Comparison of vibration visualization methods for classification of chaos based on CNN

초록

This study assessed methods for visualizing the vibrations for chaotic systems using a time series, fast Fourier transform (FFT), threshold recurrence plot, and unthresholded recurrence plot. The image classification was then performed using CNN, and the accuracy of each visualization method was compared and analyzed. The nonlinear behavior of chaotic systems was examined using the commonly known Van der pol, Rossler, and Duffing equations. The Lyapunov exponent was calculated for each model parameter change to determine the chaos. The classification accuracy was examined for the chaotic signal in each visualization method of the proposed architecture based on VGG 16 using the determined label and image. The classification accuracy for the chaos of each visualization method is the result of signals mixed randomly five times. FFT analysis showed the highest evaluation result.

키워드