Συντάχθηκε 18-09-2012 14:23
από Galateia Malandraki
Email συντάκτη: gmalandraki<στο>tuc.gr
Ενημερώθηκε:
-
Ιδιότητα: υπάλληλος ΑΡΜΗΧ.
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Τμήμα Ηλεκτρονικών Μηχανικών & Μηχανικών Υπολογιστών
ΠΑΡΟΥΣΙΑΣΗ ΜΕΤΑΠΤΥΧΙΑΚΗΣ ΔΙΑΤΡΙΒΗΣ
Φίλιππος Ν. Ανδρεάδης
με θέμα
“Bilinear Neuro-Fuzzy Indirect Adaptive Control of Unknown Nonlinear Dynamical Systems’’
Τετάρτη 19 Σεπτεμβρίου 2012, 14:00 μμ
Αίθουσα Εργαστήριο Ψηφιακής Επεξεργασίας Σήματος & Εικόνας, Κτίριο Επιστημών, Πολυτεχνειούπολη
Εξεταστική Επιτροπή
Καθηγητής κ. Ζερβάκης Μιχαήλ (Επιβλέπων)
Επίκουρος Καθηγητής κ. Καρυστινός Γεώργιος
Αναπληρωτής Καθηγητής κ. Μπούταλης Ιωάννης (Τμήμα ΗΜΜΥ, Δημοκρίτειο Πανεπ. Θράκης)
Abstract
Automatic Control of systems has played a vital role in the development of engineering. It plays a determinative role in the successful function of special systems such as spacecraft systems, automatic navigation aircraft systems, mortar shell driving systems, robotics etc. In parallel, it comprises one of the most important components of contemporary industrial and constructive processes. Theoretical development of the automatic control systems has a continuous evolution which was accelerated from the Second World War period until today. Theoretical aspects and control techniques for linear systems have met a particular progress while a special interest has been drawn during the last few decades to the control of non-linear systems, but without reaching a universal applicability.
A basic prerequisite of conventional linear or non-linear control techniques, is the presence of an accurate mathematical model of the dynamical behavior of the under control system. It is known today, that both physical and modern man-made systems can be particularly complex (multi-variable) and are characterized by many nonlinearities. This makes their dynamical mathematical description especially difficult or even impossible, a fact that often leads to their treatment as unknown (black box) systems. The complexity of these systems hinders the design of suitable control techniques. This becomes even more difficult because the dynamical mathematical model required by the conventional approaches is most of the times unknown. Even though the mathematical description is possible, there exist difficulties in the adaptation of the feedback controllers when the system is time-varying with an unknown to the designer way. These drawbacks create the demand for the development of new approximation models and control techniques that have the ability to learn and adapt to varying environmental conditions or internal dynamical behavior of the system.
Artificial neural networks and adaptive fuzzy systems constitute a reliable choice for modeling unknown systems, since during the last years they are considered as universal approximators. In this way, they can approximate any nonlinear function to any prescribed accuracy provided that sufficient hidden neurons and training data are available. Recently, the combination of artificial neural networks and adaptive fuzzy systems has lead to the creation of new approaches, fuzzy-neural or neuro-fuzzy approaches that capture the advantages of both fuzzy logic and neural networks and intend to approach systems in a more successful way. The neural and fuzzy approaches, are most of the time equivalent, differing between each other mainly in the structure of the approximator chosen.
This thesis is based on the development of an Adaptive Recurrent Neuro-Fuzzy Approximator for the identification and control of unknown multi-variable nonlinear dynamical systems, which present various nonlinearities. It extends the operational flexibility of the approximator by admiring a bilinear form in respect to the unknown parameters and proposing new weight updating laws for the on-line parameter updating. This approximator offers a new Neuro-Fuzzy (NF) dynamical description of systems that cannot be mathematically described in an accurate way. The central idea of the new approximator is an alternative description of a classical fuzzy system, which combines the definition of some Weighted Indicator Functions (WIF) with the fuzzy partitioning of the system output variables. In the sequence, the discontinuous WIF functions are approximated by High Order Neural Networks (HONNs). In other words, the central idea of the approximator regarding the fuzzy logic, is the following: Every High Order Neural Network approximates a group of fuzzy rules associated with every center that has resulted from the fuzzy partitioning of the system output variables.
Moreover, after considering the demands of the initial design assumptions in the usual neuro-fuzzy adaptive systems, it is concluded that the new neuro-fuzzy approximator used in this thesis (F-RHONNs) can perform with the existence of much less initial knowledge. It is presented the design, analysis and simulation of new neuro-fuzzy approximators and controllers that can be used for the approximation and control of non-linear affine in the control systems in bilinear form. The approximator that is being used is a dynamical neuro-fuzzy model, which separates the real system to neuro-fuzzy subsystems. Each one of the neuro-fuzzy subsystems separately approximates the respective terms in the dynamical representation of the system defined. The method of parameter hopping is suitably adapted to the new control data and it is reassured in this way that all signals in the closed-loop remain bounded making the system Lyapunov stable. Moreover, the controllers that are being proposed are designed in such a way that the closed-loop error dynamics become linear as well as stable. Introducing and suitably adapting it, the method of parameter hopping once again reassures the existence of the control signal. This method is incorporated in the weight and center updating laws and maintains the closed-loop system Lyapunov stable.
Concluding briefly, in this Master Thesis the main aspect was to develop an indirect adaptive regulation of unknown nonlinear dynamical systems in bilinear form. This method is based on a Neuro-Fuzzy Dynamical Systems definition which uses the concept of Fuzzy Dynamical Systems (FDS) operating in conjunction with High Order Neural Network Functions (F-HONNFs). In this problem the plant is considered unknown, and so it is approximated by a special form of a fuzzy dynamical system while in the sequel the fuzzy rules are approximated by appropriate HONNFs. Thus the identification scheme leads up to a Recurrent High Order Neural Network (RHONN), which however takes into account the fuzzy output partitions of the initial FDS. This scheme does not require a-priori experts' information on the number and type of input variable membership functions making it less vulnerable to initial design assumptions. At first, the system is identified around an operation point, and then it is regulated to zero adaptively. Weight and Center Updating Laws are provided for the HONNFs and the centers of the output membership functions respectively, which guarantee that both the identification error and the system states reach zero exponentially fast, while keeping all signals in the closed loop bounded. We assure the existence of the control signal by applying a method of parameter hopping, which is incorporated in the weight and center updating law. The applicability of the method is tested on a typical problem of Indirect Control, and that is on a DC Motor System, where it is shown that by following the proposed procedure one can obtain asymptotic regulation.
Συνημμένα:
-
Μεταπτυχιακή Διατριβή - Ανακοίνωση Ανδρεάδης.doc
Μεταφορτώσεις: 159,
Μέγεθος: 51 KB application/msword
