On-line learning for robotic assembly using artificial neural networks and contact force sensing

Tools

Lopez-Juarez, I., 2000. On-line learning for robotic assembly using artificial neural networks and contact force sensing. PhD, Nottingham Trent University.

Preview

Text
10183421.pdf - Published version
Download (44MB) | Preview

Official URL: http://gateway.proquest.com/openurl?url_ver=Z39.88...

Abstract

Traditionally, robotic assembly techniques have depended on simple sensing systems and the robot manufacturer's programming language, which has severely restricted the extensive use of robots in complex manufacturing operations. The research reported in this thesis is related to the creation of self-adapting robots capable of learning manipulative skills on-line. This work involves the use of Artificial Neural Networks (ANNs) and contact force sensing to "teach" the robot how to behave in poorly structured environments.

An industrial PUMA 761 robot arm was provided for this research by Rolls Royce & Associates, who are interested in autonomous robot operations. The investigation includes the design of a novel Neural Network Controller (NNC), which is based on the Adaptive Resonance Theory (ART) and a knowledge base, whose knowledge is generated by specific assembly operations.

The research used a force/torque sensor attached to the robot's wrist. This was the only sensory information available to the NNC during assembly operations since the precise location of the components was unknown. The communication with the robot controller was made through a PC master-slave architecture, which provided data acquisition and control in real-time.

The design of the NNC was founded on ART's strength to learn incrementally in combination with a dynamic knowledge base. Initially, the robot was provided with a Primitive Knowledge Base (PKB), which contained a minimum set of primitive contact force conditions and the corresponding motions to reduce these forces. The knowledge is enhanced on-line, based on the success in predicting the motion that reduces the constraint forces. New knowledge information is only accepted in the PKB when it has contributed strongly towards the success of the assembly. The robot actually enhances its overall assembly performance which is measured by a reduction in assembly time. Additionally, mistakes made earlier do not recur, which demonstrates the new expertise acquired by the robot.

The results also demonstrate the generalisation capability of the NNC by learning the assembly of different part geometries using the same PKB. The overall results show the effectiveness of the methodology and clearly define the requirements for implementing the skill acquisition onto other industrial manipulators, hence, providing an important contribution to the creation of new self-adapting robots with on-line incremental learning capability.

Item Type:

Thesis

Creators:

Lopez-Juarez, I.

Date:

2000

ISBN:

9781369316223

Identifiers: