Motor-mechanism dynamic model based neural network optimized computed torque control of a high speed parallel manipulator

Abstract

There are wide applications of the parallel manipulators in the industry field because of the high speed and accuracy, but difficulties are also unavoidable in the controlling process for the systematic time-varying and coupling characteristics. Therefore, the approaches for better control performance are vital to the application of parallel manipulator. In this paper, one newly invented 2-DOF (degree of freedom) parallel manipulator called Diamond 600 is investigated as an objective for Pick and Place operation. Firstly, the dynamic model of mechanism and servo driving subsystems are formulated using virtual work principle and field orient control principle, respectively. According to the equivalent torque principle, a mechatronic (motor-mechanism coupling) dynamic model is deduced for real-time control. Secondly, the neural network optimized computed torque control algorithm is described in detail for the proposed coupling model. At last, a series of numerical simulations and experiments are carried out, respectively, to prove the validity of the above methods, and results verify the favourable tracking ability and robustness.

Introduction

The serial manipulators have been applied in some but not the entire industrial field with their priorities. Comparing with the serial ones, the parallel manipulators have potential advantages in terms of stiffness, accuracy, high speed and payload. They can be widely applied to the following fields, like the Pick and Place operation in food, medicine, electronic industry and so on. At present, the key issues are the ways to meet the demand of high accuracy in moving process under the condition of high speed because of the systematic nonlinearity.

Actually, the dynamic model of the parallel manipulator is of nonlinearity with the time-varying and coupling characteristics, which comprises the driving motors, reducer and the parallel mechanism [1], [2], moreover, the analysis of complex motor-mechanism coupling model is hard and time-consuming. From the viewpoint of system, it is very important to design one controller with good performance in order to match the mechanism. In the history of the controller development, it can be divided into three phases: classical controller, modern controller and intelligent controller. The classical control methods, such as simple PD and PID linear control, are simple but not good performance. It is worth noting that they have been applied in the industry widely just for the simpleness. The modern control methods are the control strategy on the basis of state-space function, it is mainly applied to the multi-input and multi-output process. The intelligent control methods, which are the most complex ones, can prove the best control performance, but there are so many undetermined factors, which are hard to design, depending on human’s intelligence in the control process.

Considering the advantages of the above methods, many modern control techniques have been developed for the nonlinear mechanism systems [3], [4], such as adaptive control, fuzzy control, sliding-mode control, computed torque control and neural network control. Fuzzy control is a valid method but difficult to design the suitable fuzzy logic rules. Artificial neural network control is such a method that has the learning capability from process, and need not to formulate the ‘real’ dynamic model of objective, but it will cost a lot of time for iterative calculation. Sliding-mode control is a new method but some bounds due to systematic uncertainties must be pre-estimated. In fact, in the control process of robot, computed torque control, which is normally adopted for usage, has a good performance [5], [6], [7] by considering nonlinear compensations to the dynamic model. It is simpler than the intelligent control method, but sensitive to the systematic parameters’ change and outer disturbances. In order to solve the problem mentioned, the intelligent algorithms are normally added to the controller for improving the validity and robustness of system. In the most time, a PID type controller is often incorporated with some adaptive control schemes including sliding-mode control, neural network control and so on [8], and applied to a serial robotic system successfully [9], [10], [11]. Unfortunately, in the most researches, the objective studied is just mechanism, seldom includes driving system. Therefore, it must bring some errors for sure. Otherwise, note that computed torque control is a kind of model based control algorithm, so designing this kind of controller with fixed gain is very difficult to reach the performance demanded. According to the above reasons, an on-line updated PID algorithm is proposed [12], this algorithm can be implemented directly to other mechanisms with little modification. It is worth noting that the design of this type controller is on basis of the linear model not the nonlinear one. Furthermore, neural network hybrid controller is designed and the motor dynamics is proposed [13], [14]. In general, the above dynamic models deduced are very simple. Especially, the dynamic modeling of the motor-mechanism coupling parallel manipulator or robot has seldom been discussed.

In this paper, the 2-DOF parallel manipulator is introduced briefly as the initial objective, and the mechanism and motor dynamic models are formulated, respectively, in advance. Then the whole motor-mechanism coupling model is set up in terms of equivalent torque principle. Furthermore, the computed torque controller is designed for the coupling model of manipulator, and the neural network algorithm is used to optimize the parameters of controller for the best performance. The simulations and experiments are carried out to validate the above theory in the following part. In the last section, conclusions are drawn.