Make no mistake, we are only at the very beginning of exploring the full potential of AI across all industries. Data Science, Machine Learning, Artificial Neural Networks, Text Mining – all of these technologies, already partially mature in the online marketing and financial worlds, have much to offer manufacturing in general, and the automotive industry in particular.

From the design and development phase through to testing and production to marketing, AI has applications throughout the automotive lifecycle. The data generated by the many sensors now embedded in vehicles, extracted from production lines and compiled from customer feedback are powerful sources of information. Their analysis and interpretation provide equally powerful levers for improvement in design, testing and maintenance; as well as for understanding user needs and expectations. Looking further ahead, the challenge – as tricky as it is inspiring – is naturally the development of autonomous vehicles and the full delegation of all safety-related decisions to the vehicle itself.

New functions for new user needs

Research into the development of smart vehicle technology is focused particularly on the issue of environmental perception: infrastructures, other vehicles, pedestrians or any other object that could be considered as an obstacle to a car. Radar, sensors, cameras, weather conditions, roadworks and other extraordinary events: the machine must be able to recognise every type of external influence and evaluate its possible effect on the trajectory of the vehicle in order to make appropriate corrections to the driving control system in real time.

Expleo is currently working on this issue through the development of an autonomous parking solution called AVP (Automated Valet Parking). Using an app that connects car, driver and infrastructure, this solution allows the vehicle to enter, leave and park unassisted in an underground car park. This in-house innovation is built around image processing technologies based on Deep Learning, two Yolo-type algorithms and semantic segmentation. The combination of these components allows the vehicle to recognise its environment, detect obstacles and behave appropriately in autonomous operation.

Fault prevention and correction

The multiplicity of data available during the testing phase provides access to information that can prove extremely valuable for fault resolution. All you need to do is be able to extract the data. By detecting faults within large volumes of data, algorithms leave engineers free to focus on data interpretation and fault resolution, rather than searching for source information. This means that methods known as clustering and classification can be used during road testing to analyse and qualify vehicle responses. Using data gathered by in-vehicle sensors, it becomes possible, for example, to identify untimely braking scenarios, understand their causes and ultimately correct them. Without the algorithms, it would be significantly more complex to make use of the data. So it is clear that AI does not remove the need for people. On the contrary, it refocuses their expertise.

Capitalising on customer knowledge

If there is one area in which the effects of big data are particularly well known, it is end-user customer knowledge. Consumer data analysis applications are among some of the most mature, and are used by brands to identify their target audiences and the expectations of those audiences. This approach is a direct response to increasing demand for product and service personalisation. In the automotive industry, customer knowledge can be applied to improve component reliability. For example, Text Mining makes it possible to analyse free text data contained in customer feedback gathered via e-commerce websites, forums, etc. Recurrent fault data can be used to revise the design of particular components and avoid the need for recall campaigns.

What about the future?

“One of the major challenges will be validating the safety-related decisions taken by autonomous vehicles. Currently, the on-road use of any vehicle is subject to its ability to demonstrate that it complies fully with a series of predefined safety standards. But in the context of autonomous vehicles, safety will be ensured by AI. So although artificial neural networks are currently delivering promising results, like the capability to respond adequately in emergency braking situations, these results can neither be demonstrated nor guaranteed. So do we need to evolve the current safety demonstration standards as a result? That’s a question that remains to be answered.” David Renaud, Head of Data Science at Expleo

AI and mechanical design : a developing symbiosis

“Despite its very many recent successes in traditional applications, such as image, speech and natural text processing, Artificial Intelligence has yet to enter the field of Mechanical Design or Digital Simulation in a more general sense. The reasons lies simultaneously in their tools – data versus physical equations – and the expertise of designers and mechanical engineers, which is challenging to interpret mathematically. Nevertheless, there are many applications for AI, and especially Machine Learning, in mechanical design; examples include building low-cost approximations of design process calculations quickly. Issue detection techniques based on Machine Learning are now used to identify damage and inspect structures. Used in conjunction with Machine Learning, global optimisation makes it possible to design better structures more quickly than the traditional trial-and-error approach by exploring a more expansive design space. Beyond these known applications, AI will release the creative potential of mechanical engineers by assisting with, and facilitating, certain time-intensive tasks, such as the construction or reconstruction of parametric CAD models. Generative Adversarial Networks will automatically generate increasingly realistic and powerful 3D designs. Lastly, available storage capacity and the distributed processing of big data will make it possible to chain the many calculations required to design a component. So a great deal to look forward to, then!” Dimitri Bettebghor, Data Scientist at Expleo

An industrial chatbot

Expleo has leveraged its expertise in data science to develop an industrial chatbot that could eventually be used to assist operators on automotive production lines. Developed out of Deep Learning and Machine Learning technologies, this innovation will be able to understand and answer natural, non-standardised human questions.