Specialvideo proposes an incremental workflow to automate the quality control of finished fabrics with Artificial Intelligence.

The neural networks underlying Artificial Intelligence for computer vision must be trained with real images of the product to distinguish defective portions of finished fabrics in the continuous flow of compliant products.
It is true that AI is flexible and can give surprising results on similar products never seen before, but it is also true that the best performance is achieved by providing Artificial Intelligence with images acquired in the field of the same textile industry for which the system is intended.

Specialvideo has designed a progressive path for integrating AI into quality control for continuous textile production, which is structured in implementation and study phases:

  1. Fully operational vision system with linear camera for inspecting solid colours.
  2. Colour acquisition for printed fabrics and weave inspection to train neural networks.
  3. Extension of colour functionality and definition of specifications for print and weave defects.
  4. Updating of the first system and supply of following systems complete with Artificial Intelligence.

Therefore, the first fabric inspection system installed in the factory has two purposes:

  • inspecting solid colours

Solid colour inspection implements monochromatic processing capable of finding stains, holes and other alterations.
Each product non-conformity is stored in the digital defect map, the file that represents the reel by tracking the defective portions of the fabric and their position on the roll.

  • acquiring images of prints and weaves for the purpose of training AI

The same vision system is activated by the customer in colour image acquisition mode for AI training, in order to capture the portions of fabric with defects.
During these acquisition sessions, defects must be marked in advance with adhesive labels.

The vision system is capable of recording counts and statistics useful in the context of Industry 4.0.

To learn more about Specialvideo’s experience in the field of visual inspection with linear cameras, please read the case history of vision systems for PVC rolls.

Colour registration defect in patterned fabric
Colour registration defect in patterned fabric

Specialvideo developed the 3D vision system on board the machine dedicated to reading the DOT code on tires, in collaboration with Uteco Contec

The fully automatic 3D reading cabin detects the DOT code printed on the tire in any position, without manipulating the tire.

The code reading is based on deep learning with a hybrid solution that combines the strengths of neural networks with traditional algorithms: reliability, accuracy and generalization.

The system works at high speeds and without knowing tire diameters and heights in advance.

To learn about the system, we invite you to read the case history and contact us for more information.

Specialvideo presents a Deep Learning quality control for pizzas, that inspects the toppings by returning the count and extent of ingredients classified by type, as well as performing classic checks such as shape integrity, color conformity, and detection of possible blue pollutants.

The application made for pizzas is suitable for other food products that have the ingredients to be checked visibly arranged, e.g., ready-made salad and pasta dishes.

To learn about the system, we invite you to read the case history and contact us for more information.

In the field of industrial quality control, ultra-high resolution images are characteristics of spot checks carried out with optical instruments by specialized operators.

Specialvideo offers the automation of visual inspection at very high resolutions, through::

  • the selection of cameras combined with appropriate high quality optics
  • the study of lighting and background, to enhance the details of the product and its defects
  • the design of the motion scan to inspect 100% of the product surface
  • the design and development of the application code, with classic Computer Vision and Deep Learning, also combined in hybrid solutions based on the strengths of both types of software: reliability, precision and generalization.

The total size of the product and the cycle time, selected by the customer, will be the basis of the design of the scanner, which can be composed of a group of several cameras, which can be matrix o linear cameras depending on the type of inspection.

The following image, deliberately enlarged to allow to see the resolution of 3 microns per pixel, shows a technical fabric for filtering, with a 80 micron thick thread.

Scanner per ispezione del prodotto al 100%: tessuto tecnico per filtraggio con scala risoluzione

In the following figure, a series of images are analyzed by an Anomaly Detection Neural Network which depicts the output in blue for portions recognized as compliant and in red for non-compliant portions:

Scanner per ispezione del prodotto al 100%: tessuto tecnico per filtraggio ispezionato da rete neurale

Specialvideo presents a 2D and 3D quality control for biscuits, crackers, breadsticks and leavened products. It can be integrated into robot lines, at a rate of over 1000 pieces/min.

The vision system for the quality control of biscuits verifies the integrity of the shape of the products and checks the fillings such as jam, cream and chocolate by quantity and position.

The application made for biscuits is suitable for other non-regular-shaped products such as crackers, chocolates, breadsticks and other baked goods.

For products as leavened products, in which height is a critical characteristic, we offer three-dimensional controls.

The system visually checks the cooking level. It detects the presence of any foreign bodies and pollutants that cause color anomalies.

To learn about the system, we invite you to read the case history and contact us for more information.

Specialvideo presents a quality control of Brilliant Uncirculated coins based on learning the model coin, thanks to which it determines the differences that cause discards, tolerating the variability due to the production process.

The concept of compliant currency is learned by the system by memorizing a series of coins free from gross defects or in any case having defects at different points of the coin. From that series, the system elaborates the model coin, therefore it takes into account the acceptable variability, due to the production process.

To learn about the system, we invite you to read the case history and contact us for more information.

When it comes to Industry 4.0, what are the basic ideas and the goal?

The goal is to support and encourage companies that invest in new capital goods and tangible and intangible assets (software and IT systems) functional to the technological and digital transformation of production processes.

Innovation 4.0 does not lie in introducing cutting-edge machinery from a technological point of view, but in knowing how to combine different technologies and thus integrate the factory system and the production chains to make them an integrated, connected system in which machines, people and information systems work together to create smarter products, smarter services and smarter work environments.

Specialvideo vision systems meet the requirements for industry 4.0 and we are available to analyze with you all the aspects, which are already part of our working method, from interconnections, controls, monitoring to remote assistance.

To know the source of the National Transition Plan 4.0, you can follow this link: https://www.mise.gov.it/index.php/it/transizione40

In Specialvideo it is our custom, in receiving visiting customers, to show demos of the vision systems we have created.

The case histories published online are only a small part of the experience we have and continue to gather since 1993 in the computer vision sector.

We choose for you, among the many projects carried out, some of the most similar in terms of needs and industrial sector.

The photo in this article is a preview of a filter quality control, which we haven’t talked about on our website yet. It is an application born in the biomedical sector and characterized by being able to be applied also to different products, in search of molding defects, such as burrs, lack of material, pollutants, and other aesthetic shape anomalies. We will return to talk about this application when we publish the Deep Learning chapter: it is in fact one of Specialvideo’s industrial vision systems that implement Neural Networks. For more details do not hesitate to contact us to request a demo of this or other Specialvideo vision systems.

It is true that the health emergency for Covid keeps us from having as many live meetings as in the past, but the passion in showing our work continues, even remotely.

In person, we are prepared to welcome you with all the hygienic provisions to combat the virus we are all used to.
Telematically, with the call conference and screen sharing platform of your choice, we are available with the same technical insights and with all the enthusiasm of Specialvideo when it comes to computer vision applications.

Personalized machine vision starts with laboratory analysis.

With Specialvideo, a custom-made artificial vision project starts with a feasibility study carried out in our laboratories, based on the customer’s technical information and samples.

The sales engineer, dedicated to your project, searches for the lighting solution that brings out the surfaces and details that need to be analyzed. When the images acquired in the laboratory are satisfactory, the design passes to the software and hardware that will be put into production.
The interaction with the Customer characterizes our method, already from this preliminary study phase. It allows us to produce a reasoned commercial offer, oriented to the required performance.