Prediction of the acoustic form function by neural network techniques for immersed tubes

Area of Science:

• Acoustics
• Computational Mechanics
• Machine Learning

Background:

• Predicting acoustic form functions (FF) for cylindrical shells is crucial for structural acoustics analysis.
• Traditional analytical methods can be computationally intensive and complex for certain geometries.

Purpose of the Study:

• To develop and evaluate artificial neural network (ANN) techniques for predicting the acoustic form function (FF) of infinite cylindrical shells.
• To compare ANN predictions with analytical solutions using the Wigner-Ville distribution.

Main Methods:

• An artificial neural network (ANN) model was designed and trained to predict the acoustic form function (FF).
• The Wigner-Ville distribution was employed to validate ANN predictions against analytical calculations for a stainless steel tube.
• Various network configurations were tested, focusing on different radius ratios (b/a).

Main Results:

• The optimal ANN model featured a single hidden layer.
• The developed ANN achieved a mean relative error of approximately 1.61% in predicting the acoustic form function.
• ANN predictions showed good agreement with analytical results.

Conclusions:

• Artificial neural network techniques provide an efficient and accurate method for predicting acoustic form functions in cylindrical shells.
• The study demonstrates the potential of ANNs as a powerful tool in computational acoustics and structural analysis.