Speaker
Description
Indentation is a versatile method to assess the hardness of different materials along with their elastic properties. Recently, powerful approaches have been developed to determine further material properties, like yield-strength, ultimate tensile strength, work-hardening rate and even cyclic plastic properties by a combination of indentation testing and computer simulations. The basic idea of these approaches is to simulate the indentation with known process parameters and to iteratively optimize the initially unknown material properties until a minimum in the error between numerical and experimental results is achieved. Such inverse methods have been shown to work in a robust way for macroscopic hardness tests, for which the indenter is large compared to the microstructural length scale. However, the repeated finite element (FE) simulation of the indentation process with a number of different combinations of material parameters is a tedious and time-consuming effort. In this work, we first confirm for some materials the accuracy of the FE simulations and the appropriateness of the chosen material model by comparing measured load-displacement curves from indentations with a spherical indenter of 30µm radius with that of the simulations. Then we investigate the use of machine learning to render this optimization procedure more efficient. The machine-learning algorithm is trained with data obtained from FE simulations of indentations with various combinations of material parameters covering a certain range of material properties. Once, the training is completed, the machine learning algorithm can serve as numerically efficient surrogate model and, thus, replace the FE simulations. The advantage of this approach is that the training effort occurs only once, and then the machine learning algorithm can be used to execute the inverse methods for different materials, whose properties, however, must lie within the range of the training parameters.
| Speaker Country | Germany |
|---|