Hidden markov model based fuzzy controller for flexible-link manipulator

A major problem with fuzzy rule-based systems is that with an increasing number of inputs and linguistic variables, the possible number of rules for the system increases exponentially. Unfortunately, there is no systematic approach to learning of rule-base of fuzzy logic (FLC) controller if there is no control expert available, then it must be constructed from the controlled environment or a suitable data set should be available. The adaptive neuro-fuzzy inference system (ANFIS) proposed by Roger Jang, which reduces the development time involved in constructing the rule-base requires a set of input-output data. The problem is now how to cope with developing an FLC where a priori information such as a set of input-output behaviour or expert knowledge is not directly available. The hidden Markov model (HMM) is a probabilistic finite-state machine used in finding structures in sequential data. A rule-base of an FLC can be compared to a finite state machine which can produce a sequence of output MFs. Therefore, the main interest of this research lies in finding a functional mapping from a rule-base of FLC to a hidden Markov model (HMM) and train the HMM using the available data source. The developed controller is then applied to a flexible-link manipulator to verify the performance of the methodology.