Tactile Based Active Perception of Structural Members in Truss Structures
- Publication Type:
- Thesis
- Issue Date:
- 2020
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Complex Three-Dimensional (3D) truss structures such as power transmission towers require regular inspection and maintenance during their service life. Developing a robot to climb and explore such complex structures is challenging. Changing lighting conditions can render vision sensors unreliable; therefore, the robot should be endowed with a complementary sensory modality such as touch for accurate perception of the environment, including recognising a structural beam member and its properties of cross-sectional shape, size and the grasping Angle-of-Approach (AoA).
The research presented in this thesis addresses three questions related to grasping and touch based perception of beam members in truss structures. (1) Methods for designing adaptive grippers for grasping a wide variety of structural beam member cross-sectional shapes and sizes; (2) Sensing for data collection and methods for classifying beam member properties; and (3) Efficient methods for selecting the next best grasping action to confidently recognise a beam member.
A stiffness constrained topology optimisation design method is developed and applied in designing a soft gripper for grasping a variety of cross-sectional shapes of beam members. The gripper design is verified through both simulation and experiments. It is found that the gripper is proficient in grasping different shapes and sizes of beam members, with adequate contact points.
A comparative study of commonly used machine learning classifiers is conducted to analyse the effectiveness of recognising a structural beam member and its properties. Using data collected during grasping with a soft gripper, the cross-sectional shape, size and grasping AoA of a beam member are classified. Evaluation of the various classifiers revealed that a Random Forest (RF) classifier with 100 trees achieved high classification accuracies, with short training and classification times.
An information-based method for selecting the next best grasping AoA to confidently recognise a beam member is developed. This method is verified through simulation using grasping data collected with a soft gripper. The results show that this method can correctly recognise a structural beam member and its properties, typically with fewer than four grasping actions. This method can be generally used with many different gripper designs and sensor arrangements.
Please use this identifier to cite or link to this item: