Online updating system and method for abnormal sound detection

Abstract

The invention provides an online updating system for abnormal sound detection. The online updating system comprises an embedded end system and a server end system. The embedded end system collects working audios of a to-be-tested sound source and inputs the working audios to the neural network module for offline abnormal sound classification. And when the offline abnormal sound classification result is that the abnormal sound is not identified, an unknown abnormal reporting module reports the unknown abnormal audio. A server-side system classifies unknown abnormal audios according to equipmenttypes to complete data cleaning and clustering, regulate the network structure of the neural network module and generate a new training set and a verification set, the abnormal sound detection modeltraining is performed on equipment of the model, the training information is issued to the embedded end system without abnormal identification after training is completed to update the abnormal sounddetection system. The present invention further comprises an online updating method for abnormal sound detection. According to the invention, dynamic updating of the abnormal sound detection model isrealized, and the system and the method can adapt to a changing abnormal sound diagnosis working environment.

Description

Technical field [0001] The invention relates to the field of embedded devices, in particular to an online update system and method for abnormal sound detection. Background technique [0002] Sound is a convenient, effective and quick way to convey information. At present, the abnormal detection of vehicle work, the failure detection of compressors, motors and other mechanical equipment, the detection of abnormal sounds in the room, and the detection of children's crying are mainly based on human judgment and relying too much on human subjective experience, which leads to large errors in positioning these abnormalities. And the cost of consumption is high. [0003] In recent years, some abnormal sound detection methods based on deep learning have appeared, which can show good practical application effects, but there are also some shortcomings: [0004] 1. The system is complex and computationally intensive, and relies on complex computing units and even GPU (Graphic Process Unit, vi...

AI Technical Summary

Problems solved by technology

[0008] When real-time audio is transmitted over the network, frame drops are extremely prone to occur. When frame drops occur in the audio stream, its spectral characteristics may change accordingly, causing abnormal sound detection to fail.

[0009] Deploying the detection system to an offline environment also has problems: the structure of the large-scale machinery and machinery waiting for the sound source to be measured is complex, and there are many types of abnormalities that may occur. When using deep learning for abnormal sound detection, when there are unknown abnormal sounds that the detection system has not learned , the abnormal sound cannot be located

Therefore, it is not possible to update the abnormal types very well, resulting in poor recognition of abnormal sounds

Images