Ensemble Grey and Black-box Nonlinear System Identification of a Positioning System
In this work, grey and black-box approaches are used in order to model a electromechanical positioning system (EMPS). An ensemble model is then constructed by combining these two approaches, by using the predictions of both models in order to generate an improved estimated output. Four friction models, in their symmetric and asymmetric versions, namely (i) Coulomb model with finite slope at zero velocity and viscous friction, (ii) Coulomb model with viscous friction, (iii) Tustin friction model, (iv) Coulomb model with viscous friction and Stribeck effect were used to describe the dynamic behavior of the EMPS. The results have shown that the combination of grey and black-box models was able to perform better than the grey-box model and that the proposed friction models are also able to improve the relative error. This encourages further research on the application of the concept of ensemble model construction from machine learning to the nonlinear system identication context towards more accurate model construction.