Deep learning-assisted efficient design of biomimetic fluorescent dielectric elastomers

Dielectric elastomers (DEs), as a class of artificial muscle materials, possess broad application in actuators, biomimetic structures, and soft robotics. However, their practical implementation is hindered by high driving voltages requirements and limited functionality. Strategies such as tuning polymer network architectures and introducing fillers have been explored to enhance performance, but experimentally identifying optimal parameters is time-consuming and labor-intensive. Artificial intelligence offers new opportunities to address these challenges, with tools such as artificial neural networks significantly accelerating the design processes of materials. In this study, we employed deep learning and follow the core idea of active learning for model iteration to predict fluorescent DEs with potentially high actuation performance. A predictive model was trained using 10 sets of pre-experimental data and 117 sets of data from an database. Guided by the predictions, we selected appropriate materials and fabrication pathways to synthesize a high-performance DE with dual actuation–fluorescence functionality. The resulting DE achieved an in-plane actuation strain of 101.5% under an electric field of 6 kV/mm, significantly reducing the required driving field and voltage. A soft biomimetic insect fabricated from the optimized fluorescent DE demonstrated directional crawling, showing potential for applications such as task execution in environments unsuitable for humans.