Machine learning

Audi looks to revolutionise the testing process in production with artificial intelligence.

Audi is promoting artificial intelligence in a new, quality control application.

22 October, 2018


 

Artificial intelligence is entering the production and quality control sphere at Audi , thanks to new software developed by the brand that recognises and marks the finest cracks in sheet metal parts – automatically, reliably and in a matter of seconds.

With the increasingly sophisticated design of its cars and the high quality standards at Audi, the company inspects all components directly after production in the press shop. As well as visual inspection by employees, several small cameras are installed directly in the presses, which evaluate the captured images with the help of image-recognition software. This process will soon be replaced by the new ‘machine learning’ (ML) procedure – the software based on a complex artificial neural network that operates in the background of this innovative procedure. 

The software detects the finest cracks in sheet metal with the utmost precision and reliably marks the spot. The solution is based on deep learning, a special form of machine learning that can operate with very unstructured and high-dimensional amounts of data such as with images. The team spent months training the artificial neural network with several million test images. The biggest challenges were on the one hand, the creation of a sufficiently large database, and on the other hand, the so-called labelling of the images. The team marked cracks in the sample images with pixel precision – the highest degree of accuracy was required. The effort was worth it because the neural network now learns independently from the examples and detects 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.

In the future, it will be possible to apply the ML approach also for other visual quality inspections. If a sufficiently large number of labelled datasets are available, the system can also support paint shops or assembly shops. The sky is the limit.

Machine learning will ensure even higher standards of finish