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. Experiments show that neural network surrogates with physics-guided features have better accuracy than other ML models across different STL models. Furthermore, sensitivity analysis methods are used to assess how physically coherent the analyzed surrogates are.

Enhancing Road Holding and Vehicle Comfort for an Active Suspension System utilizing Model Predictive Control and Deep Learning

“…An Artificial Neural Network (ANN) is a powerful and adaptive approach that can be applied effectively to improve the performance of ASSs in vehicles by learning from real-time sensor data and intelligently adjusting suspension components. It effectively models complex relationships and improves ride comfort, stability, and safety, reducing body roll, pitch, and squat during maneuvers [20]. The multifaceted PSO approach has been applied on quarter-car thorax and pelvis models, demonstrating zero head acceleration, low sprung mass acceleration, and equivalent road-holding qualities.…”

Research on the Sound Insulation Performance of Composite Rubber Reinforced with Hollow Glass Microsphere Based on Acoustic Finite Element Simulation

“…Afterward, the influences of the parameters on the STL of the composite materials were investigated one by one, such as the density of the composite rubber and that of the HGM, the acoustic velocity in the polymer and that in the inorganic particle, the frequency of the incident wave, the thickness of the sound insulator, and the diameter, volume ratio and hollow ratio of the HGM. Later, based on the achieved effect behaviors of the influencing parameters, the weighted STL of the composite material for the limited thickness of the sound insulator was optimized through parameter optimization with the neural network algorithm [ 45 , 46 , 47 , 48 ], which aimed to obtain the optimal sound insulation effect with certain constraint conditions. The proposed sound insulation material of composite rubber reinforced with HGM could be considered as a highly efficient sound insulator with little occupied space, which could be favorable for promoting its practical application in the industrial field.…”

A Parametric Physics-Informed Deep Learning Method for Probabilistic Design of Thermal Protection Systems

“…However, this approach necessitates calling the system model multiple times for computation, resulting in significant computational costs and reduced computational efficiency. Alternatively, the original high-precision complicated model can be substituted with a surrogate model to enhance computational efficiency while preserving accuracy [6][7][8][9][10]. A surrogate model refers to an approximation model that describes the relationship between the input and output of a system.…”