Statistical modeling of collision frequencies of non-spherical agglomerates

点击翻译按钮获取中文摘要

Agglomeration is one of the key mechanisms for particle and aerosol growth. This growth process can easily span several orders of magnitude from the nanoscale to the microscale under process conditions that are typical for particle flame synthesis. Particle collision leads to particle growth and – depending on material properties and temperature – to agglomerate with clearly varying morphology. The mechanisms, however, that drive relative particle motion and collision depend on the agglomerates’ size. While collision mechanisms for very small and very large, spherical particles are well understood, there is no thoroughly validated model for particle collision of non-spherical particles in the transition region. In the present work, new models for collision frequencies of non-spherical agglomerates at sizes typical for particle flame synthesis will therefore be analysed and validated.Numerical simulations suitable for validation need to allow for agglomerate growth across several orders of magnitude, they need to account for the correct ratio of all relevant length scales such as primary particle diameter and turbulent length scales, and they need allow for effects of fractal agglomerate morphology on growth dynamics. Conventional simplifications that are based on scale separation between the small and the large scales are no longer valid due to the significant growth of the agglomerates and a wide range of scales needs to be resolved. The distinction of the project is the use of a coarse-graining approach that allows to represent the smallest scales by larger (coarser) scales and to significantly reduce computational requirements. Only by coarse-graining, the entire range of length scales can be computed.Firstly, the coarse-graining approach will be further developed to allow for a direct tracking of all agglomerates while resolving all relevant turbulent scales. Only then, an explicit analysis of collision frequencies between agglomerates of different sizes can be conducted and direct evaluation of the accuracy of existing models is possible. Secondly, the large data sets of these direct simulations will be used to automatize the additional characteristics of the coarse-grained particles that allow for the reduction in the degrees of freedom and to generate alternative models for the collision frequencies using deep learning methods.