Research On Identification Of Dynamic Characteristics And Kinetic Parameters Of Robot

Extracting the dynamic characteristics of the mechanical system has important significance and applied value for the accurate control of kinematics and dynamics of the system itself. Typically represented by the industrial robot, the mechanical system is mainly dealt with in the dissertation, which inquires into a series of questions in extracting the impulse response function (IRF) of robot system. The existing methods of extracting IRF, i.e. time domain method, frequency domain method and wavelet method are analyzed. The dissertation also proposes an identifying method in time domain based on step excitation, which is propitious to robot. In addition, some correlative problems in identifying inertial parameters of manipulator of robot are explored. What is more, a method of applying the artificial neural network (ANN) to parameters identification of system is developed. The dissertation is organized as following:1. In chapter one, the developing status of industrial robot is overviewed at first. And then, some methods of modeling the kinematics and dynamics of robot are generalized. It points out the significance and some key questions of identifying dynamic characteristics and dynamic parameters of the manipulators of robot. After that, the methods of systematic identification are summarized. The purposes and jobs of the project in this dissertation are given at the end of the chapter.2. In chapter two, the relations of several kinds of methods in extracting IRF of system are analyzed theoretically, which prepares for the following chapters in the field of theory. Firstly, the existing methods of extracting IRF of the system are given. And then, based on wavelet transform, two kinds of methods, circular wavelet method and correlative wavelet method, are compared with the time domain method in the meaning of layers of wavelet decompose. It is pointed out that the time domain method is a special case of those two methods and it is more efficient than those methods based on wavelet transform in case of obtaining the same result. It is based on the theoretic judgment, as well as proved by simulation and experiment. Meanwhile, the disadvantage of time domain method, that is, the poor function in antinoise is pointed out, which shows the researching directions clearly for the following chapters.3. The method of extracting dynamic characteristics of system based on step response is researched mainly in chapter three. Firstly, it is interpreted that based on impact excitation, the average spectrum method in test of manipulators of robot is inapplicable, while time domain method based on step excitation is feasible. Then, the characteristics of step excitation in time domain and frequency domain are analyzed. There exists some problems in practical test, i.e. it is hard to measure step force accurately by using force sensor, and it will bring serious errors by replacing the step signal with rectangular pulse. After that,reasons for these questions are analyzed, and the methods to resolve them are put forward. Making use of the linearity of system, the technique of obtaining the step force from step response is developed. Applying step response to obtain parameters in the transfer function of the system, the typical area method is analyzed. The dissertation also presents the sensitivity of time domain method and frequency domain method to different kinds of error. According to the calculus relationship between impulse and step signal, the difference method, which can extract IRF from step response, is researched. Results of simulation indicate that difference method is more efficient than time domain method, and there exists no error when there is noise.4. In order to resolve the questions put forward in chapter two that time domain method is sensitive to noise, in chapter four, it emphasizes on how to carry out noise reduction by using averaging technique, and extract the IRF of system at the same time. Firstly, different signal averaging techniques are analyzed, it is pointed out that those methods neither average different kinds of signal, nor fit for extracting IRF. And then, the cause of ill-conditioned, problem in identification aroused by noise in response signal is interpreted by Riemann-Lebesgue lemma, singular value of matrix and the theory in frequency domain respectively. Based on these theories, the algorithm of compensating difference value and algorithm of partial derivative of error are presented. In the former, the analysis of error shows that the error in IRF is a accumulative value. In the later, algorithms for input error model and output error model are given respectively. Results of simulation show that those methods can average the test signal in different excitation, and the result of identification is close to the theoretical value.5. The method of identifying inertial parameters of manipulator of robot is researched in chapter 5. Firstly, five existing methods of inertial parameters identification are generalized and the disadvantage of them are pointed out, i.e. incapability of identifying on line, requirement for disassembling the robot, being unable to obtain independent parameters, and being incapable of considering the characteristics of joint of manipulator. Method of identifying inertial parameters of end effector by using the wrist force sensor is proposed, and it is proved by experiment. Based on this method, there is an algorithm of identifying inertial parameters of links by using wrist force sensor, which can obtain those independent parameters and doesn't need to model the characteristics of joints. The results of simulation prove the validity of the method.6. In chapter 6, it presents that the parameter identification of system can be achieved by making the best of ANN. The basic theory and the developing status of ANN, and the problems of parameter identification are introduced. The technique of ANN has turned into a rising method in identification of system. The series-parallel connection model of ANN in identification of system is given. It is pointed outthat some typical methods of parameters identification are unable to be used in the case of colored noise and time varying system, the general BP ANN has no specific physical meanings, and there is no general law in structuring the network. This chapter also introduces a method of structuring the network based on mathematic model of system. The parameters waiting for identification are acted as the weighted values of network. A general law and algorithm of structuring the network are given, which are proved by the mathematic model of inertial parameters identification in the fourth chapter. A simulation is carried out in a non-linear system. All of the results show the validity of the method.7. In the last chapter, the advantages, progresses and the lack of the dissertation are summarized, and the dissertation ends with presenting the keystones, directions in the future research and the problems waiting to be solved.