US 12,442,684 B1
Method and apparatus for acquiring aerodynamic noise of compressor, medium, and product
Chen Liu, Harbin (CN); Xu Zhan, Harbin (CN); Zexi Wu, Harbin (CN); Yipeng Cao, Harbin (CN); Yang Liu, Harbin (CN); Jie Yang, Harbin (CN); Jie Guo, Harbin (CN); and Xinyu Zhang, Harbin (CN)
Assigned to Harbin Engineering University, Harbin (CN)
Filed by Harbin Engineering University, Harbin (CN)
Filed on Jan. 17, 2025, as Appl. No. 19/028,180.
Claims priority of application No. 202410431700.X (CN), filed on Apr. 11, 2024.
Int. Cl. G01H 17/00 (2006.01)
CPC G01H 17/00 (2013.01) 8 Claims
OG exemplary drawing
 
1. A method for acquiring aerodynamic noise of a compressor, comprising:
acquiring first aerodynamic noise frequency-domain data of a compressor in a historical time period under a target working condition continuously, wherein the first aerodynamic noise frequency-domain data comprises corresponding sound pressure levels of aerodynamic noise at different frequencies;
performing inverse Fourier transform on the first aerodynamic noise frequency-domain data to obtain corresponding first aerodynamic noise time-domain data, wherein the first aerodynamic noise time-domain data comprises a sound pressure corresponding to each historical time step;
determining an actual acquisition time of the first aerodynamic noise frequency-domain data according to all the historical time steps;
determining a predicted aerodynamic noise time according to a target acquisition time and the actual acquisition time, wherein the predicted aerodynamic noise time comprises each future time step;
with each future time step as an input, outputting a sound pressure corresponding to each future time step by a trained nonlinear autoregressive neural network model trained on the first aerodynamic noise frequency-domain data, wherein a training process of the nonlinear autoregressive neural network model comprises:
selecting a certain proportion of the first aerodynamic noise time-domain data as training samples;
training the nonlinear autoregressive neural network model with each historical time step in the training samples as an input and a predicted sound pressure as a label;
optimizing the nonlinear autoregressive neural network model using Bayesian regularization algorithm;
calculating a loss error between the predicted sound pressure output by the optimized nonlinear autoregressive neural network model and a corresponding actual sound pressure using a loss function; and
when the loss error is less than a threshold, completing the training of the nonlinear autoregressive neural network model;
obtaining second aerodynamic noise time-domain data corresponding to the target acquisition time according to the sound pressure corresponding to each historical time step and the sound pressure corresponding to each future time step;
performing Fourier transform on the second aerodynamic noise time-domain data to obtain second aerodynamic noise frequency-domain data, wherein a frequency range corresponding to the second aerodynamic noise frequency-domain data is the same as a frequency range corresponding to the first aerodynamic noise frequency-domain data;
identifying a noise source based on the second aerodynamic noise frequency-domain data to determine a component of the compressor generating the aerodynamic noise; and
controlling the component of the compressor generating the aerodynamic noise to reduce impact on surrounding environment.