Inverse engineering of micro-perforated plates for targeted acoustic characteristics

2026-04-27T03:53:26+00:00

The inverse design of micro-perforated panels (MPPs) for target sound absorption remains challenging due to the complex nonlinear relationship between structural parameters and acoustic performance. This study proposes a tandem neural network (TNN) framework to achieve efficient inverse design of single-layer MPPs. A forward multi-layer perceptron (MLP) is first trained to accurately predict the absorption coefficient curve from three key structural parameters: perforation diameter, panel thickness, and cavity depth. The forward model achieves superior accuracy compared to GA-SVR, GridSearch-SVR, and random forest models, with an R² of 0.999 and MAE of 0.007. Subsequently, an inverse design network is connected in series with the frozen forward network, taking a target absorption curve as input and outputting the corresponding structural parameters. The activation function of the output layer constrains the parameters within physically feasible ranges. The framework is validated by designing an MPP with a distinct absorption peak in the 300–600 Hz range. The predicted parameters (diameter 0.93 mm, thickness 0.9 mm, cavity depth 9.9 mm) yield an absorption curve that matches the target with an R² of 0.997. This work demonstrates that deep learning can effectively automate the inverse design of MPPs, offering a flexible and efficient alternative to traditional trial-and-error methods.

Fuzzy-stochastic coupled models for broadband noise radiation from flexible

2026-04-30T07:40:28+00:00

Broadband noise radiation from flexible panels is governed by the coupling of random excitation fields with uncertain structural and boundary parameters. This paper develops a fuzzy-stochastic vibro-acoustic framework that separates (i) aleatory uncertainty in the broadband pressure field from (ii) epistemic uncertainty in panel properties and mount conditions. The panel dynamics are modeled in the frequency domain using a modal or finite-element representation of the thin-plate operator, while acoustic radiation is evaluated through a baffle-mounted radiation model leading to radiated sound power spectra. Random excitation is represented by the pressure cross spectral density, enabling direct propagation of spectral statistics to displacement and velocity cross-spectra via linear transfer functions. Epistemic uncertainty in Young's modulus, thickness, and loss factor is represented by fuzzy numbers and propagated through α-cuts, yielding interval-valued parameter sets at each α level. The coupling is implemented using an α-cut outer loop with a stochastic inner solver that computes mean and variance of radiated sound power; α-level interval extrema then provide fuzzy envelopes of stochastic response metrics. Verification is performed through modal truncation, frequency-grid stability, and α-grid refinement. Numerical demonstrations using representative datasets show that epistemic uncertainty can induce wide bands in band-integrated sound power level (≈9.9 dB in the 100–1,000 Hz band at α = 0), and that percentile metrics (e.g., 95th percentile under a lognormal approximation) provide conservative bounds for design decision-making. The proposed framework offers a transparent and computationally tractable route to uncertainty-aware broadband vibro-acoustic prediction for panels in vehicles, buildings, and machinery enclosures.

Sound & Vibration

Inverse engineering of micro-perforated plates for targeted acoustic characteristics

Fuzzy-stochastic coupled models for broadband noise radiation from flexible