Adaptive Pole-Allocation Control with Multi-Rate Neural Network for Pneumatic Servo System.

Abstract

Pneumatic servo systems contain non-linear elements caused by compressibility of the air, various frictions and so on. And the discrete-time model of the plant usually becomes non-minimum phase caused by above one. Therefore, it is difficult for ordinary linear control methods to accomplish satisfactory control performance in a pneumatic servo system. In this note, we propose a design scheme which combines an adaptive pole-allocation control with a multi-rate type neural network (NN) for the pneumatic servo system. In this design scheme, the role of the NN is to compensate for constructiong a linealized model of the non-linear plant of the electro-pneumatic servo system. 0n the other hand, the role of the adaptive controller is to control above linealized model. As the NN has excellent nonlinear mapping and learning capabilities, this method has accomplished excellent control performance. However, this design scheme whose structure is too complex has a defect that a calculating-time of weights in the NN exceeds one sampling-time. To overcome this problem, we introduced multi-rate type NN which update-time of weights is selected as multiple of the sampling-time. The effectiveness of the proposed design scheme is confirmed by experiments using the existent pneumatic servo system.