Cascading deep neural networks for inverse design of acoustic-mechanical multifunctional metastructures

Multifunctional, lightweight, and compact structures have become increasingly essential in modern applications, as exemplified by airplanes. This paper presents a methodology for the concurrent inverse design of acoustic-mechanical properties in curved-crease origami foldcore sandwich structures, utilizing deep neural networks. Initially, a forward neural network was developed to predict sound absorption coefficients, buckling strength, and failure strength based on geometric parameters. Subsequently, an inverse neural network was constructed and cascaded with the pretrained forward neural network to achieve inverse design, mapping functional requirements to geometric characteristics. Through two practical engineering cases and experimental validations, one of the inverse-designed structures with a thickness of 21.4 mm exhibits excellent sound absorption (α > 0.85) at the target frequencies of 720 Hz and 900 Hz. Another inverse-designed structure, with a thickness of 22.4 mm, achieves 75% sound absorption within the target frequency band (600–1000 Hz). Furthermore, the inverse-designed structures exhibit exceptional mechanical performance, with failure strength exceeding 60 MPa and specific strength reaching up to 32.3 MPa/g. This work accomplishes the inverse design of a metastructure characterized by lightweight, ultra-thin, and functionality-customizable attributes, highlighting its potential applications in aerospace, transportation, and architectural acoustics engineering.