Soft computing methods for performance improvement of EAMA robot in fusion reactor application
Wu, Jing (2018-03-20)
Väitöskirja
Wu, Jing
20.03.2018
Lappeenranta University of Technology
Acta Universitatis Lappeenrantaensis
Julkaisun pysyvä osoite on
https://urn.fi/URN:ISBN:978-952-335-208-7
https://urn.fi/URN:ISBN:978-952-335-208-7
Tiivistelmä
The experimental advanced superconducting Tokamak (EAST) has achieved a series of important research results and scientific discoveries. However, the EAST inner components of the first wall will also face an increasingly tough operating environment with high heat loads. Therefore, in order to ensure adequate running time, studying remote handling (RH) maintenance of the EAST device during physical experiments is a challenging task. The EAST's articulated maintenance arm system (EAMA) is developed for real-time detection and maintenance operations during plasma discharges without breaking the ultra-high vacuum conditions. To achieve the desired performance, EAMA must guarantee accuracy and stability. Building up the foundations needed for developing sensor fusion in a timely way can be facilitated by familiar hypothesisdriven or first principles approaches but also by engaging modem data-driven statistical methods. These methods feature machine learning (ML), an exciting R&D approach that is increasingly deployed in many scientific and industrial domains. An especially time-urgent and very challenging task in the development of intelligent RH services today is to reliably deal with large-scale major disruptions in magnetically-confined tokamak devices of the near future. Prediction methods with better predictive capability are required to provide sufficient advanced result for distribution mitigation or optimization strategies to be effectively applied to system remaining to be improved. This truly formidable task, outlined in this work, demands accuracy beyond the nearterm reach of hypothesis-driven or first-principle simulations that dominate current research and development in the field. The ML methods deal with very large data sets hold significant promise for delivering the much-needed EAMA predictive tools that can be generalized at the basic level and used in multiple application domains. In particular, the signal data from the superconducting tokamak plasmas of high temperature (80-120°C ), and high vacuum(~ 10-5Pa), such as the EAST, is of significant interest to explore. In addition, the topic of vibration control, as an extension of our ML capabilities, is also a viable and timely subject to be studied. The main contributions of this dissertation include: the architecture, communication and model analysis of the entire EAMA software system; the optimization of EAMA trajectory by a genetic algorithm minimizing the end-point jerk; the study of two different methods, the extended Kalman estimator and the adaptive neural fuzzy system, to predict the pitch and yaw joint errors of the manipulator; and the eventual development of an estimation algorithm of EAMA dynamic vibration to predict the EAMA system operation.
Firstly, the design of the EAMA system should guarantee that the robot can stably run in the harsh environment of high temperature (80-120 'C) and high vacuum (about 10-5Pa). The EAMA manipulator is a typical multi-body system; the overall speed is not high with an average joint angular velocity of about -0.5-1 ° /s. Meanwhile, the EAMA has a reduced structural stiffness and strictly limited operating speed. The inertial forces generated from acceleration could generate exaggerated, unwanted displacement and vibration. Any inappropriate motions can significantly cause system performance degradation by reducing positioning accuracy and aggravating the settling time which could result in system instability. To overcome the flexibility weakness of EAMA, a series of measures are taken to enhance the accuracy of EAMA in several fields: the mechanical flexibility of multi-body system dynamics, the accurate control in high performance systems, and the stability-optimized motion plan.
Secondly, we present a trajectory optimization method that pursues the stable movement of the ?-degree-of-freedom-articulated arm, which maintains the mounted inspection camera anti-vibration. Based on dynamics analysis, the trajectory optimization algorithm adopts multi-order polynomial interpolation in the joint space. The object of the optimization algorithm is to suppress the end-effector vibration by minimizing the root mean square value of jerk. The proposed solution has such characteristics that can satisfy kinematic constraints of EAMA' s motion and ensure that the arm runs within the boundaries of absolute values of velocity, acceleration and torque. The genetic algorithm is employed to search for a global and robust solution of this problem by mapping a jerk transformation under 0.5 m/s3 .
