Research On Machine Abnormal Sound Detection Based On Deep Learning

With the development of modern science and technology,mechanical equipment is widely used in industrial production,and the failure rate increases gradually with the increase of service life.The health monitoring and anomaly detection of industrial machines are of great significance for production safety.Machine learning is the main method to extract abnormal sound data and collect abnormal sound data.This method has two main shortcomings: one is that some abnormal sound data are difficult to simulate,manufacture and collect;the other is that the model has limited learning of data features and low prediction accuracy.In view of the above problems,the original machine abnormal sound detection method is improved by deep learning,which can realize the rapid detection of mechanical equipment and further improve the accuracy.The specific work is as follows:(1)In view of the lack of abnormal sound data in real life,a machine abnormal sound detection model based on supervised learning based on mobilenetv3 is proposed.Using the collected machine normal operation sound,the experimental data set is divided according to the machine number in the data set,and the model is trained to realize the machine abnormal sound detection,Through this model,the AUC(Area Under roc Curve)value of the baseline system based on self encoder network is increased by 9.72%,and the performance and accuracy of the model are greatly improved.(2)In order to improve the effectiveness of data feature learning,a machine abnormal sound detection method based on sound feature extractor and auto encoder(MMNSFE-AE,Machine Mobile Net V3 Sound Feature Extractor-Auto Encoder)is proposed.It is improved on the basis of Mobile Net V3 model,which is used as sound feature extractor(MMNSFE),and the extracted data is input into the auto encoder(AE)model,Realize the machine abnormal sound detection task.The experimental results show that the AUC value of this synthesis method is 14.6% higher than that of the baseline system based on auto encoder network,which verifies the effectiveness of sound feature extractor and the superiority of this method.