In the field of acoustic diagnostic fault audio monitoring of mechanical equipment, the fault audio sample data plays a crucial role in enhancing the operational safety of mechanical equipment. To this end, this paper designs a mechanical fault audio acquisition framework based on a high-resolution A/D digital-to-analog converter, and performs noise reduction on the acquired acoustic audio signals by improving the variable-step-length LMS adaptive filtering algorithm. Then the compression-aware algorithm is utilized to realize the sampling data compression, transmission and recovery of the mechanical fault audio information, and the sampling accuracy of the mechanical fault audio information is improved by the additive random sampling algorithm, which further reduces the power consumption of the sampling of the acoustic audio information of mechanical faults. The transformer equipment fault audio information of a power grid is selected as the research object, the BIOES annotation specification is adopted to automatically annotate the mechanical fault acoustic automatic audio information, and the mechanical fault audio information knowledge base is established. The BiLSTM+CRF and Transformer models are utilized for named entity recognition and entity relationship extraction of mechanical fault diagnosis audio information knowledge respectively. The results show that the audio information acquisition method designed in this paper for acoustic diagnosis of mechanical faults has higher efficiency and lower power consumption, and the F1 values of entity recognition and entity relationship extraction for BiLSTM+CRF and Transformer models reach 95.87% and 86.31%, respectively.
