Pro4AI - AI Models for Strength
AI-supported prediction of the failure probability of plastic components
As part of the “Pro4AI” project (Probabilistic Prognosis of Product Propertiesby Artificial Intelligence) project, AI-supported methods are being developed that enable more efficient simulation and more precise prediction of the failure probability of components. In particular, the topics of probabilistic strength prognosis and limit load are being addressed. The project is being carried out in collaboration with the ipf - Institute for Polymer Research in Dresden, Germany.
Deterministic vs. probabilistic simulation
A simulation is deterministic; it is based on the numerical solution of differential equations that represent a physical relationship. In reality, however, product properties are subject to unavoidable variations. In the case of short-fiber-reinforced plastics, for example, scattering of strength due to the statistical fiber configuration, among other factors (Fig. 1).The usual FE simulation cannot capture this, which is why safety factors are used (Fig. 2). These compensate for the scatter in order to avoid product failure. They are therefore chosen generously. This is at the expense of economic efficiency (increased use of material). To overcome these drawbacks, the probabilistic simulation was developed, which takes into account scatter in the material properties and loads. Here, the input data is also subject to a distribution, which then also leads to a distribution of the output data (Fig. 2). This method makes it possible to make more realistic predictions about the behavior of components under load and thus avoid excessive safety factors. However, the more accurate prediction comes at the cost of a considerably higher computational effort. For this reason, the numerical FE model of a RVE (Fig. 1) is being replaced by an AI-based model that works orders of magnitude faster. This means that a prediction of the failure probability can still be made with considerably less effort.
Strength vs. limit load
In practice, the mechanical integrity of components is usually assessed by evaluating a local material failure, i.e. the strength of the material as such. The strength is then determined by the micromechanical structure of the material and its composition on an atomic (metals) or molecular (polymers) level. Failure occurs locally and leads to the material's strength limits being exceeded (Fig. 3, reaching the elastic load), but not necessarily to a total failure of the component. Another approach is to evaluate the component in terms of its ability to withstand a given load situation without losing its load-bearing capacity.
Loss of load-bearing capacity (limit load) refers to the global failure of load-bearing cross-sections due to yielding or fracture, which always leads to component failure (Fig. 3, component cross-section fails). The limit load is a technological failure limit that also depends on the design parameters. An AI model is being developed that considerably simplifies the limit load assessment, as numerically difficult and tedious non-linear FEM analyses up to the plastic collapse range are avoided.