Research On The Application Of Deep Self-encoding Network In Gearbox Faults

Artificial intelligence technology development has opened up the field of intelligent failure.Deep learning effective fit learning on fault data can avoid manual fault feature extraction.There are mainly supervised and unsupervised learning models in deep learning.Among these,supervised learning uses fault-labeled data for model training to fit fault category features,which often requires large amounts of labeled data.Unsupervised learning style automates the aggregation and classification of unlabeled data through the unsupervised model.A common type of neural network in the context of unsupervised learning is the Deep Autoencoder which is a model architecture with equal input and output sizes.Deep Autoencoder enables unsupervised learning of unlabeled data,and this unsupervised feature extraction can effectively handle the large amount of unlabeled data collected during production in mechanical fault diagnosis.In this paper,we focus on the application of Deep Autoencoder networks in gearbox fault diagnosis.The main elements of the study are as follows.(1)The large amount of data collected in industrial production,of which the largest proportion of unlabeled data,in order to make better use of this large amount of unlabeled data,using unsupervised learning to generate pseudo-labels data for a large amount of unlabeled data,by adding classifiers for fault classification,improve the accuracy of Deep Autoencoder diagnosis,and in the gearbox for the diagnosis of bearings and gears,respectively,multi-service diagnosis and diagnosis of bearing and gear mixed fault data.(2)Deep Autoencoder contains multiple network layers,where the activation function provides nonlinear data processing capabilities as a key link in the deep network learning layer’s processing of data.Using publicly available bearing and gear data as an object,Deep Autoencoder integration of multiple activation functions was used in the study to improve fault diagnosis stability.(3)In deep Autoencoder,convolutional autoencoder has efficient feature extraction capability.A fusion model of deformed convolutional and deep autoencoder is proposed for convolutional autoencoder,which enhances autoencoder feature extraction by deformed convolution and improves data reconstruction capability.(4)Through the deployment of multi-class Deep Autoencoder models using Tensor Flow-serving,Node-Red data real-time visualization and three.js 3D model rendering,providing a remote fault diagnosis basis for the actual production process.