Machine learning drives fast and accurate calibration of carbon nanotube contact parameters

The contact parameters of carbon nanotubes (CNTs) particles play a crucial role in measuring mobility, predicting structure, optimizing materials, and many other aspects. However, the complexity and severe challenges posed by the micro- and nano-scale sizes, along with the inhomogeneity of the particles, significantly impede experimental investigations, making it difficult to acquire precise contact parameters. To date, CNTs have not been found as a contact model for agglomerated particles. In this paper, a new method of contact parameter refinement is proposed for determining the contact parameters of CNTs agglomerated particles during fluidization. Initially, the angle of repose (AOR) and density of CNTs agglomerated particles are determined based on physical tests, leading to the identification of recommended ranges for six contact parameters. Subsequently, the three contact parameters with the most significant effects are screened based on Plackett-Burman. The Response Surface Methodology, Sympy, and Fully Connected Neural Network (FCNN) are then employed as prediction models for comparison against numerical simulations for validation, resulting in a reduction of the FCNN model's relative error from 14.81% to 2.09%. Finally, the optimal multi-objective parameter combination coefficients (0.45/0.45/0.121/0.76/0.74/0.13) are determined by optimizing the inversion based on the selected best FCNN model under the experimentally measured AOR of 40.3°. Numerical simulations based on this parameter and comparative experiments with an error of only 0.07° fully demonstrate the effectiveness of the proposed strategy. This study provides the possibility to accurately simulate the fluidization pattern of CNTs and further investigate their growth mechanism.