Inverse design of a thermally self-adaptive carbon fiber acoustic metastructure using a residual MLP architecture

To address the dual challenges of low-frequency and broadband noise mitigation in variable-temperature aerospace environments, this study proposes an intelligent inverse design framework that integrates a deep residual multilayer perceptron (Res-MLP) surrogate model with a genetic algorithm (GA) optimizer. The resulting thermally self-adaptive carbon-fiber acoustic metastructure (TS-CFAM) employs embedded shape memory alloy (SMA) actuators to convert ambient thermal stimuli into controlled mechanical displacement, thereby enabling real-time reconfiguration of its resonant microarchitecture and overcoming the fixed-bandwidth limitation inherent to conventional passive acoustic metamaterials. Numerical and experimental results confirm that the Res-MLP surrogate model accurately captures the high-dimensional, strongly nonlinear acoustic–thermal–mechanical coupling, achieving a test-set coefficient of determination ( R 2 ) of 0.99. Validation further demonstrates that TS-CFAM delivers a continuous thermal frequency-tuning range of 1240 Hz—spanning 620 Hz to 1860 Hz—over the 30–50 °C operational range. Under weighted multi-objective optimization, the framework attains a 14.23% reduction in total structural mass and a 33.30% decrease in radial thickness, while sustaining an average sound absorption coefficient above 0.88 across the target band. Collectively, this work establishes a scalable, physics-informed digital design paradigm for structural–functional integration in next-generation aviation noise-mitigation systems.