A multi-scale neural network accelerated complex permittivity homogenization framework for frequency-dependent electrical responses in MXene-based nanocomposites with physics-based interface effects

Physics-based interface effects serve as the key mechanisms for illustrating the excellent frequency-dependent electrical behaviors in MXene/polymer nanocomposites. In this research, we propose a novel multi-scale complex computational homogenization framework for AC electrical behaviors in MXene-based nanocomposites on realistic configurations. First, a direct complex FE² (cFE²) model is developed to evaluate the frequency-dependent complex permittivity of MXene-based nanocomposites. Several categories of physics-based functional interface effects are considered in the frequency domain, including the tunneling effect, frequency-dependent Maxwell-Wagner-Sillars polarization, and electron hopping. To address the high computational cost of the cFE² model, a complex artificial neural network (cANN) surrogate model is developed to rapidly evaluate the nonlinear electric responses of microscopic RVE. A neural network-based homogenization model (cANN-cFE) of complex permittivity is established to accelerate the multi-scale numerical simulation for MXene-based nanocomposites. The predicted AC electric responses by the direct cFE² and accelerated cANN-cFE models are both calibrated with the experiment. Compared with the direct cFE² model, the cANN-cFE model further reduces the computational cost by several orders of magnitude while preserving high computational accuracy. The effective permittivity decreases with the frequency, while the effective conductivity reveals an opposite trend. This research can provide the design aid of MXene-based nanocomposite device.