Audi uses machine learning software to look for defects in fresh parts of car bodies | Robotics
Manufacturing cars is not an easy business. Thousands of different parts go into making a single car and they have to be manufactured to the company’s standards. Audi is known for its high standards for quality and modern sophisticated designs. Now Audi is becoming one of the first automobile manufacturers worldwide to apply machine learning in series production.
Car body parts are made in special presses that stamp sheet metal into various shapes needed to construct a car. These presses are very powerful and sheet metal is relatively thin, which is why sometimes cracks may appear. Audi’s employees check for cracks and there are some cameras using image recognition software that are scanning fresh parts for defects as well, but now this function is going to be taken over by a complex artificial neural network. This software will reliably recognize the defects and mark them. This will be based on machine learning algorithm, working on unstructured and quite big pools of data in a form of images.
Of course, creating this type of software presents a series of challenges. Engineers had to create a big database for the algorithm to learn from. And then they had to mark cracks in images with high precision so that the machine could learn the task. These efforts paid off, because now this software can detect cracks in previously unseen images. This could speed up the quality control and help workers in these press shops. Artificial intelligence is a huge part of the future strategy of Audi, because it could improve manufacturing efficiency as well as the quality of the finished product.
Frank Loydl, Chief Information Officer at Audi, said: “In this cross-divisional project, we are jointly developing a production-ready solution that Audi will use exclusively in the company and which is unique in the market”. Interestingly, this software was mainly developed in-house. It requires a tremendous amount of effort, but the result is worth it, because current cameras have to be repositioned for different parts of the cars body. The new AI system will allow keeping the camera stationary, which will reduce the time needed for preparing the system. Furthermore, current cameras are prone to false detections, because they depend on ambient factors such as lighting conditions and surface properties.
Currently automated component inspection system is being tested in series production at our Ingolstadt press shop, but later is will spread through Audi’s facilities. Further developments of machine learning algorithms could make it useful in other areas of car manufacturing as well, such as paint shops or assembly shops.