Systemizing the load case definition for component tests

Component tests are expensive and time-consuming. Therefore, those tests should be performed in such a way that a maximum benefit out of the measured data can be achieved.

Crashing a component — the smart way!

Within the AIMM project (Artificial Intelligence for Material Models), new material models based on machine learning (ML) algorithms are developed. Those models give for each strain input the corresponding stress output. We use component tests to validate the models, in the best case at all strain states that can occur theoretically. Therefore, such tests are preferred, during which as many strain states as possible can be observed in the component during loading. 

Development of a tool to rate different tests

Fraunhofer EMI developed a Python tool that can be used to rate different tests concerning the strain states that occur in the component, before only the most suitable test is conducted for real. For this purpose, simulations of the tests are performed. The tool then calculates several characteristic numbers out of the simulation data for each test. One example is the relative area coverage of the strain space, which is the ratio of the area covered with data points to the total area of the strain space. The higher this number, the more suitable is the corresponding test for validation.

Another application of this tool is the geometry optimization of specimens used for the generation of experimental training data for ML material models. The aim of such an optimization is to obtain training data, which provides information for the material model at many different points in strain space