The Future of Simulation

With simulation models, products can be better analyzed and developments predicted. A research team has developed an AI-based decision support system which increases process efficiency.

The image of the CAD model of a polypropylene component with milled text (left) serves as the basis for the Convolutional Neural Network (CNN). The finite element simulation can be used to train the CNN of the assistance system in order to determine the stress distribution over the component.

When our imagination overwhelms us, we reach for a pen and put down the chaos in our head on paper. By visualizing a complex issue, we keep it manageable –at least for the moment. Today’s simulation models in engineering are built on this fundamental concept. The computer-based models are designed to help us understand the properties of products and predict future developments. Various methods can be applied in simulation models, such as finite element analysis (FEA), computational fluid dynamics (CFD), and discrete event simulation methods (DES). FEA is used, for example, to study the stresses and deformations in a component to which force has been applied. On the other hand, CFD methods can be used to position or deform geometries of components for optimal flow around them. And DES simulation can be used to examine entire manufacturing workflows virtually. What these methods have in common is that they do not require physical prototypes, thereby saving time and money.

However, creating such simulations requires a great deal of experience, and usually involves significant effort. Particularly when analyzing different variants or configurations of a product, the effort required for specific use cases is enormous.


Intelligent simulation …

AI-based decision support systems do an excellent job in assisting engineers with this. With their help, findings from previous simulations can be analyzed and applied to new products with similar configurations. Furthermore, engineers can use similar simulations to perform an intelligent selection of parameters and boundary conditions for a current problem. In addition, substitute models, i.e., reduced digital models, of the simulation can be generated. Researchers at Fraunhofer IPK have recognized the considerable functional possibilities of such intelligent systems and are researching new approaches for assisting engineers even better in their day-to-day work.

In the project, approaches to finding a solution for a use case in product design were researched. Specifically, the project addressed the risk of weakening the material when text is milled into polypropylene components, which may lead to material failure when it is subjected to bending stresses. However, it would have been too time-consuming for the engineers to perform an FE analysis using a simulation model before each milling process, which would have identified weak points in the milling path. Therefore, an assistive system was needed that evaluates the components in advance in terms of their properties so as to avoid time-consuming simulations. The solution was an AI-based decision support system that analyzes the individual components via intelligent image evaluation. A convolutional neural network (CNN), an artificial neural network from the field of machine learning, was used for the decision support system. The CNN was adapted to image recognition for evaluating the image data of the polypropylene components. During this process, the system was trained to determine the different configurations of a component simulation using the resulting image data as well as the corresponding FE results. This resulted in a reduced model, with the help of which the decision support system was subsequently able to recognize – from an image of the component with the text to be milled in – when the component appears robust enough and when it is too fragile for the milling process. The use of AI eliminated the need for a simulation and FE analysis prior to milling. As a result,  faulty components could be identified in real time, allowing correspondingly detailed simulations to be initiated for them.


… Excellent Results

The study demonstrates the potential of image recognition as a replacement for time-consuming finite element analysis in simulation models. One thing is clear: The possibilities for decision making with AI and simulations are manifold. In the future, not only will FEA
be replaced by the decision support system, but the system’s rate of success will also be continuously optimized with the aid of a hybrid substitute model based on real-world data during the milling process. A digital twin, a fully-fledged substitute model, is also being planned. The advantages of AI-based decision support systems cannot be denied: they increase efficiency and significantly reduce costs.