In this work, the classical problem of acoustic radiation from a baffled plate under harmonic excitation is considered. The objective of the present work is to determine the location of a suitable lumped mass on the plate such that the radiated sound power is minimal. Towards this end, the elementary radiator approach is used to compute the radiated sound power from the vibrating structure. The vibrating structure comprises of a baffled simply-supported plate with or without lumped masses. For wider applicability of the present work, all the parameters associated with the problem are non-dimensionalized. A numerical optimization technique is employed in order to determine the location of the masses for minimal sound power radiation. The optimization exercise is implemented in MATLAB using classical algorithms.
Additionally, finite element method based simulation is conducted in COMSOL Multiphysics to validate various vibration and acoustic results.
As there are several parameters involved in the problem, the numerical optimization exercise is computationally intensive. As such, re-runs for any change in design parameters are prohibitively expensive. In order to mitigate this problem, different machine-learning models (Artificial neural network and K-nearest neighbors) are developed for computing the optimal location of the lumped masses for minimal sound power radiation. Data generated using the elementary radiator approach is used for training the machine-learning models. In order to solve the different types of loading conditions, a pre-trained ANN architecture is used with transfer learning. Similarly, a KNN model is employed for different loading conditions associated with the problem. The results obtained from the ML models are verified against the solution obtained using the classical optimization technique. The present results show the applicability of ML models in obtaining a quick and accurate solution for the present problem. A comparison between different machine learning models visa-a-vis the traditional approaches is presented.