A vibration data feature extraction method based on a sliding window self-attention mechanism

Abstract

The application belongs to the field of artificial intelligence health monitoring, and relates to a vibration data feature extraction method based on a sliding window self-attention mechanism, comprising the following steps: step one, collecting multi-channel original vibration signals on a complex device and performing signal data preprocessing; step two, based on the sliding window self-attention mechanism, performing feature calculation and integration on the preprocessed data to extract key feature data; step three, according to the preprocessing mode of step one and the sliding window self-attention mechanism based on step two, referring to a residual neural network, constructing a deep artificial neural network, designing a classification head combined with a specific task, analyzing the key feature data and outputting specific analysis results. The application can directly process original vibration data with different sampling frequencies, different collection time lengths and different channel numbers, is convenient for general deployment in different equipment, different collection sensors and different data processing algorithms, has small influence on data distribution and modal, and has strong self-adaptability.

Images