Thirdly, for sensors in the EAMA position control, two algorithms are implemented for estimating and compensating segment position error. For yaw joint, the error model uses curve fitting, which has unnegligible nonlinearities. The extended Kalman filter is adapted to make the segment position compensation error accurate based on the curve fitted model. For pitch joint, it has two distinct tasks, shaft rotation direction signal processing and discrete data classification. Meanwhile, the ideas of neural network and expert system are applied to complete these tasks respectively. In this part, the use of an adaptive neuro-fuzzy inference system for estimating the compensation error from an unformulated cluster of data forming a disclosed hysteresis loop. The experiment results have shown that the root mean squared error is significantly improved, and the final results satisfy the accuracy requirement of up to 0.02 degrees.
Finally, an open software architecture developed for the EAST articulated maintenance arm(EAMA) is described. In the control point of view, it offers robust and proper performance and an easy-going experience based on Open Robot Control Software (OROCOS). The software architecture is a multi-layer structure including: an end layer, an up layer, a middle, and a down layer. In the end layer, the components are defined off-line in the task planner manner. The distributed architecture of the control system associating each processing node with each joint is mapped to a component with all functioning features of the framework.
Firstly, the design of the EAMA system should guarantee that the robot can stably run in the harsh environment of high temperature (80-120 'C) and high vacuum (about 10-5Pa). The EAMA manipulator is a typical multi-body system; the overall speed is not high with an average joint angular velocity of about -0.5-1 ° /s. Meanwhile, the EAMA has a reduced structural stiffness and strictly limited operating speed. The inertial forces generated from acceleration could generate exaggerated, unwanted displacement and vibration. Any inappropriate motions can significantly cause system performance degradation by reducing positioning accuracy and aggravating the settling time which could result in system instability. To overcome the flexibility weakness of EAMA, a series of measures are taken to enhance the accuracy of EAMA in several fields: the mechanical flexibility of multi-body system dynamics, the accurate control in high performance systems, and the stability-optimized motion plan.
Secondly, we present a trajectory optimization method that pursues the stable movement of the ?-degree-of-freedom-articulated arm, which maintains the mounted inspection camera anti-vibration. Based on dynamics analysis, the trajectory optimization algorithm adopts multi-order polynomial interpolation in the joint space. The object of the optimization algorithm is to suppress the end-effector vibration by minimizing the root mean square value of jerk. The proposed solution has such characteristics that can satisfy kinematic constraints of EAMA' s motion and ensure that the arm runs within the boundaries of absolute values of velocity, acceleration and torque. The genetic algorithm is employed to search for a global and robust solution of this problem by mapping a jerk transformation under 0.5 m/s3 .
Thirdly, for sensors in the EAMA position control, two algorithms are implemented for estimating and compensating segment position error. For yaw joint, the error model uses curve fitting, which has unnegligible nonlinearities. The extended Kalman filter is adapted to make the segment position compensation error accurate based on the curve fitted model. For pitch joint, it has two distinct tasks, shaft rotation direction signal processing and discrete data classification. Meanwhile, the ideas of neural network and expert system are applied to complete these tasks respectively. In this part, the use of an adaptive neuro-fuzzy inference system for estimating the compensation error from an unformulated cluster of data forming a disclosed hysteresis loop. The experiment results have shown that the root mean squared error is significantly improved, and the final results satisfy the accuracy requirement of up to 0.02 degrees.
Finally, an open software architecture developed for the EAST articulated maintenance arm(EAMA) is described. In the control point of view, it offers robust and proper performance and an easy-going experience based on Open Robot Control Software (OROCOS). The software architecture is a multi-layer structure including: an end layer, an up layer, a middle, and a down layer. In the end layer, the components are defined off-line in the task planner manner. The distributed architecture of the control system associating each processing node with each joint is mapped to a component with all functioning features of the framework.
Kokoelmat
- Väitöskirjat [1037]