Machine learning is coming to series production to make quality inspection of components even more precise.
Audi, a brand which is traditionally focused on technical innovation, will be amongst the first in the automotive sector to use machine learning (ML) in series production. More specifically, it will be used in the inspection process for sheet metal parts – the software Audi has developed will be able to recognise and mark the finest cracks in record time thanks to artificial intelligence.
Due to the increasingly sophisticated design of its cars and the high quality standards at Audi, the company inspects all the components directly after production in the press shop. Here, a visual inspection is performed by specialised technicians, supported by several small cameras installed directly in the presses which capture images to be analysed with image-recognition software. This process will soon be replaced by the one based on machine learning: it will be able to detect even the finest cracks in each piece of sheet metal with the utmost precision and reliably mark their position, thanks to software based on a complex artificial neural network.
Learning to learn
The plan is to transform Audi’s facilities into smart factories thanks to technologies such as deep learning, a special form of machine learning designed to operate with significant quantities of unstructured data, e.g. images. In the initial phase, the artificial neural network was “trained” with several million test images to create a sufficiently large database. Thereafter, imperfections such as cracks in the images were identified with pixel precision – the highest degree of accuracy possible. The neural network is now able to learn independently from these previously analysed examples, and can detect cracks even in new, previously unknown images. The database consists of several terabytes of test images from seven presses at Audi’s Ingolstadt plant and from several Volkswagen plants.
Moving from cameras to AI
The software was developed almost completely in-house, from the original idea up to the finished prototype. Since mid-2016, the innovation department of Audi IT has been working hand in hand with the Production Technology division of the Equipment and Metal Forming Technology Competence Centre.
The future will see further evolutions of the system: quality inspection using ML will replace the current optical crack detection with smart cameras. This will save a great deal of manual effort. Whether doors, engine hoods or fenders – the cameras currently have to be reconfigured for every new component produced in the press shop. In addition, false positives regularly occur, since the current algorithm is highly dependent on ambient factors such as lighting conditions and surface properties.
In the future, it will also be possible to apply the machine learning approach to other kinds of visual inspections, as in paint shops or assembly shops, for example. All that is needed is a sufficiently large number of labelled datasets.
Source: Volkswagen AG