Our second Czech team in Batch 11 is 24 Vision System, a company specialised in using computer vision to observe items on a production line and detect defects as well as configuration errors. If you are a quality manager in a factory, you have one and only one obsession: to make sure that not a single product with error leaves your production line. There could be multiple causes of errors such as plain and simple defects but also configuration mismatch. What is unique about 24 Vision System is their ability to check multiple problem areas in one model and guarantee 99.9% accuracy.
We sat down with Martin Hrisko, co-founder and CEO of 24 Vision System to learn more about how their deep learning models are already being used by popular brands.
Hi Martin! What prompted you to start developing 24 Vision System? What did you see as fundamentally wrong with the way quality managers work today?
We started to develop 24 Vision System as a comprehensive quality monitoring tool for our pilot customer Hyundai Transys. It was used to resolve quality control issues in seat production. Previously, Hyundai used a standard vision system and manual control by quality control staff. However, this classic system had a big problem finding an error in an image with a different visual appearance. This is due to the Just in Sequence and Just in Time production types, where the line alternates products according to customer orders and products have different parameters every time.
We saw the problem as a challenge, and developed and deployed a system core based on deep learning artificial intelligence for test operation in 12 months. We then extended our system by configuration correctness control, where we had to distinguish whether we were looking for an error or a model of the right part of the product.
In practice, the most common quality control is now carried out by an operator who has to check many parameters based on checklists. The operator records these results electronically or manually and then marks the product as OK or NG (No Good). This work is time consuming and tedious. With today’s production dynamics, it’s impossible for an operator to know all product variants – his work performance generally reaches a maximum 70% to 80% success rate. Our system knows all product configurations and also knows where to look for errors. By deploying 24 Vision System, the quality operator gets an immediate real-time result, showing both defects and confusions in the configuration and can only concentrate on eliminating them. The entire process takes seconds, with no errors or mistakes, and with a 99.9% success rate.
How big is the team and why are they a good match for this project?
A team of ten great and enthusiastic people work on the project. The company was founded by three founders whose roles have been distributed well, and therefore complement each other. As a CEO, I’m in charge of business strategy, marketing, finance, management and team motivation. My experience is based on the management of a successful IT company TINT and the 24 Vision System product follows in the original development of this company. Martin Cviček, CTO, leads the research and development and is responsible for the product’s technical development. Martin has many years’ experience in software development. He has spent his entire life in the IT environment, led many large projects, and worked on software automation. Zdeněk Neustupa is a system architect who specialises in programming and new technologies in the artificial intelligence and neural networks fields. We all have one thing in common – we’re innovators who like to bring visions into reality.
What are you trying to accomplish with 24 Vision System that you think no other platform has been able or willing to do?
So far, everyone has focused on solving partial problems and individual defects. This is due to the technology, hardware and the production environment’s complexity. We look at the product from the last operator’s perspective, who has a complete product in front of him and doesn’t know the possible defects or individual components. By ensuring real-time communication with the production system, we find all properties for each product and use our system to compare them with defined patterns in the obtained image. There can be a large number of these patterns and combinations, and the recognition result is delivered to the operator in real time.
Can you tell us a bit more about how your technology really works? How unique is your solution?
We receive visual data from cameras located on the production line. We have all the information concerning the product and its properties for every obtained image. Firstly, it’s necessary to teach our neural network what a defect or the right part of a product looks like. Learning is done by presenting images, where individual areas with patterns are simply marked and then the whole network is trained. When we have a basic model ready, the network will be tutored in the form of operator feedback in real operation in several iterations. The neural network is constantly improving itself until it reaches the required recognition accuracy – typically 99.9% (human control accuracy is around 70%).
New features and patterns are defined by the customer in the 24 Vision System portal and he can gradually add more and more requirements when changing production. If the pattern or defect is the same for different product variants, the network doesn’t need to be re-taught. All you have to do is assign the desired pattern to the product’s properties and that’s it.
Recognition results are displayed in real-time on output displays, which can also be customised by the customer. The quality manager has an analytical and reporting module where he sees breakdowns of individual defects or confusion in configurations for each product at any time. For example, this data is available to perform LEAN analysis.
It sounds like any industry could use a solution like yours. How much of a problem is that for a small start-up? Are you able to focus on one segment or are you serving anyone who comes through your door?
Our solution is universal and can be used in general in any mass production industry. However, we now focus mainly on Just in Sequence or Just in Time production. The automotive sector is a typical representative of this production, where different cars are produced on different lines in different user modifications. We were close to this, as the Hyundai car factory is located close to our headquarters, where we developed and tested our system and gained extensive experience. In the automotive industry, it’s only very difficult to deal with final quality control when the vehicle is completely assembled. Any mistakes at this stage are very costly and sometimes irreversible. Moreover, stopping the production line is very expensive. All automotive suppliers are bound by strict contracts with high penalties for failing to meet a delivery deadline or failing to meet quality standards. This is why we chose the automotive sector as there are high quality requirements and our 24 Vision System system ensures that no defective products come off the production line.
We plan to address other industrial sectors (consumer electronics, food, healthcare) in the coming years. We can’t solve everything at once. Each implementation must be preceded by a feasibility study and rapid prototyping to assess whether our system is suitable for the customer and whether it will be implemented. Experienced consultants, traders and developers are needed for such activities. There are ten of us in the team and we mainly develop our knowledge in the automotive field. We want to become the leading suppliers in the field of comprehensive final quality before we reach out to other industrial sectors.
Can you explain what is special about the Czech industrial market? Why do you think your home market is a good market to start?
The Czech market is strongly focused on automotive, which accounts for around 9% of GDP. We have three important representatives in car manufacturing and hundreds of TIER1 and TIER2 suppliers are connected to them. The Czech Republic has a long history in vehicle’S production and developed cooperation in the research and development field. The feedback that we get directly from production is extremely important to us as a start-up. This allows us to dynamically develop the product and respond to market demands. Slovakia has a similar situation, where automotive is also the largest industry.
So if you are a quality manager in the automotive industry, what do you need to get started with 24 Vision System?
This is the easy part – contact us or one of our partners. Everything is available at www.24vs.io. Together we will analyse any problems that occur on the production line. If rapid prototyping is successful, we will design the entire solution and prepare an offer within six weeks. We offer three variants of our product, depending on the features used – Basic, Advanced and Enterprise. The price includes the license fee, system administration and updates. The price for hardware and implementation depends on the complexity of the install and includes cameras, networking and a server for running the whole system. At the same time, we will train the system operator, who will quickly understand how to work with the system. As a result, the quality manager gets a tool to improve quality, record rejects and detect defective products.
The production manager will have the data to optimize production and carry out LEAN analysis so he will be able to accelerate production and find any weaknesses in the production process. A benefit for everyone is the elimination of tedious jobs, reduced costs and a sense of security resulting from avoiding non-compliance with the supply contract.
Talk a bit about your experience with StartupYard. What were the surprises, challenges, highs and lows?
We have had a very positive experience with StartupYard. It was the right decision from us to approach this company and sign up for the accelerator. Right from the start we were fully involved in all activities and we were surprised by the quality and knowledge of the mentors. The meetings were brisk and intensive; it was difficult, but we acquired a lot of experience in business, presentations and strategy. In addition, we met interesting people who are interested in helping us and sharing their experience. In general, today we are much better prepared to enter the market, we have set a clear business and sales strategy, we are working with real risks and we know what pitfalls await us. All this knowledge will help us become a successful company and establish ourselves on the market.