Multi-modal Defect Detection of Fillet Joints in Gas Metal Arc Welding

Failure in pipelines may have financial and life-threatening consequences. High-quality weld joints in pipelines can avoid these consequences and improve the company’s productivity and reputation. To improve the weld quality, an efficient solution is to detect the defects. A collaboration of human and robots can provide an accurate, robust defect detection system. This system requires sensors to observe and record the information during the welding process. It also requires an Artificial Intelligence (AI) model to distinguish the defective and normal data. This project aims to develop an appropriate AI model to detect the defects in fillet joints during Gas Metal Arc Welding (GMAW) process. We will use a collaborative welding robot to collect sufficient training data including the welding images, sound signals, and welding parameter. Then we will label the data and use transfer learning and fusion learning rules to train an AI model on the prepared dataset.

Faculty Supervisor:

Guy Dumont;Klaske van Heusden

Student:

Partner:

Novarc Technologies

Discipline:

Engineering

Sector:

Manufacturing; Professional, scientific and technical services

University:

The University of British Columbia

Program: