Understanding the limitations of state-of-the-art nonlinear system identification approaches applied to experimental systems
More Info
expand_more
Abstract
In recent years, there has been a lot of focus on system identification using vibration data instead of building mathematical models based on domain knowledge. The Sparse Identification of Nonlinear Dynamical System (SINDy) is a method that has been a great tool to identify the nonlinear dynamics. Therefore, in this report, the nonlinear systems such as duffing oscillator, Single Degree of Freedom (SDoF) with friction are identified using SINDy algorithm. An impact of noise in the measurement data is studied briefly. The components used in the process of identification are smooth finite difference-based differentiator, Sequential Threshold Least Squares (STLSq) optimizer and custom candidate
library. Further, a comparison is made between the numerical models and the models obtained from PySINDy. The Root Mean Squares Errors are calculated for all the cases. It is seen that SINDy is capable of identifying these nonlinearities with good accuracy.