Comparison of different neural networks models for identification of manipulator arm driven by fluidic muscles

Abstract

The main subject of the study, which is summarized in this article, was to compare three different models of neural networks (Linear Neural Unit (LNU), Quadratic Neural Unit (QNU) and Multi-Layer Perceptron Network (MLP)) to identify of the real system of the manipulator arm. The arm is powered by FESTO fluidic muscles, which allows for two degrees of freedom. The data obtained by the measurements were processed in Python. This program served as a tool for compiling individual dynamic models, predicting measured data, and then graphically interpreting the resulting models. Levenberg-Marquardt (LM) was used as the learning algorithm because it is more suitable for learning neural networks than for example, the Gauss-Newton method.