
5 minute read
SYMPOSIUM MARKS SUCCESSFUL CONCLUSION OF MACHINE VISION PROJECT FOR QUALITY CONTROL
On 17 September, around 80 professionals gathered at KU Leuven-Bruges Campus for the symposium on Machine Vision for Quality Control. The symposium marked the conclusion of the VLAIO TETRA project “Machine Vision 4 Quality Control” where researchers from KU Leuven Bruges Campus and Vives University of Applied Sciences in Kortrijk worked for two years to transfer their practical knowledge of machine vision to companies in the manufacturing industry. Project coordinator Matthias De Ryck reflects on the project and the final symposium.
Machine vision is essentially the technology for capturing and analyzing digital images to extract useful information that can be applied in industrial applications, such as defect detection, product measurement, or counting,” begins dr. De Ryck. “While computer vision focuses on developing high-performance vision algorithms, machine vision is centered on the practical application of these algorithms to solve industrial problems and the selection of the right hardware and software for optimal results,” Matthias explains.
The TETRA project specifically focused on using machine vision for quality control, addressing the needs of the manufacturing industry. “In the manufacturing industry, products must meet specific, often strict, quality standards. Currently, many companies still rely on manual visual quality inspections. These manual checks are often subjective and inconsistent. Additionally, products may move at high speeds, making it difficult to detect errors,” says Matthias. Despite these manual inspections, many products of insufficient quality still reach the customer, leading to additional costs and damage to the company’s reputation.
Machine vision techniques that assess quality through camera images can make these inspections faster, more consistent, and more dependable. Thanks to extensive innovative research, machine vision technology has already advanced significantly. “However, we often see that its practical application is lacking in the industry. This is partly due to a lack of internal knowledge or resources, or because the current market offerings do not fully address the complexity of the challenges faced by the industry.”
Emerging technology
Through the TETRA project, the researchers bridged this knowledge gap by working on six industrial use cases over two years. The results were shared with a wide audience of companies through workshops, masterclasses, and interim project updates. “We provided companies with the necessary knowledge, inspiration, and tools so they could start integrating machine vision into their applications. The final symposium also contributed to this,” says Matthias.
“Of the six use cases, the biggest challenge was measuring the fat percentage in donuts after frying, as variations in fat content are simply not visible to the naked eye or to a standard camera. For this use case, we turned to emerging hyperspectral technology. A regular RGB camera collects information in three color bands (red, green, and blue), just as the human eyes do. These cameras can only detect how much red, green, and blue is reflected by an object. A hyperspectral camera, however, can divide the light spectrum into many more bands and measure the intensity of the incoming light in each of these bands. This allows the camera to detect how much of a specific wavelength is reflected by the object.
Depending on the (liquid) substances present, the reflected wavelength varies. In this way, we were able to detect fat,” Matthias explains. To calibrate the camera, the researchers collaborated with a professional baker who made dough balls with known fat percentages. The researchers analyzed which part of the spectrum changed as the fat percentage increased in the measurements taken by the hyperspectral camera. “This changing part of the spectrum provided us with information about the fat content in the donut. By mapping this correlation, we were able to predict the fat content of other dough balls based on the hyperspectral image,” says Matthias.
At the final symposium on 17 September, Professor Lien Smeesters (B-PHOT, Brussels Photonics) gave a keynote on the potential of spectrometry and hyperspectral cameras in food and manufacturing industries. Additionally, the project results and developed use cases were presented. “During the symposium, we introduced participants to the market offerings in the field of machine vision through, among other things, a mini expo where various technology providers and system integrators displayed their products, demo materials, and successful use cases. Participants also had the opportunity to test the demonstrators we developed during the project, giving them hands-on experience with the practical implementation of machine vision,” says Mathhias.
Next Steps
Matthias and his fellow researchers are already looking ahead to the next step. “With the implementation of machine vision, more production errors are detected during quality inspections, but ideally, you want to prevent these errors in the first place. To avoid these errors, you first need to understand how they occur. This involves detecting the errors, linking them to the process parameters in place at the time, and identifying which parameters cause the quality deviations. By analyzing this data, we can find correlations and, for example, support operators in adjusting the process parameters more effectively to prevent quality reduction.” An application for a follow-up project (COOCK+ project) on this topic has already been submitted. “If the follow-up project is approved, it will likely start in January 2025. Professor Mathias Verbeke, who was also involved in the TETRA project on machine vision, will lead the project from KU Leuven,” concludes Matthias.
Pauline Van Springel
www.iiw.kuleuven.be/onderzoek/m-group
www.mv4qc.be
