AI based Twin Arm Robot for Textile Fabric Analysis

Abstract

The use of Artificial Intelligence together with automation is increasingly becoming the new norm in almost all aspects of life. The Textile and Composite industries have been lagging in this field and a huge gap has been identified that can be filled by AI, in particular computer vision to aid automation in fabric analysis during manufacturing and quality assurance. For this, a twin arm robot has been developed that incorporates computer vision and various sensors to assess and analyse fabrics. Computer vision is key in making automation flexible, where it is rigid and cannot be easily changed for different types of materials. This work endeavours to use artificial intelligence, in particular deep learning and object tracking techniques, to identify salient features of fabrics. It is also used to make decisions and predictions on a range of attributes, positively to recognise, classify and track the object of interest, such as a particular fabric landmark or feature, up to macro scale. This information can subsequently be used to assess fabrics for strain measurement, despite the compliant behaviour of joints, defects and mechanical properties. This can provide immense flexibility to the overall system in its capacity to handle a range of fabrics of distinct characteristics. This work starts by using computer vision techniques to understand the nature and features of the image. The work also includes low level image processing methods to match an image and detect the presence of features of interest in an object. Subsequently, it deploys deep neural networks, in particular convolutional neural networks to classify an image as belonging to a certain class of fabrics. The network complexity is increased to segment the fabric and detect anomalies or regions of interest, which might result in the positive detection of certain areas within a larger image containing many different or multiple instances of the same objects. Equipped with the above discussed abilities, a twin arm robot has been developed from scratch as part of this research. It has been used to track areas of interest, conduct automatic calculations of shear angle and tensile properties of the fabrics, measure buckling rigidity, heat flux through materials and compression and friction properties of fabrics. The focus has been to prove that the developed system, equipped with vision, has the ability to overcome the joint compliance in the robot (arm assembly) and give accurate results for strain measurements. The fabric handle measurements on the robot are compared to the SDL Atlas Fabric Touch Tester. This research validates the premise that vision significantly reduces noise and enhances accuracy of tensile properties and strain measurements. It also demonstrates high correlation between heat properties, fabric friction, roughness and compression, measured on the twin arm robot and the Fabric Touch tester. Given the integration of such AI based vision systems in textile fabric analysis, this research provides a step change for objective testing and analysis of fabrics.

Date of Award	23 Feb 2024
Original language	English
Awarding Institution	The University of Manchester
Supervisor	William Sampson (Supervisor) & Prasad Potluri (Supervisor)