Research On Early Warning Methods Of Bearing Failure Based On Neural Network

As the essential component supporting and rotates in standard mechanical equipment,rolling bearing is highly vulnerable to damage during the conducting process.When applied on-site,it is hard to find the cause of the accident in time,and the failure will lead to a severe impact on the device’s safety,thus causing severe production safety hazards.This thesis takes bearing data from Case Western Reserve University(CWRU)as the research object.An optimal operation strategy is proposed after combining theoretical knowledge with production practice and conducting indepth research about signal decomposition and neural network prediction technology.A failure early warning system for rolling bearings is realized.Firstly,regarding the difficulty of characterizing the vibration signal fault,the internal structure of the rolling bearings is deeply analyzed,and the reasons for the failure are listed.A new signal processing method-Hilbert Yellow Transform(HHT)-is adopted to decompose,reduce,and transform the collected vibration signal of the drive end to get the Hilbert spectral envelope containing the main fault feature.With the extracted feature vector set,the accurate extraction of the fault feature information is realized.The neural network algorithm is applied better to improve the performance of the fault prediction model.Three typical neural network algorithms are studied based on the drive end’s vibration signal,and a fault prediction model is proposed.The normalized feature vector set is artificially shuffled and used as the input to the system.The comparative prediction experiment shows that the Elman neural network has the best accuracy and has a phenomenon that signals in the input layer are lost quickly.In response to this defect,this thesis proposes an improved Elman neural network fault prediction model.The enhanced model introduces time delay in the input layer,selects the appropriate time delay according to the actual vector dimension,and corrects the data input entry of the input layer.In experiments under the same conditions and different conditions,it is found that the improved Elman neural network fault prediction model conducts the best effect.The accuracy of the prediction fluctuates between 80% and 90%,the time for the forecast is short,and the antiinterference is potent,stable,and reliable.A flexible and intelligent fault warning platform is designed and implemented in this article relying on the experimental data and methods.The system is convenient for users and realizes the fault warning,monitoring,recording,analysis,and storage functions of rolling bearings,which effectively improves the safety of the equipment and helps the staff find the faults early.The actual application shows that the approach has high engineering application value.