Non-Parametric Modeling of Motion Control Systems Using an Hybrid MODE-NARX Algorithm

Abstract

In practical motion control systems, high fidelity model of the system is fundamental for design, analysis and implementation of control algorithm. Linear Time Invariant (LTI) model approaches provide simplify approach to classical controller design and simulation. However, such approach usually leads to poor real-time performance on the actual system. On the other hand, obtaining nonlinear parametric model has been an arduous task. This paper presents non-parametric modeling approach using an optimized Nonlinear Autoregressive with eXogenous inputs network (NARX-network) with Multiobjective Differential Evolution (MODE). The hybrid algorithm, MODE-NARX addresses challenges of network parameters determination in the conventional NARX network, while providing optimal performance. A laboratory scale motion control systems is used to evaluate the performance of the algorithm. Based on simulation and comparative results analysis performed the proposed hybrid technique outperformed the common well-known PEM-ARMA model with up to 80% better accuracy, and better generalization performance across varying datasets.