Neurocontrol of Chaotic Vibration

This study is concerned with several aspects of the nonlinear vibration of a body levitated by an electromagnetic force. Firstly, from the viewpoint of analyzing the qualitative properties of the vibrations, the fundamental characteristics of the electromagnetic levitation system are investigated using phase plane portraits and Poincare maps. Secondly, aperiodic motions are identified as chaotic vibrations by their quantitative properties, such as Lyapunov exponents, fractal dimension and power spectra. Finally, on the basis of the above-mentioned study, the control of chaotic vibration using neural networks containing recurrent paths is investigated. Two architectures of the neural networks, Jordan-type and Elman-type networks, are shown to work as adaptive nonlinear feedback controllers. It is further shown that the neural networks controller can suppress the chaotic vibration more effectively than a PD feedback controller can.