PlauSim: Plausibility check for simulations with enhanced transferability and reduced application limitation

• Applicants: Dr.-Ing. Stefan Götz; Professor Dr.-Ing. Sandro Wartzack
• Subject Area: Engineering Design, Machine Elements, Product Development
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 456585803

In the current industrial environment, linear finite element simulations accompanying design are often carried out by product developers and not exclusively by calculation engineers with several years of professional experience. This leads to frequent iterations in the product development process and can lead to incorrect decisions based on inadequately confirmed results. An automatic plausibility check for linear structural-mechanical FE simulations is an important method to support product developers. The use of Convolutional Neural Networks (CNN) and machine learning methods offers enormous potential for recognizing correlations in data and setting up a model with high predictive quality. In the first phase of the research project, it was shown that a plausibility check for FE calculations is possible using deep learning and CNNs. The prediction quality was further increased by adapting the CNN architecture and hyperparameters, and local areas of the FE simulation were also investigated, particularly to recognize numerical errors such as singularities. In addition, a method was developed that enables a specific prediction of the cause of plausibility so that better feedback can be provided to the user. The second phase of the research project aims to implement a generalized plausibility check of FE simulations and thus ensure even better comprehensible support for users with less effort. The basis of the entire procedure is the data set from the first phase, which serves as the foundation for the new project. In addition, the necessary unknown test data must be generated so that a comparison of the transferability to unknown simulations is possible at all. The first step is to develop a similarity measure for FE simulations, which can be used to compare the quantitative analysis and describe the limits of achieved transferability. These two data sets will subsequently be used to analyse the factors influencing transferability to unknown simulations. Following this, the focus is on the detailed reasoning of the plausibility results by coupling the trained model with ontologies. The necessary need for training data is also part of the investigations in the project, which is why the reduction of the training data and the necessary labels is also worked on in a separate work program. At the same time, all newly developed methods will be incorporated into an assistance system, which will be constantly updated and evaluated in line with the current state of research.

• DFG Programme: Research Grants