Research On Rolling Bearing Fault Diagnosis Based On Deep Learning

Rolling bearing is one of the most commonly used parts in rotating machinery,and its working state directly affects the operation of the whole equipment.Therefore,the diagnosis of its failure is very meaningful,It has important practical significance how to use the intelligent technology to diagnose rolling bearing faults,with the advent of the big data era.Deep learning can simulate human brain's hierarchical data processing method,and extract low-level and specific raw data into high-level abstract features.Through deep structure,we can learn complex nonlinear relationships from training data,achieving the approximation of complex data distribution.It has powerful data processing ability,multi state recognition ability and generalization ability under variable conditions.Therefore,for traditional methods such as support vector machine,BP neural network and others are difficult to solve large data,rolling bearing failure in variable working conditions and multi state recognition of the degree of fault are very applicable.However,if the low level original feature data contain less or less characteristic information,the accuracy of state recognition would be greatly reduced.The set of empirical mode decomposition is an adaptive method of signal time-frequency localization.It can decompose any complex signal into several eigenmode functions.Each intrinsic mode function can be regarded as a single component signal,so every mode has its physical meaning.And each component contains some original feature information.In combination with above two methods,this paper presents a method for fault diagnosis of rolling bearing based on the combination of empirical mode decomposition and deep confidence network.This method first decomposes the original time domain signal by the set of empirical mode,then transforms the spectrum of the former eigenfunction of the decomposition,and connects the transformed components in turn,and constructs the high dimensional feature data containing the rich feature information.The high dimensional data are divided intoseveral data sets with the different load.As the input of deep learning,a load training other load test is used for fault diagnosis of rolling bearings.This method combines the deep learning of big powerful data processing ability and generalization ability of variable condition and state recognition ability,big data is solved by using the variable condition under the premise of rolling bearing fault characteristics of difficult state recognition problems.