Development of robotics gripper with variable stiffness joints via numerical modeling and machine learning
Soft grippers in automation, especially bionic finger grippers, hold promise for precise manipulation. However, their complexity makes predicting finger joint bending angles challenging. This dissertation focuses on bionic finger grippers, aiming to improve angle prediction accuracy in light of thes...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Thesis-Master by Coursework |
Language: | English |
Published: |
Nanyang Technological University
2023
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/171245 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Soft grippers in automation, especially bionic finger grippers, hold promise for precise manipulation. However, their complexity makes predicting finger joint bending angles challenging. This dissertation focuses on bionic finger grippers, aiming to improve angle prediction accuracy in light of these challenges.
The choice of a gripper design with three segments and two knuckles was deliberate, as it emulates the complex structure of bionic finger grippers, making it an ideal testbed for investigating the intricate challenges associated with predicting finger joint bending angles. This design mirrors real-world applications and offers valuable insights into addressing the complexities of soft grippers in automation.
This dissertation investigates soft grippers in automation, specifically bionic finger grippers, and the need for an accurate angle prediction model to achieve precise control of finger joint bending angles. The gripper designed in this study uses polymer 3D printing with TPU and PLA materials and is composed of three segments with two knuckles. ANSYS is used to simulate the bending shape of the gripper when pulled to measure bending angles for different material properties. A regression model is built using deep learning techniques to predict temperature, pull distance, and knuckle angle, with the K-neighbour regression model yielding the best results. The model is validated with an average error of no more than 15%. |
---|