Extruded Tube Surface Defect Detection Using Machine Learning



Hydro Norsk, a leading company in the steel industry with worldwide operations and factories, faced a critical concern of early defect detection in their production pipeline. The presence of even minor defects resulted in significant waste of raw materials. Although Hydro had an existing solution based on real-time image collection of their production pipelines, they identified the potential for improvement using contemporary techniques. Consequently, they approached our team to investigate and explore new solutions.


By leveraging our deep experience in AI-powered image processing, we started to design and develop an algorithm capable of accurately identifying the various types of defects in their product. Deploying the solution in the initial pilot factory introduced a new set of challenges.The data processing pipeline needed to be exceptionally fast, making intelligent decisions about which data were worth analyzing to enable prompt defect detection. This was crucial to ensure swift actions could be taken when necessary. The data processing pipeline also needed to be resilient to failures, ensuring it continued to function even in the face of unexpected glitches or issues. Additionally, the output of the data processing pipeline needed to be readily accessible via a real-time dashboard. This dashboard would highlight problematic patterns in the production lines, providing insight into deeper issues that required attention.


Our collaboration with Hydro resulted in a substantial enhancement of Hydro's defect detection capabilities. The project's success can be attributed to the combination of Hydro's industry expertise and our proficiency in AI, image processing techniques and interactive dashboard creation.