The morphological features in silica aerogels can be modelled by means of the diffusion-limited cluster-cluster aggregation (DLCA) algorithm [1]. However, the DLCA modelling process is slow and computationally expensive for potential reverse engineering applications. In this poster, we demonstrate the capability of a recently developed artificial neural network (ANN) for predicting the fractal properties of silica aerogels [2]. The input variables to the ANN are the DLCA model parameters, while the output of the ANN is the fractal dimension. The same ANN is then inverted to predict the input variables for a given target fractal dimension. The problem of non-uniqueness is overcome by using a guided gradient descent approach in the input space. The forward mapping of the fractal dimension is very accurate with an R² score of 0.973, while the constrained inversion results in predicting DLCA parameters within an error of 2%. Possibilities of extending such an approach for modelling the properties of fibrillar aerogels is also highlighted at the end. 

References

[1] R. Abdusalamov, C. Scherdel, M. Itskov, B. Milow, G. Reichenauer, and A. Rege, J. Phys. Chem. B 125: 1944-1950 (2021)

[2] R. Abdusalamov, P. Pandit, B. Milow, M. Itskov, and A. Rege, Soft Matter (2021) doi: 10.1039/D1SM00307K