنترل
سال پنجم شماره 2 (تابستان 1390)

Stability of SDRE observers is usually analyzed in a deterministic framework, where the performance of the observer is not perturbed by external noises. However, plant uncertainty and measurement noises are not negligible in real implementations. In this paper based on the stability results developed for Kalman-Bucy filters and stochastic stability analysis of nonlinear systems, the stability of an SDRE filter is analyzed in discrete-time domain. Sufficient conditions are given to limit the estimation errors in sense of mean-squared measure. Furthermore, it is shown that in practical situations there exist feasible regions to satisfy the stability conditions, provided that the initial error and the norm of plant uncertainty and measurement noises are sufficiently small. Finally, the stability conditions are verified through simulation of a typical nonlinear system.

The class of switched linear systems with state-input logic constrained can be used for modeling of many systems. This article presents a new stability analysis and controller design method for this class of hybrid systems. Proposed method is based on quadratic lyapunov function. Computational approach for stability analysis and design is convex optimization (Linear Matrix Inequality type). Simulation results on dc-dc buck converter show the effectiveness of the proposed method.

Due to the status of Rotary Kiln Cements (RKCs) in different industries and lack of a mature model for these systems, identification and prediction of the Kiln system are necessary for any simulation and automation approaches. Intrinsically, RKCs are non-linear and time variant systems. This paper proposes a novel approach of using ANSFI to predict the status of a RKC system in a scale of few minutes in advance. Since the data used in this research has been extracted from a real system, pre-analysis of data is one of the critical parts of identification process. In addition to the system inputs, dynamic of the system which has been selected according to the LIPSCHITZ method with a system’s genuine delay are applied as inputs for Neural Network system with one step phase lag. Genetic algorithm has been utilized as a characteristic selection and phasor rules reduction method due to the existing challenges on the number of rules in phasor systems specifically with a large number of variables to be applied to the Neural Network. To verify the performance of the proposed identification and prediction method on a non-linear industrial system, simulation results have been carried out on a real data extracted from SAVEH Cement Company.

The μ synthesis problem has not been completely solved, and this is attributed to existing challenges and issues in calculation of the structure singular value. The most common solution for μ Synthesis problem is called D-K Iteration. Even though this specific method is not the complete solution for the μ Synthesis, but the controllers obtained through this method have proven to be one of the most complete forms of robust control technique, based on robust stability and performance. One of the major disadvantages with the D-K Iteration is a high order controller. In this paper an evolutionary algorithm called PSO has been used to design a robust controller. The main objective is to find a solution for μ Synthesis that can better improve the robust stability and performance compared to same order (reduced order) controller obtained through D K Iteration. To evaluate the proposed algorithm, it has been used on the mass-spring-damper benchmark system. The simulation results from the proposed algorithm show that this method has a more robust stability and performance for closed loop systems than the same order controllers obtained through D-K Iteration.

The SVM is the well known technique for power converters with medium power and high voltage. However, the whole modulating procedure could be time-consuming for implementation purposes because of seeking the location of the reference vector in addition to performing so many necessary computations. Furthermore, computational cost for conventional SVM rises when level numbers of the multilevel converters increases. This paper proposes a modulation technique that directly concentrates on a three-phase system, engaging two independent line voltages in the procedure of finding switching states and their duty ratios. Interestingly, the proposed method is not only simple and fast, but also nearly eliminates the procedure of positioning the reference vector for multilevel converters that is vital for the conventional SVM. Through examples and simulations, the validity of the proposed method in modulation process is demonstrated.

Type-2 fuzzy neural networks have a good ability in identification and control of nonlinear systems, time varying systems and also system with uncertainties. In this paper a new method for designing adaptive inverse type 2 fuzzy neural controllers for online control of nonlinear dynamical system has been introduced. The proposed network has seven layers that the first two layers consist of type-2 fuzzy neurons with uncertainty in mean of Gaussian membership functions, are used for fuzzification part. Third layer is the fuzzy rules layers. Reduction type is done in fourth layer with adaptive nodes. Reminder layers are used for consequent left–right firing points, two end-points and output of network respectively. In this paper, gradient descent with adaptive learning rate backpropagation is used for learning phase. Finally, Type-2 online Sugeno fuzzy neural network is used for tracking control of nonlinear dynamical water bath temperature system. Results are compared with Adaptive-Network-Based Fuzzy Inference System (ANFIS). Simulation results show the proposed method has a good efficiency.

An important shortcoming about designing robust control for models generated from prediction error identification is that very few control design methods directly match the ellipsoidal parametric uncertainty regions that are generated by such identification methods. Models obtained from identification experiments in prediction error framework lie in an ellipsoid uncertainty region. In this contribution we present a joint robust control/input design procedure which guaranties stability and prescribed closed-loop performance using models identified from experimental data. Finite dimensional parameterization of the input spectrum is used to represent the input design problem as a convex optimization. A method for fixed-order H controller design for systems with ellipsoidal uncertainty is used and given H specifications on the closed-loop transfer function are translated into sufficient requirements on the input signal used to identify the system.