Author(s): Braz-César, Manuel ; Oliveira, Kellie ; Barros, Rui
Date: 2015
Persistent ID: http://hdl.handle.net/10198/17511
Origin: Biblioteca Digital do IPB
Subject(s): Semi-active control; MR damper; Structural control; Fuzzy control; Neuro-fuzzy; Dynamics
Numerical modelling of magneto-rheological (MR) dampers based on parametric models constitutes one of the main methodologies to simulate the response of these actuators. However, the highly non-linear nature of these devices and also their inherent rheological behaviour make this modelling approach harsh and complicated hindering the development of simple numerical models. Usually complex parametric models comprising numerous parameters are required to achieve a reliable and accurate representation of the hysteretic behaviour of MR dampers. On the other hand, non-parametric models seems to be an alternative modelling approach that can deal with the complex non-linear behaviour of MR dampers without the need of to define or identify a large number of model parameters. In this context, this paper provides detailed information about a non-parametric technique based on an adaptive neuro-fuzzy inference system (ANFIS) to create neuro-fuzzy models for MR dampers. An ANFIS is used to optimize a fuzzy inference system by training a family of membership functions in accordance with a predetermined input and output data set related with the damper behaviour. This data optimization algorithm presents the advantage of providing automatic tuning of a fuzzy inference system to relate the device inputs (mechanical excitations and operating currents) to obtain the desired damping force output. Initially, the background and basic concepts of fuzzy modelling with an ANFIS algorithm are described. General guidelines are also provided to improve the optimization procedure with this type of modelling technique. A framework for modelling MR dampers with ANFIS was implemented and its effectiveness in simulating the response of a commercial MR damper was verified with both numerical training data. The results obtained with the resultant neuro-fuzzy model are compared with those of experimental tests and also with an established parametric model (i.e., the modified Bouc-Wen model).
