Automated Welding Quality Inspection and Component Rejection System Using AI and Computer Vision
Participants:
ORKLI
LORTEK
Programme:
Sectors:
Automation
Digitalisation
Oil&Gas
Technologies:
Digital platforms
Computer vision
Scope
To implement an inspection system for thermocouples and magnetic groups in Orkli’s Gas Safety range, critical components that require rigorous quality control. Traditionally, this inspection was performed visually by operators, a labor-intensive process due to the small size of the parts and the high production volume.
The new system integrates computer vision into the production line, enabling the automated detection of defects and the immediate rejection of non-compliant parts. Using artificial intelligence algorithms, welding quality defects—such as lack of fusion and porosity—as well as geometric irregularities in the weld are identified.
Solution
Quality inspection is carried out at a post-welding inspection station, adapted to the type of component and production line. For thermocouples, two types of inspection are implemented across two different production lines, while for magnetic groups, a single type of inspection is used, always assessing welding quality.
The design of the computer vision system is customized according to the characteristics of the part and process. Suitable cameras and lenses are selected to capture a very small area, typically between 2-4 mm, depending on the part-handling mechanism. In some cases, up to three cameras are used to obtain a complete image of the weld while the part remains stationary, or a single camera captures five images using a mechanism that moves the part. Additionally, lighting plays a crucial role in achieving high-quality images, especially with metallic materials that may cause unwanted reflections. Therefore, diffuse LED lighting, provided through a dome, is used to minimize glare and ensure optimal image capture.
To detect defects, Deep Learning algorithms are trained using a dataset of approximately 1,000 labeled images of real defects, previously captured by the vision system. These algorithms can identify welding defects such as lack of fusion and porosity, as well as non-conforming geometric characteristics.
The developed software integrates the entire process: from image capture to analysis through the defect detection algorithm, including a data visualization platform. Additionally, continuous communication is ensured with the production line’s PLCs and automation systems, allowing for the automatic rejection of defective parts.
The initial system developments were implemented in various projects and have been replicated across multiple production lines, maintaining the same inspection system at the Ordizia plant, as well as in Brazil and China. Currently, 21 inspection systems are successfully operating in these plants.
