Big data for microstructure-property relationships: a case study of predicting effective conductivities

The analysis of big data is changing industries, businesses and research since large amounts of data are available nowadays. In the area of microstructures, acquisition of (3D tomographic image) data is difficult and time-consuming. It is shown that large amounts of data representing the geometry of virtual, but realistic 3D microstructures can be generated using stochastic microstructure modeling. Combining the model output with physical simulations and data mining techniques, microstructure-property relationships can be quantitatively characterized. Exemplarily, we aim to predict effective conductivities given the 1 microstructure characteristics volume fraction, mean geodesic tortuosity and constrictivity. Therefore, we analyze 8119 microstructures generated by two different stochastic 3D microstructure models. This is to the best of our knowledge by far the largest set of microstructures that has ever been analyzed. Fitting artificial neural networks, random forests and classical equations, the prediction of effective conductivities based on geometric microstructure characteristics is possible.