AI-tailored Wish Fulfillment

The first draft rarely is the final one in product development. With the help of artificial intelligence, this labor-intensive but essential stage can be accelerated.

Whether it is a gasoline car or an electric vehicle, an SUV or a compact car, a sports car or a family vehicle – comfort is a crucial factor in the purchase of a car. The damping system, consisting of shock absorbers and additional springs, is essential for this. To ideally adapt the latter to the specific needs of certain cars and achieve the best possible comfort, iterative tuning processes and expert knowledge have always been a requirement. Commissioned by BASF Polyurethanes GmbH, a manufacturer of additional springs in the automotive industry, researchers at Fraunhofer IPK have now developed a process that aims to simplify the design of such additional springs using artificial intelligence. The goal is for AI to derive a suitable component directly from the customer’s requirements.

© BASF
In addition to shock absorbers, such additional springs are essential for a car’s ride comfort. Their design is to be made much easier in the future – with the help of AI.

As a methodological approach to training an AI, the researchers used the CRISP-DM standard process. After an initial situational analysis, the available data is examined, and specific goals are derived. These serve as the basis for training and testing an AI model.

Additional springs are rotationally symmetrical and can therefore be represented in simplified models with few parameters. Compared to other training sets with more complex historical data, this approach significantly simplifies AI training.

Using the simple models, the researchers made systematic small changes. This allowed them to generate over 10,000 different designs from a single model. The team then simulated how these simplified and readjus­ted virtual springs behave under various loads. The data obtained from the simulations served as the training basis for a neural network.

In an initial random sample test, it was shown that the AI model is capable of generating components that come close to the specified requirements. In practice, this means that development teams, for example, could »feed« a new customer’s specifications into the AI and receive a functional initial design for an additional spring. They could then refine the design with detail adjustments.

What our customers say

»The project has shown the limitations of current ML systems and the need for new approaches. The identified solution is not yet final, but it shows a lot of potential and has helped us in our search for suitable ML solutions. A completely finished solution was never the goal, but the result achieved a good match with little input data.« Mario Wolff, Group Leader Product Lifecycle Services, BASF Polyurethanes GmbH

 

Case Study - Semi-automated Design of Jounce Bumpers for Shock Absorbers

Get in touch

 

We would be happy to discuss your challenge with you and present our solutions. Let us advise you without obligation.

 

With the help of applied research, we develop solutions along the entire industrial value chain.