A comparative study of linear ARX and nonlinear ANFIS modeling of an electro-hydraulic actuator system

Abstract

The existence of a high degree of nonlinearity in Electro-Hydraulic Actuator (EHA) has imposed a challenge in development of a representable model for the system such as that significant control performance can be proposed. In this work, linear Autoregressive with Exogenous (ARX) model and nonlinear Adaptive Neuro-Fuzzy Inference System (ANFIS) model of an EHA system are obtained based on the mathematical model of the system. Linear ARX modeling technique has been widely applied on EHA system and satisfying result has been obtained. On the other hand, ANFIS modeling technique can model nonlinear system at high accuracy. Both models are validated offline using data set obtained and using different stimulus signals when doing online validation. Offline validation test shows that ANFIS model has 99.37% best fitting accuracy, which is more accurate than 93.75% in ARX model. ARX model fails in some online validation tests, while ANFIS model has been consistently accurate in all tests with RMSE lower than 0.25