Machine learning to optimize nonlinear conductive performance of composites for self‐adaptive electromagnetic shielding

Abstract Polymer‐based composites that exhibit a unique nonlinear response to high‐power electric fields have the potential to serve as intelligent electromagnetic shielding materials. The optimization of switching fields (E_b) and nonlinear coefficient (α) of polymer‐based composites is of great interests for nonlinear conductive performance. Based on literature data, the prediction models for E_b and α are first successfully established using machine learning (ML) methods. A stacking ensemble learning (SEL) strategy was used to combine five base machine learning models, showing superior predictive performance. The research focuses on the effect of key process parameters on nonlinear conducting composites. The feature importance analysis shows that the nonlinear properties of the composites are considerably impacted by the mass fraction, filler size, and sample thickness. The parameter optimization method to improve the performance of the composites was explored by using partial dependence plots analysis. By measuring the nonlinear response of CNT/ZnO composites under high electric fields, the effectiveness of the optimization strategy is experimentally verified. This work establishes the intrinsic relationship between composition and performance, which is helpful in designing intelligent self‐adaptive electromagnetic shielding for switchable electronic devices. Highlights Prediction models for E_b and α using machine learning. Stacking ensemble learning for superior predictive performance Focuses on the effect of key process parameters on performance. Optimization strategy validated through experimental testing.