Research On Abnormal Sound Detection Technology Of Compressor Based On Depth Learning

Natural gas compressor is a key equipment in the field of gas gathering station in oil and gas field.It has been widely used in the field of pressurized input natural gas network and so on.Once abnormal,it will seriously affect the normal operation of production.Robot inspection technology can collect the sound of natural gas compressor in oil gas gathering station,and realize the real-time monitoring of natural gas compressor status in operation site.Taking the sound of natural gas compressor in the gas gathering station of oil and gas field as the research object,this paper determines the abnormal sound detection scheme of natural gas compressor based on the combination of feature extraction technology and deep learning detection technology,compressor sound feature extraction based on improved MFCC and VGG16 fusion compressor sound recognition algorithm based on improved Transformer model and compressor abnormal sound detection algorithm based on Inception-DCGAN are proposed.It has certain guiding significance for abnormal sound detection of natural gas compressor in gas gathering station of oil and gas field.The main research contents are as follows:(1)Aiming at the problem of high similarity between abnormal sound classes of natural gas compressor collected by gas gathering station,a compressor sound feature extraction method based on improved MFCC and VGG16 fusion is proposed.By extracting low-level features(LLDs)and calculating the feature values of each frame for abnormal sounds in natural gas compressors,a feature level fusion module is introduced for information compression,achieving better differentiation of abnormal sounds in natural gas compressors where the feature differences are not obvious.Simulation experiments have shown that this method effectively solves the problem of high similarity of abnormal sounds in natural gas compressors.(2)A compressor sound recognition algorithm based on improved Transformer is proposed to solve the problem that the dense arrangement of equipment in gas gathering station will lead to the interference noise collected from other equipment and affect the detection accuracy.Extract the time series characteristics of the natural gas compressor sound in the oil-gas field gas gathering station,introduce the spatial attention mechanism,focus on the decisive area in the data characteristics,combine the transfer learning method,improve the detection efficiency,and build a recognition model for the abnormal sound of the natural gas compressor.Simulation experiments have shown that this method improves the recognition accuracy of abnormal sounds in natural gas compressors under interference from other devices.(3)Aiming at the problem that there are few abnormal sound data of natural gas compressor collected by gas gathering station,an abnormal sound detection algorithm of compressor based on Inception-DCGAN is proposed.Extract the sound spectrogram features of natural gas compressors to achieve multi-dimensional and multi-level sound feature representation.DCGAN combines Inception network to efficiently expand the network from two different angles of depth and height,and uses SSIM values to detect anomalies in unknown categories of sound.Simulation experiments show that this method achieves abnormal sound detection of natural gas compressors with limited abnormal data.