100% REWORK, a company that provides services primarily in the field of quality control, has made an investment in the start-up company 24 VISION, which develops applications based on neural networks. The connection with the development company is expected to bring 100% REWORK an even better level of control and expansion into other fields.
The QPAG Group, of which 100% REWORK is a part, provides quality services and products for clients. It supplies measuring technology, calibration and inspection, the latter service, inspection, being provided by 100% REWORK for more than 20 years. Customers today are largely companies in the automotive sector, but it also has clients in the aerospace and marine industries.
“We were looking for a partner with whom we could automate certain inspections demanded by our clients, because some activities require automated inspection and a high level of repeatability in the long term,” says Tomáš Pavlík, Managing Director of 100% REWORK, when asked how and why the companies were connected. He adds that of course not everything can be automated, but it can be a very good complement to quality control.
The company 24 VISION, which develops a system for visual inspection of products and uses, among other things, neural networks for this purpose, attracted Tomáš Pavlík with this interesting idea, but also with its overall effort to find solutions for clients. The cooperation was established in February last year. He joined the company in November.
The companies did not disclose how much investment was involved and what share of 100% REWORK acquired in the start-up. However, it is not a majority stake. Y Soft is the largest investor in 24 VISION.
Solutions for large series
Founded in 2019, 24 VISION builds on the previous years of experience of developers in the field of applications working with image data in the parent company TINT. “There, we gradually moved from developing applications for, for example, license plate reading to the quality area, where we started to perform object recognition using standard methods. However, we soon came across the limitations of these methods – little flexibility, the need to ensure constant lighting conditions, ensuring absolutely accurate product placement, and many others. This feedback from production was a big challenge for us, and one way was to use neural networks. And this is the moment when my colleagues and I founded 24 VISION,” says Martin Hriško, the director of the start-up 24 VISION.
“With the use of neutron networks, control can be even more perfect.”
In a nutshell, the product that 24 VISION offers is a comprehensive AI-driven quality control using neural networks and deep learning. It focuses on the end-to-end quality control of products that leave the production line in large batches and currently have to be inspected by humans. For each product, there is an inventory of visual inspection points, a so-called defect catalogue, which can also vary according to the model or type currently being produced. And it is this variability or diversity that provides the “ideal conditions” for deploying this quality control solution directly into the production process.
Both Tomáš Pavlík and Martin Hriško say that the neural network inspection method is a revolutionary solution that competitors are still only testing, while the 24 VISION product has already been deployed with several customers.
How it works
For a neural network to work properly, it must be trained. The neural network is presented with images with defects or, conversely, images without defects, and these defects are classified into different classes according to type or configuration. These are trained on, tested, the results validated, and as the training progresses, the control reaches higher and higher confidence values for detecting the correct class.
Multiple types and kinds of neural networks run simultaneously, each suitable for something different. Segmentation looks for subjective defects such as scratches or surface damage, while detection detects and recognizes objects such as various buttons and components, and can also count them, for example. The last possibility is that the system has trained a correct pattern and looks for the degree of deviation from this pattern, here it is classification.
The whole solution contains multiple modules that are interconnected and logically linked. The system contains a complete user interface which is used for the actual work of the operator. The operator on the line has an output display where he can see in real time the results of the checks and can react interactively to the defect found. Reporting is then provided for managers, which includes various views and statistics. There is also an archive of all results and images, so there is evidence of the condition in which the product left the line, including links to serial number and time.
It’s just the beginning
“Thanks to the use of neural networks, the limit of control reliability is pushed up to 99 percent and higher,” says Martin Hriško. This is also appreciated by the head of 100% REWORK, Tomáš Pavlík. He says: “For long-term inspection, cameras and high-quality software based on artificial intelligence are a very suitable approach and the future for the industry. I expect that the cooperation with the QPAG group and 24 VISION will enable us to expand our inspection to other industries where quality is a major focus. In addition to the highest quality of service, customers will gain, for example, the traceability of inspections and thus potentially less financial impact in the event of callbacks.”
At the beginning of the cooperation, 100% REWORK will provide 24 VISION with the installation of the system and the production of any inspection stations that will be equipped with cameras and adapted to the application. “If everything goes according to plan, in a year’s time we will have a top quality control solution ready for worldwide expansion,” believes Martin Hriško.
In the future, the two companies want to jointly look for other product packages where human inspection is very difficult and cannot yet be implemented by machine. “Offering such control in a supplier-managed way is one of our goals,” concludes Tomáš Pavlík.
Next articles and interviews
Next articles and interviews