Design of labyrinthine ventilated sound absorption structures driven by machine learning

Abstract Ventilated sound absorbers have garnered considerable attention due to their ability to allow airflow while effectively reducing low-frequency noise. However, their efficient design remains highly challenging due to the complex and nonlinear relationship between geometry and acoustic performance. The present study proposes a novel deep learning (DL) based method for optimizing the parameters of ventilated sound absorbers. The method employs a convolutional neural network in conjunction with a unique functional design network to facilitate inverse design, transforming target acoustic absorption performance into corresponding structural parameters. To effectively optimize structural parameters, an improved fitness function-based particle swarm optimization algorithm is introduced to guide the training of the neural network. The proposed algorithm accounts for inter-structure frequency coupling by optimizing the geometric parameters of labyrinthine absorbers, thereby redistributing their frequency-domain distributions to reduce detrimental impedance interactions. Notably, the trained surrogate model demonstrates high prediction accuracy, with the mean squared error ( MSE ) of the predicted absorption coefficient remaining within 3 × 10 −5 across the investigated frequency range. The optimized structure achieves near-perfect broadband low-frequency absorption, with absorption coefficients exceeding 0.85 within the target frequency band, and effectiveness of the proposed framework is validated through both simulations and experimental measurements. Results demonstrate that the DL based optimization approach not only enhances model performance but also improves impedance matching in multi-absorber systems. This improvement stabilizes their acoustic response and broadens the effective absorption bandwidth across the target frequency range.