Rapid prediction of effective absorption bandwidth in PEEK/CF additive manufacturing metastructure via interpretable machine learning

Abstract

The development of machine learning has provided a new perspective for the design of electromagnetic metastructures, particularly in the rapid design of key performance metrics such as effective absorption bandwidth. Traditional methods, grounded in electromagnetic theory and empirical approaches, often lacked sufficient flexibility and adaptability. In this work, three types of machine learning models were developed to establish the relationship between effective absorption bandwidth and structural parameters. The results indicated that the random forest model achieved the most accurate and efficient design for this task. Then, the additive manufacturing optimal metastructure obtained using this approach outperformed existing designs in terms of both effective absorption bandwidth and reflectivity, while also exhibiting superior radar stealth performance and mechanical load-bearing capacity. Furthermore, through interpretable machine learning and data analysis, the intrinsic mechanisms underlying the relationship between effective absorption bandwidth and structural parameters were revealed. Overall, this work introduced a novel approach to metastructure design and enhanced the understanding of the relationship between structural parameters and electromagnetic properties, providing a key foundation for future design.