The team consists of Béla Paragh and Tamás Kovács is part of the Experimental Engine Production Centre (in German the Motoranlaufcenter (MAC)), yet in their work they function more like an independent start-up workshop.
The idea of establishing a development team within the MAC emerged a few years ago, and became a reality in 2019 by expanding what had been a one-person field. That year, Béla came from maintenance as a mechatronics engineer, and Tamás, a qualified mechanical engineer, arrived at the same time, and the two of them now make up the team.
What gave birth to the Makerspace concept?
Innovative technological developments that combine professional experience, in-house development, market intelligence, innovative solutions from competitors and service providers, and seek unique solutions to challenges that external partners could not solve or could only do so at a significantly higher cost. The creative, development work that is often difficult to fit into the framework of a multinational company does not produce return on investment figures based on Excel spreadsheets. The time invested in research and development and the knowledge gained there will be profitable for the projects implemented later. It is up to the professional competences of the development team to decide where to look for projects to solve, to introduce techniques already developed in the factory or to turn towards new technological directions. Béla and Tamás have mainly deepened their knowledge in machine vision, camera systems and industrial acoustics, where they have recently successfully implemented two artificial intelligence-based projects in engine and vehicle production, adding to the 20-30 previous projects based on conventional camera technology.
What are the benefits of AI-based solutions?
Tamás: More complex problems can be solved than with traditional methods. When processing images or sounds, artificial intelligence comes up with results based on small details that we wouldn't even think of. It can better track changes in the environment, from lighting conditions for camera work to background noise for acoustic projects. In general, it requires less manual programming and yields higher accuracy. It is important to note that this cannot guarantee 100% verification accuracy, nor can any other verification method based on image processing.
How does the process of acoustic analysis based on artificial intelligence work? How did you develop the method?
Béla: In both the vehicle and engine plants, there was a need to find a way to check the correct installation of the “plugs” (electrical connectors) during the installation of electrical connections, i.e. to filter out and identify the clicking sound of the locking tab in noisy conditions. Since the parameters are constantly changing, a forklift truck passing by, people talking in the background, etc., it was necessary to involve artificial intelligence, which can achieve efficient results under unconventional conditions, i.e. changing environmental parameters, after learning the patterns in the signal. In the field of speech recognition, AI-based voice recognition technology was already available, but acoustic signal recognition had not yet been applied in this industrial field. We could not find any working solutions for this either within the Group or outside it, so we developed this special technology in cooperation with acoustic experts from Széchenyi István University.
Throughout the project, we validated our concept after countless factory tests and proved that our method could work, in a way that is unique in the industry. We implemented our first series project on the MDB-EVO engine production line, where it was a natural expectation for line workers to get immediate feedback on the correctness of the assembly. Therefore, a signal lamp and stopping the workpiece ensure correct assembly monitoring. For data security reasons, cloud-based systems were out of the question. So the AI models run locally in an on-site device integrated into the production line, thus enabling acoustic voice recognition under industrial conditions.
In addition to voice recognition, you are also using AI for image recognition. What need gave rise to the method during the assembly of the tank covers?
Tamás: One of the main indicators of Audi Hungaria's flexibility is that several models are produced on the same production line. Currently, five models, the old and new generation Audi Q3 and Q3 Sportback, as well as the Cupra Terramar, are produced in parallel. Consequently, two reasons justified the use of AI in our project at the paint shop to check tank cover preparation. The first is that very similar parts that are hard to distinguish with the naked eye need to be handled correctly by the staff next to the line. The second is the ever-changing working environment, because although the lighting in the workplace itself is stable, the movement of the employees can affect the illumination of the part, and may partially obscure the part to be inspected. This phenomenon, combined with the slightly variable position and the small variations in the tank cover manufacturing technology, justified the use of AI.
How was the method developed and how does the process work?
Tamás: Very simply put, "teaching" AI is like teaching small children about different fruits. We show them a variety of pictures of apples, pears, bananas in different situations, colours and light. Convolutional neural networks learn on their own what the main features are and what makes them different. As with all such procedures, the first step was to have the right amount and quality of data. These were collected under production-line conditions and the data set was prepared for teaching. Once the model worked well in the lab and the software of our embedded system was tailored to the project, we were ready to test it in production. During operation, the employee receives immediate feedback of the faulty assembly, indicated by a red light, plus the system intervenes to stop the body from moving down the line until the fault is corrected.
The process of innovation and development never stops, of course, and today the data is being gathered and the supplies are lined up in the Makerspace for the next unique solution. Béla and Tamás are also keen to move towards AI-based robot control and keep themselves at the forefront of technological advances. Like creative thinking, their tasks are hard to limit when they get into the flow, to use a fashionable phrase, they might keep thinking about their projects at night, while running, or even during their holidays, noting down their ideas whenever they come. The phrase in the title, “The right people in the right place”, really makes sense in their office, their workshop, where you can feel the harmony and dedication to their work every minute.
