The previous Special article looked at the role of AI behind the self-driving technology. The second series of this Special features the role of AI in auto manufacturing, design and material development.

Collaborative robots work alongside humans

The manufacturing industry is undergoing a significant transformation. Under the banner of Industry 4.0 in Germany and Connected Industries in Japan, for example, the industry is accelerating its effort to become smarter. Labor shortage and the emerging 5G network will also drive the transformation. The automotive sector is no exception, and AI is used in many areas of car manufacturing, including assembly, inspection, predictive maintenance.

In the same way as in other manufacturing sector, robots have been used to collect, move, and sort items, as well as take part in assembly in the auto sector. In today’s assembly lines, wearable robots (exoskeleton) are used to assist human workers and reduce the risk of injury. South Korea’s Hyundai Motor Group, for example, developed a wearable robot called Hyundai Vest Exoskeleton (H-VEX) to assist workers. The H-CEX weighs only 1.6 kg and equipped with two artificial legs. It is endurable enough to support a body weight of up to 150kg, reducing the impacts on workers’ knees.

Thanks to advanced computing power, better algorithms and machine vision, these robots can now handle more complicated tasks. These flexible, non-special-purpose robots can also react to changes in their environment. They monitor the movement of human employees working around them and adjust their motions to avoid injuring humans. In the past, it was necessary to separate an area where robots work and humans work for safety reasons, but with the robots becoming smarter, human and robots can work together. This also gives greater flexibility in the design of factory floor.

Inspection and predictive maintenance with AI

AI is also used in other areas of manufacturing process, such as inspection and welding. Audi uses AI to detect cracks in sheet metal. Conventional approach for crack detection is visual inspection by employees with the help of several small cameras. The cameras installed directly in the presses evaluate the captured images using image-recognition software. However, this process involves a great deal of manual effort because camera has to be reconfigured for every new component produced in the press shop. What is more, the accuracy of the inspection depends heavily on ambient factors such as lighting conditions and surface properties. The conventional mechanism based on simple algorithms of the image-processing program often results in false detection. Audi is Audi is replacing this time-consuming process with fast and reliable system based on AI, and especially deep learning technology.

The team which developed the new system spent months training the artificial neural network with several million test images taken from seven presses at Audi’s Ingolstadt plant and from several Volkswagen plants. Thanks to deep learning, the neural network now learns independently from the examples, and the software can detect cracks even in new, previously unknown images automatically, reliably and in a matter of seconds. Audi is testing this automated component inspections at its Ingolstadt press shop. Audi also aims to apply this system for other visual quality inspection such as paint shop. As in the case of press shop, however, a large number of datasets is required to train the artificial neural network to develop such a system.

At BMW plant in Munich, each blank sheet is given a laser code at the start of production since 2019. The laser code is linked with other material and process parameters, such as the temperature and speed of the presses and the thickness of the metal and oil layer, and the parameters can be cross-checked with the quality of the parts produced. The AI application is also used in the assembly line, determining deviations from the standard in real time. It checks whether all required parts have been mounted in the right place. Predictive maintenance is also introduced in BMW’s welding tongs, but at the Munich plant, sensors have been fitted to all the tongs to measure friction levels three times per shift and report any abnormalities. It is quite costly and time-consuming if the tongs break down during the operation and need replacement. Thanks to the real-time monitoring, maintenance work can be planned in advance, saving time and cost.

Design by AI

The use of AI is not limited in the production and assembly lines. Designworks, a subsidiary of BMW, uses generative design techniques to design new wheel rims and car seats. Generative design is a design exploration process. Designers or engineers input design goals into the generative design software, along with parameters such as performance or spatial requirements, materials, manufacturing methods, and cost constraints. The software then quickly generates various design alternatives based on the design goals and specifications. Generative design has several advantages, for example, the speed of exploring different options and the design human bias and customs. NASA, for example, is using generative design to create a space lander prototype.

AI for materials discovery

AI is also used in material science. In January 2016, Toyota established Toyota Research Institute (TRI) in Silicon Valley as an R&D enterprise with an initial focus on artificial intelligence and robotics. TRI announced in March 2017 that it is investing $35 million over the next four years into collaborative research. Partnering with various institutions and companies, including MIT and Stanford University of the US and Ilika of the UK, TRI is leveraging AI to expand the boundaries of discovering new materials as well as accelerating the pace to material discovery. With the help of AI, Toyota aims to develop smaller, lighter, more powerful and less costly advanced batteries and hydrogen fuel cells, which help the company to achieve its vision of reducing global average new-vehicle CO2 emissions by 90% by 2050.

As one of the outcomes of this collaborative research, TRI announced in March 2019 that it found the way to accurately predict the useful life of lithium-ion batteries before their capacities started to wane. By combining comprehensive experimental data and artificial intelligence, the algorithm can predict how many more cycles each battery would last, based on voltage declines and several other factors. The algorithm can also categorize batteries as either long or short life expectancy based on just the first five charge/discharge cycles. The researchers at MIT, Stanford and TRI made the data publicly available.

From material discovery to design, production and driving, AI increasingly becomes an integral part of the auto industry. In 2018, BMW launched the prototype of Level-5 fully autonomous self-driving car at Mobile World Congress 2018, the world’s largest exhibition for the mobile industry. To what extent the design and manufacturing of cars will be automated when Level-5 self-driving cars are on the road?