Optimization design of energy storage performance in ferroelectric nanocomposites based on phase field simulation and machine learning

Ferroelectric nanocomposites are promising candidates for dielectric energy storage devices due to their rapid charge–discharge capability and high breakdown strength. However, their energy storage performance is highly dependent on the composition, morphology, and volume fraction of nanofillers. Traditionally, identifying suitable nanofillers has relied on either trial-and-error experimentation or time-consuming computation. Here, by combining the phase-field simulations and machine learning, we propose an approach to rapidly identify optimal nanofillers in PbZr1-xTixO3/SrTiO3 (PZT/STO) nanocomposites to enhance energy storage performance. Phase-field simulations are conducted on the energy storage density and efficiency of the ferroelectric nanocomposites with different compositions, morphologies and volume fractions of PZT nanofillers to construct a dataset, which is employed to train a neural network model. The trained model predicts that the maximum energy storage density is achieved with a nanofiller composition of x = 0.55, a volume fraction of 40%, and an aspect ratio of 7, whereas optimal energy storage efficiency occurs at x = 0.55, 25% volume fraction, and an aspect ratio of 0.636. This work establishes an effective machine learning-based strategy for optimizing the energy storage performance of ferroelectric nanocomposites by efficiently searching the nanofiller design space.