Gonvarri Industries MS R&D — Machine vision and deep learning for quality control
Low-cost system for the automation of surface inspection using new machine vision and deep learning techniques
What issues are being addressed or improved?
Gonvarri MS R&D needed to improve the detection of surface defects on steel plates on the pickling line at its plant in Asturias, while minimising operator risk.
Until now, the detection of surface defects was based on inefficient and incomplete manual methods by the worker on the production line.
The aim of this project is to automate surface inspection on the plant’s pickling line within the framework of Industry 4.0 and smart manufacturing, providing a robust and reliable quality control system.
The project is led by Rubén Usamentiaga, from the Computer Systems Self-Management Group at the University of Oviedo.
What digital technology is used to address the problem?
With the University of Oviedo as a digital enabler, Gonvarri has implemented a low-cost system for automating surface inspection using new computer vision and deep learning (deep learning) techniques to detect anomalies.
Combining deep learning and computer vision, the operator can see on a screen in real time the defects in the plates.
The results of the project’s proof of concept covered the end-to-end design of an inspection system, including camera, optics, and lighting.
Another key part is a software tool capable of image acquisition, defect detection, remote viewing and information management. This tool is integrated with the plant, exchanging information with the PLC that controls the manufacturing process.
A real-time visualisation system was also incorporated, allowing both operators and quality engineers to monitor the state of the plant.
What results are achieved and what impact does it have?
The project has resulted in the development of a continuous learning system that automatically identifies relevant images for tagging and generates tasks in a centralised system.
These tasks are tagged by plant operators, generating data sets with tags. Over the course of the project, 525 labelling tasks have been generated, leading to a total of 9,878 images with 9,748 defects of twenty different types being labelled.
From the labelled dataset, an experimental design was carried out to train detection models.
The result is a model that exhibits an average defect detection accuracy of 0.625 according to the F1 metric and 0.590 according to the mAP50 metric, highly positive metrics.
The project thus demonstrated the feasibility of the automated surface inspection process using advanced artificial intelligence and computer vision techniques.
In this link, there is more information about the project of Gonvarri and the University of Oviedo.
Proof of Concept Premiums
This project was one of the three financed through the Gonvarri Proof of Concept 2022-23 Premiums, organised by the SEKUENS Agency, the University of Oviedo and Gonvarri MS R&D.
Through the Proof of Concept Premiums, the Government of the Principality of Asturias promotes a public-private financing instrument to support open innovation models in leading companies in the region.
The aim is to promote proofs of concept that validate in real industrial environments the research results of university groups in the Research and Innovation Areas proposed in the S3 Asturias 2021-2027 strategy, promoting open innovation models.
About Gonvarri Industries MS R&D
Gonvarri Industries was founded in 1958 and has continued its growth by expanding globally and diversifying. It is a leader in the steel and aluminium transformation business, with an emphasis on sustainable, profitable growth and continuous improvement of products and services with the aim of strengthening relationships with customers and suppliers.
In the service centres, the metal is subjected to a series of high-tech processes that enable the product to be tailored to the customer’s needs.
As an expert in the steel sector, Gonvarri Industries defines the following business units: Service Centres, Metal Structures, Material Handling and Precision Tubes.
