본문 바로가기
HOME> 논문 > 논문 검색상세

학위논문 상세정보

Development of Enhanced Robot Calibration Techniques through Consideration of Robot Structures, Measured Data Dimension and the Neural Network Compensation 원문보기

  • 저자

    웬 호아이 난

  • 학위수여기관

    울산대학교

  • 학위구분

    국내박사

  • 학과

    전기전자정보시스템공학과 Robotics

  • 지도교수

    Professor Kang Hee-Jun, PhD

  • 발행년도

    2014

  • 총페이지

    155

  • 키워드

    calibration neural network kinematic model;

  • 언어

    eng

  • 원문 URL

    http://www.riss.kr/link?id=T13540254&outLink=K  

  • 초록

    This dissertation presents the development of enhanced robot calibration techniques through consideration of different robot structures, effects of the measurement dimensions on robot calibration accuracy, and compensation for non-geometric error sources using Artificial Neural Network (ANN). First, kinematic analyses are carried out for three robotic manipulators which have different structures such as a robot having an open chain mechanism (Hyundai HA06 robot), a robot having one closed chain mechanism (Hyundai HX165 robot), and a robot having multiple closed chain mechanisms (Hyundai HP160 robot). In order to identify the robot model parameters, a mathematical formulation for parameter identification is necessarily performed. For the robot with an open chain mechanism, parameter identification equations are simply derived by partial differentiating the position and orientation vectors of robot end effector in term of the robot parameters. For the robot having one or many closed chain mechanisms, the formulation is more complicated. The differential relationships of the closed mechanisms are initially obtained, and then embedded into the differential kinematic equations of the robot principle open linkage to make global identification equations for overall robot parameters. Subsequently, the effects of measurement dimension on robot identification accuracy are studied for three Hyundai robots such as HA06 robot, HX165 robot and HP160 robot. Simulated identification processes are performed on the individual manipulators in two cases, the first case, calibration using full pose (position and orientation) measurements of robot end effector, the second case, calibration using partial pose (position) measurements of robot end effector. For each manipulator after calibration, its identified parameters and position accuracy between the two cases are compared. A validation of robot position accuracy is carried out by using measurement points which are different from the one used in the calibration. By comparing the robot position accuracy of the validating measurements, some remarks are withdrawn. The parameter identification requires robot position and orientation measurements, however, acquiring robot orientations is difficult, laborious, time consuming, and needs special supporting tools. To overcome these difficulties, a new full pose measurement methodology is proposed with advantages such as simple, accurate, fast, not use any special tool, therefore no manufacturing cost and no requirement of pre-calibration before using. The method provides robot full pose based on the feature analysis of the discrete points on a circular trajectory, measured by a non-contact 3D measuring device (e.g., a laser tracker). The accuracy of the proposed method is evaluated via simulation on the Puma robot before applying for experimental calibration on the Hyundai HA06 robot. The experimental results will demonstrate the effectiveness and correctness of the method. Next, to identify the robot parameters, an identification algorithm plays a very important role. There are several algorithms such as maximum likelihood method, Newton Rapson method, nonlinear optimization method, Levenberg-Marquardt method, linear least squares estimation (LSE), extended Kalmand filter (EKF), and etc. Among them LSE and EKF are used widely due to the advantages as fast convergence speed, high accuracy and ease to use. The accuracy of robot identified parameters is compared between using EKF and LSE via computer simulation for the Puma robot. Finally, in robot structure, besides the geometric errors (such as link offset, link length, link twist, joint offset errors) which are easily modeled, there are non-geometric errors (such as link deflection, joint compliance, gear backlash, etc.) which are normally difficult to be modeled. So, artificial neural networks, which have many advantages of high flexibility, good approximation, are utilized to compensate the effects of non-geometric error sources on robot position accuracy. A new calibration method is developed by means of identifying geometric parameter errors by using EKF and compensating non-geometric errors by using ANN. Simulated calibration is performed on Puma robot demonstrate the theoretical correctness and effectiveness of the proposed method. The experimental calibration results of Hyundai HH800 robot confirm the practical effectiveness and correctness of the method.


 활용도 분석

  • 상세보기

    amChart 영역
  • 원문보기

    amChart 영역