Research on Near Acoustic Field Noise Prediction Methods for Scroll Compressors Based on Vibration Signals Using Transformer and BiLSTM

This study addresses the challenge of measuring near acoustic field noise on the surface of scroll compressor casings by proposing a deep learning approach for noise prediction. We employ a hybrid network model that integrates a Transformer and a Bi-directional Long Short-Term Memory (BiLSTM) network. Initially, vibration acceleration data from the compressor casing surface and near acoustic field noise pressure data are collected in an anechoic chamber and preprocessed. Subsequently, the Transformer network is utilized to identify and extract features from the vibration acceleration and noise pressure data. Finally, the BiLSTM network, coupled with an attention mechanism, is employed to train and refine the model's performance on the basis of the input vibration acceleration data features. The attention mechanism enhances the model's ability to capture critical information, thereby accurately predicting the near acoustic field noise pressure at the compressor casing surface. This method effectively resolves the issue of inaccurate noise measurements for scroll compressors in situ.