Project Details
Projekt
Fracture mechanism in heterogenous structures – combining microstructure of EnAM and fracture/ liberation behavior
Applicant
Professor Dr.-Ing. Urs Peuker
Subject Area
Mechanical Process Engineering
Term
since 2021
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 470551727
The mechanical processing of slags containing EnAM requires the liberation of the EnAM grains prior to the physical sorting. As the structure and material properties of slags with amorphous and crystalline structures differ from those of mineral ores, it is necessary to take a closer look at their fracture behaviour, with particular emphasis on liberation. The latter describes the generation of single phase particles, either matrix or EnAM material, during comminution. The overall idea is to generate a new process model that combines information about the internal heterostructure of slags containing EnAM particles with their macroscopic liberation behaviour. It is known that the internal structure of a mineral or slag particle can interact with a crack during comminution. The liberation efficiency depends on whether the crack propagation is completely random or influenced by local material properties. The key point is to analyse the fracture plains within the heterogeneous structure of the slag containing EnAM to find a relationship to the cracking mechanism. The latter determines the liberation. Since the entire fracture plain can only be resolved from 3D images, its identification and quantification is based on computed tomography (CT) data. A series of slags containing EnAM particles are crushed in-situ in the CT, allowing both the initial structure and the resulting debris particles to be captured in 3D. The images of the debris particles can be virtually rearranged to identify one or more fracture plains within the heterostructure. The process model has to correlate complex information, mainly describing the internal structure of the initial particle and its debris with regard to liberation. A direct analytical correlation of the data is not possible, so the structure of the particles and the fracture plain is described by proxies. The proxies are either known geometrical key parameters or abstract parameters that can be created by aggregating the image data. This aggregation uses so-called autoencoders, which compress the image data into an abstract pattern. Both types of proxies are used as training data for a machine learning approach that aims to correlate the initial structure of the slags containing EnAM with the liberation mechanism and liberation performance. Since experimental training data is limited, artificial images are generated either based on statistical information or in collaboration using the structure models of the DEM simulations in the SPP. The artificial data provide the opportunity to train specific fracture phenomena defined in the literature but not yet observed experimentally. The validation uses 3D image data of Li-bearing EnAM slags. As the machine learning model works with proxies, it will be evaluated whether the model is also able to use proxies from less sophisticated, i.e. 2D image, sources.
DFG Programme
Priority Programmes
