Looking for design patterns
Tomorrow’s computers will be capable of “thinking” like a design engineer. Daimler is committed to eliminating repetitive manual tasks. This will benefit the automotive industry – and many other sectors as well.
Share
Computers perform complex calculations and simulations in the area of numerical design – a classical example of rational and efficient interaction between humans and machines. Will artificial intelligence (AI) change this division of roles in the design and engineering sector?
Human expertise is still an essential factor for the evaluation of design results. CAx – the collective term for computer-aided technologies such as computer-aided design (CAD) and computer-aided engineering (CAE) – is evolving at a rapid rate. The next step in digital product development will be AIAx, short for “Artificial Intelligence Aided x”. The goal is to deploy artificial intelligence to digitize the product development process, manage large quantities of complex data and thus eliminate time-consuming routine assignments.
“It is a matter of reducing the number of standard assignments so that experts have more time to reach complex decisions and thus enhance the quality of these decisions. In the latter case we aim to condense the entire sequence of evaluations – including evaluations of co-workers – into patterns which then serve as a basis for decision-making,” explains Dr. Steven Peters, Head of Artificial Intelligence Research at Daimler. This Stuttgart-based carmaker initiated the research project and recruited external partners from industry and science.
More time for innovation
Evaluating design solutions is a challenging process due to the complex criteria and the existence of conflicting goals. “Soft” criteria such as experience, intuition and expert judgement play a crucial role. Everything must be weighed up in order to reach the best possible compromise. So far this has been the province of human experts. These empirical values are difficult to formalize and pass on to the next generation of designers. However with the aid of certain machine learning techniques it is possible to detect and exploit patterns in CAD data. Each individual simulation generates masses of data. At present this data has to be evaluated by human operatives in order to identify flaws in design and initiate continuous improvements. Specially developed machine learning techniques will facilitate smart analysis and generate proposals for potential improvements. Daimler assumes that automated analysis would not only speed up the design process but would also give engineering staff greater time and scope to develop innovative ideas and design solutions. “We believe that the results will be relevant to many high-tech companies in Germany,” says Dr. Peters.
The digital welding assistant
One of them is Endress+Hauser. The company collaborates on the project with Daimler, DYNAmore and USU Software as well as with the Technical University of Berlin and the Karlsruhe Institute of Technology (KIT). “We speed up our own development process by applying artificial intelligence ‒ for example, when a virtual assistant takes on tasks performed by a developer in the CAD system and generates suggestions. A good example is the design of a welding seam. Today, a designer would go to a welding engineer to have a weld seam approved. In future, a virtual welding assistant would help at this stage,” explains the company’s strategic expert Dr. Volker Frey, adding that the partners had also developed an appropriate business model. “During the course of the project we develop machine-learning algorithms which are validated in our application scenarios. The two software houses in the consortium have developed products based on these algorithms which can be sold to other companies because they are applicable to any solution and do not require our specific product know-how.”
With the advent of artificial intelligence it should be possible to equip computers with expert knowledge and harness their decision-making skills to complex simulation results. In other words, human experience will be fed into computers in order to generate usable evaluations. To increase practical acceptance and promote implementation in industry these machine-learning processes must be transparent and easily explicable. Why has my machine-learning system reached this particular decision? “Within the framework of a user survey we aim to investigate various visualization modes for the decision-making process. How should explanations be presented? Which items of information are useful? Explicability is an important factor in the acceptance of machine-learning processes. After all, final responsibility always lies with the designer,” explains Klaus-Robert Müller from Berlin Technical University.
Related Exhibitors
Interested in news about exhibitors, top offers and trends in the industry?
Browser Notice
Your web browser is outdated. Update your browser for more security, speed and optimal presentation of this page.
Update Browser