Model reference robot control using fuzzy neural network

Abstract

Multi degree of freedom robot manipulators have nonlinear and complex dynamic characteristics. The use of computed torque etc. conventional type controllers require detailed manipulator dynamics model. Because of coupling effects between links, external disturbances, variation of payloads, frictions, saturation of actuators etc. obtaining the complicated nonlinear manipulator model and its real time implementation in control algorithm are difficult. In this study, fuzzy logic and artificial neural network have been combined together under model reference adaptive control structure to obtain a self-learning and self-tuning sophisticated adaptive robot controller (BUYSA). To achieve this controller structure, fuzzy logic control, neural networks, neuro-fuzzy control subjects have been investigated and a new neuro-fuzzy controller has been developed under the model reference adaptive system structure. Real time structure development and parameter learning algorithms, forming the basis of BUYSA, have been introduced. In the simulations implemented by proposed BUYSA, membership functions and number of fuzzy rules changed in real time in relation to dynamic characteristics of system, variation of system parameters and external disturbances, saturation of actuators, complexity of motion tasks and expected control quality. If degree of freedom of robot and/or complexity of motion tasks increase then huge calculations caused by necessity of bigger fuzzy neural network with bigger number of neurons have to be avoided. To achieve this, a separated structure that formed by fuzzy neural networks has been proposed.

Özet

Çok serbestlik dereceli robotların doğrusal olmayan dinamik modelinin detaylı bir şekilde elde edilmesi ve bu modelin denetim algoritması içinde gerçek zamanda uygulanması zordur. Bu çalışma kapsamında, yaklaşık robot modelini kullanan, uygulamaya yönelik, öğrenebilen ve öz uyarlamalı gelişmiş bir adaptif robot denetim yapısı elde edilmesi amaçlanmıştır. Bu yapıya ulaşabilmek için bulanık mantıkla denetim, yapay sinir ağları ve bunların bileşimi olan yapay sinir ağı temelli bulanık denetleyiciler incelenmiş ve model referans adaptif izleme sistemi yapısında, robot denetimi amaçlı bir bulanık yapay sinir ağı denetleyici (BUYSA) önerilmiştir. Bulanık yapay sinir ağı denetleyicinin temelini oluşturan gerçek zamanda yapı geliştirme ve parametre öğrenme algoritmaları tanıtılmıştır.