University of Twente Student Theses
Data Driven Feedforward Control of a 2-DOF Redundantly Actuated Manipulator with Flexure Joints using Machine Learning Techniques
Sathiyanarayanan, Prasanna (2021) Data Driven Feedforward Control of a 2-DOF Redundantly Actuated Manipulator with Flexure Joints using Machine Learning Techniques.
|
PDF
6MB |
| Abstract: | Robotic manipulators are capable of performing various tasks at speeds and accuracies that far exceed those of human operators. They find applications in various industries ranging from manufacturing to precision engineering. To perform their tasks reliably at such high speeds with precision the end effectors of the manipulator are controlled digitally. Control of robot manipulators can be greatly improved with the use of feedforward control. However, the effectiveness of the feedforward control heavily relies on the accuracy of the inverse dynamics model of the manipulator. Classical physical modelling approaches suffers from modelling inaccuracies due to uncertain factors of the manipulator such as flexibility, friction and hysteresis, etc. Alternatively, artificial intelligence techniques are becoming increasingly popular in robotics applications in the recent years. It is hypothesised that Machine Learning techniques offer an alternative for the (complex) modelling approach and still enable an efficient and accurate implementation of the feedforward control. In this thesis, a data-driven approach using machine learning techniques for modeling the inverse dynamics of the manipulator is studied. It is researched which type of algorithms are suited for this purpose. Based on the research three neural network-based machine learning techniques are proposed and developed, namely, LSTM network, CNN-LSTM network and TCN network. A comparison study on the performance of these networks are drawn. Investigations are made on the feasibility of applying these techniques for a feedforward control. In this study, an existing flexure based 2 DOF redundantly actuated manipulator is considered as the real time test setup upon which feedforward control using the proposed machine learning techniques are applied. The machine learning algorithms are first developed on a simulation model of the test setup, later it is developed for the real-time setup. Feedforward control is designed based on the developed algorithms and they are put to test on both the simulation environment and on the real-time system. Several use cases were investigated and experiments were performed. The results showcased that machine learning algorithms are a powerful tool for modeling the inverse dynamics, and the designed feedforward control based on these techniques improved the manipulator’s precision and tracking performance. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 50 technical science in general, 52 mechanical engineering, 53 electrotechnology, 54 computer science |
| Programme: | Systems and Control MSc (60359) |
| Link to this item: | https://purl.utwente.nl/essays/89139 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page