Dynamics Identification of Fluidic Muscle-actuated Planar Manipulator Using Two Nonlinear Models

摘要

A model is basic instrument, through that is possible describe systems. Approaches, that are used to design models, are double of kind-empirical or theoretical. Since the essence of the empirical approach is observation based on the real system and the measured data, this approach was chosen to identify the dynamics of the system described in the article. It is a manipulator device whose drive is provided by the fluidic muscles of the manufacturer Festo - two pairs of muscles are engaged antagonistically (one pair for the upper axis, one pair for the lower axis). The models, that are identify for this system, represents the MISO structure (it has multi inputs and single output). In each axis of the manipulator arm were monitored the tension and pressure difference in the muscles at the input, and the angle of rotation of the joint was output. The input (voltage) was generated by a jump or trapezoidal excitation signal. Nonlinear dynamic models-Hammerstein-Wiener and forward neural model of MLP type (Multi-Layer Perceptron) - were created in the System Identification Toolbox and Simulink with using the Matlab program. The Gauss-Newton algorithm was used to estimate the parameters for the training and test phases. The resulting simulation models were evaluated based on the Normal Root Mean Square Error. The resulting models, as well as the initialization of their parameters, are the content of the article. The aim of the research described in this article was to identify the real system as much as possible.