Identification of DMSP-5 Fluidic Muscle Dynamics Using Hammerstein Model

Abstract

Fluidic muscles are pneumatic artificial muscles with specific construction and improved characteristics over the muscles with separate tube and outer fibre shell (McKibben type). In the paper we investigate the dynamics of DMSP-5 fluidic muscle, which is treated as a SISO system with voltage as an input variable and its displacement as an output variable. Based on the random step excitation, it is shown here that Hammerstein model, which is an attractive model from control viewpoint, can be a suitable model architecture for modeling its dynamics. Possible structure of this model with 2-zeros, 3-poles linear transfer function and sigmoid neural network approximation of static nonlinearity is identified and used for initialization of the estimation algorithm. The resulting model is based on triangular excitation signal using Levenberg-Marquardt optimization algorithm and achieves 93% fit on test data with total of 12 parameters.