Prediction of compressor aerodynamic noise based on transfer learning Random Forest

Area of Science:

• Mechanical Engineering
• Acoustics
• Computational Fluid Dynamics

Background:

• Aerodynamic noise is critical for high-pressure ratio centrifugal compressors.
• Traditional prediction methods are computationally expensive and time-consuming.
• Rapid assessment of compressor noise is needed.

Purpose of the Study:

• To develop a rapid and accurate method for predicting aerodynamic noise in centrifugal compressors.
• To apply transfer learning for noise prediction using limited target data.
• To validate the effectiveness of the transfer learning approach.

Main Methods:

• Experimental acquisition of aerodynamic noise data from baseline and target compressors.
• Transferability evaluation using the maximum mean discrepancy method.
• Transfer learning: pre-training on baseline data, fine-tuning on target compressor data.

Main Results:

• The transfer learning model accurately predicts aerodynamic noise for target compressors.
• Overall sound pressure level error is less than 3 dB.
• Achieved high prediction accuracy with minimal training data.

Conclusions:

• Transfer learning offers an effective and efficient solution for aerodynamic noise prediction in centrifugal compressors.
• This method significantly reduces computational cost compared to traditional approaches.
• Enables faster design iterations and performance assessments.