Abstract
This study explores the application of machine learning techniques for identifying acoustic modes within turbomachinery assemblies. A multilayer perceptron neural network is employed with supervision from a forward model for modal identification generating extensive training data. The goal is to extract hidden modal features from the transfer function of the forward model. This paper offers guidelines for creating such networks and investigates their robustness under noise interference. Results show that simple network structures consisting of just a single hidden layer can already accurately predict azimuthal modes. Critical factors include the number of neurons and activation functions for effectively solving this problem. The proposed method is substantiated using experimental data from a duct acoustic testing rig. Furthermore, it is shown that the method can be combined with compressive sensing (CS) to reduce sampling requirements in acoustic signal processing and duct acoustic applications.
Funding Information
  • Suzhou Gusu Innovation and Entrepreneurship Leading Talents Program (ZXL2023163)
  • National Natural Science Foundation of China (12202364)
  • Aeronautical Science Foundation of China (2024Z073053007)
  • Fundamental Research Funds for the Central Universities (G2022WD01008)

This publication has 24 references indexed in Scilit: