ニューラルネットワークの油圧サーボ系への適用に関する研究

第1報 系の信号伝達特性の同定と振動波形制御入力の創成…シミュレーションによる検討

抄録

Several neural network studies have been carried out to apply both learning and adaptability of the neural networks to servo systems. Genarally, those applications have been carried out for the servo systems, which have rather slow responses, so as to perform a comparative long sampling time interval so as to calculate a learning algorithm of the neural networks.
In this research, the vibratory waveform control method of an electro-hydraulic servo system to reproduce the reference input waveform is discussed. Improvement of the reproducibility of the input signal can be carried out at a comparative high frequency domain by the original control method using two neural networks.
In this control method, in the first place, the signal transfer characteristics of the electro-hydraulic servo system is identified by the neural network which is called the neural network model. Next, the vibratory control input signal is created off-line by using the neural network model as the controlled object, then the control input signal is input into the electro-hydraulic servo system. The output waveform of the piston displacement of the electro-hydraulic servo system is evaluated by performance index J which is defined to evaluate waveform distortion.
This control method applied to the digital simulation to confirm reproducibility of the reference input signal, and the results are as follows :
(1) It is possible to identify the signal transfer characteristics of the electro-hydraulic servo system by the linear neural network using a rectangular input signal and its response output signal, and
(2) It is possible to create a vibratory control input signal by using the neural network model. The digital simulation results show the good performance such that J=4.28% for the rectangular input signal with the fundamental frequency of 1Hz and the frequency characteristics of the controlled system which has a bandwidth of 2Hz.