Neural Network-Based High-Accuracy Motion Control of a Class of Torque-Controlled Motor Servo Systems with Input Saturation

Abstract

The torque-controlled motor servo system has been widely used in the industrial applications in recent years due to its low pollution and high efficiency compared with its hydraulic counterpart [1]. How to design a high-performance controller for the motor servo system has been a topic of great interesting in domestic and foreign research fields [2]. Adaptive control of nonlinear systems has received much attention for obtaining the global asymptotical stability of the closed-loop system [3]. However, all kinds of uncertainties always exist in the system, and these uncertainties could deteriorate the systematic control performance severely [4]. Input saturation is one such uncertainty in the motor servo system, and the typical input saturation is shown in Fig. 1. When the input of the actuator reaches a certain limit value, the output of the actuator does not increase with the increase of the input, which is defined as input saturation. Input saturation can deteriorate the dynamic performance and even result in the instability of the system [5].To deal with the input saturation, one approach is to introduce an additional system to analyse the effect of it. The states of the auxiliary system are employed for controller design and stability analysis [6]. In [7], an auxiliary design system is introduced to analyse the input saturation effect. In [8], in order to overcome the problem of input saturation, a new auxiliary design system and Nussbaum gain functions are incorporated into the control scheme. The control performance could be improved by establishing this kind of auxiliary system, but the model of the system becomes complex, which makes it difficult to be used in engineering applications. A second approach simply takes the input saturation as a kind of disturbance, and an observer is designed for it. An RBF neural network disturbance observer • A single-hidden-layer neural network has been adopted to estimate the input saturation, which cannot be measured.• Experimental results show the effectiveness of the proposed control strategy.

Hybrid DDPG Approach for Vehicle Motion Planning

“…Besides the classical methods, approaches based on artificial neural networks gain more and more interest thanks to their high performance in learning, adaptation and generalization. Supervised learning techniques were used for motion prediction of road vehicles (Yim and Oh, 2004) as well as motor control for industrial robots in dynamic environments (Liu et al, 2017). In recent years, reinforcement learning (RL) was also used successfully for motion planning of car-like mobile robots.…”