On Machine Learning-Driven Surrogates for Sound Transmission Loss Simulations

Abstract

Surrogate models are data-based approximations of computationally expensive simulations that enable efficient exploration of the model's design space and informed decision-making in many physical domains. The usage of surrogate models in the vibroacoustic domain, however, is challenging due to the non-smooth, complex behavior of wave phenomena. This paper investigates four Machine Learning (ML) approaches in the modelling of surrogates of Sound Transmission Loss (STL). Feature importance and feature engineering are used to improve the models' accuracy while increasing their interpretability and physical consistency. The transfer of the proposed techniques to other problems in the vibroacoustic domain and possible limitations of the models are discussed.

A review of machine learning methods applied to structural dynamics and vibroacoustic

“…When selecting a suitable algorithm for surrogate modeling, several factors need to be considered, such as the size, input format, and dimensionality of the dataset, the smoothness and nonlinearity of the function, and the need for prediction variance, according to the guidelines in Section 2. Statistical methods, such as PCE [238,239], polynomial response surface model (RSM) [240][241][242][243], RBF interpolation [244,245], lowrank tensor approximations [235], and spectral expansions, [231] are largely used to construct surrogates in SD&V. ML supervised regressors, such as SVM [246], GPR [247,248], NN [234], RF, and gradient-boosting decision trees [249], are also commonly employed due to their capabilities to approximate arbitrary functions, as they pose weak assumptions on the format of the underlying function.…”

“…The challenge of modeling surrogates for SD&V problems with non-smooth behavior was addressed in [249]. A benchmark of ML algorithms was conducted for the sound transmission loss problem of fluidstructure interaction, with complexities ranging from analytical to FEM models.…”

“…The mean decrease in impurity [366], a by-product of the RF training, was also used to perform GSA and improve surrogate interpretability in the case of sound transmission analyses [249]. An alternative method to evaluate feature importance with RF is the out-of-bag- based sensitivity.…”