Recent research has shown the potential of machine learning based constitutive models. Herein, classic plasticity modeling frameworks are replaced by e.g. neural network based approaches. Data-driven models are a promising method to model complex mechanical behaviours and represent an important field of research for the laboratory.
Machine-learning based approaches can be used to replace complex hardening laws. The stress response of randon and non-monotonic strain paths can be accurately described by neural network models.
Forming Limit Curves (FLC) are used widely in sheet metal forming to predict the onset of necking in a material. Using neural networks, for proportional pre-straining histories the evolution of the FLC can be predicted.
Crack detection in mechanical experimets can be a tedious and user-dependent task. Using optical methods together with neural networks automated crack detection can be carried out in an objective manner.
Over the past years, there exist growing efforts to determine the effective properties of more and more complex structures and materials. Ultimately, aiming at bridging the gap between observable measures and mechanically significant quantities. Neural networks help to directly translate geometrical features into mechanical properties and effective material behavior.
Thermoplastics exhibit a complex mechanical response for a range of strain rates and temperatures. Leveraging the potential of neural network, accurate constitutive models can be developed.
The figures below show the material (dashed) and the model (solid) response for a range of strain rates and temperatures for a Polypropylene (PP).
