Abnormal Monitoring Method of Voiceprint Signal of Rolling Bearing Based on CAE-ECA Model
In order to solve the problem that traditional voiceprint analysis methods do not express enough features man-ually and rely too much on expert knowledge,a model based on Convolutional Auto-Encode network and Effective Channel Attention mechanism is proposed to realize adaptive voiceprint feature extraction and efficient recognition of abnormal states of rolling bearings.First of all,the one-dimensional time sequence voiceprint signal is sampled repeatedly at intervals to con-struct input samples,and the deep features of bearing running sound signals are adaptively extracted by using neural net-work structures such as convolution and pooling.Then,the weight of the convolutional self-coding network to the key features is enhanced by the effective channel attention module.The self-coding network structure is used to reconstruct the sound signal and only the normal running sound signal data of rolling bearings are used for model training.Finally,the abnormal state of rolling bearing is evaluated by the reconstruction deviation of the abnormal state data of the model.
rolling bearingvoiceprint anomaly monitoringconvolution self-codingeffective channel attention mechanism
NETL
NSTL
万方数据
