Research On Bearing Compound Fault Diagnosis Technology Based On Multi-source Information Fusion

With the rapid development of big data,Internet,sensors and artificial intelligence technology,the upgrade of traditional machinery manufacturing industry by using the artificial intelligence technology is an inevitable trend.As one of the most common mechanical parts,bearings play an important role in mechanical equipment.And with the rapid development of sensor technology,data processing technology and intelligent diagnosis technology,intelligent diagnosis of bearings' health has also become a hot area in academic research.In actual production,an initial fault that doesn't affect normal production or doesn't reach the replacement standard may make a combined fault(compound fault),and the complex operating environment of the equipment often leads to the occurrence of compound faults.Therefore,there are still great difficulties in bearing compound fault detection and diagnosis.Because of these difficulties,the research on the bearing compound fault detection and diagnosis has become a hot area in domestic and foreign research.In addition,with the increase of sensors' data,it is possible to mine useful information from massive multi-source production equipment data.The method of intelligently acquiring target data characteristics has also made great progress.This paper focuses on the research of intelligent fault diagnosis and fusion technology of bearing multi-source information.The main research work is as follows:Firstly,an improved D-S(Dempster-Shafer)evidence theory algorithm for the conflict problems is proposed.This algorithm uses Pearson Correlation Coefficient(PCC)and 0 element correction to improve traditional D-S evidence theory.The improved method largely considers the importance of each evidence in the overall recognition framework.The result of the fusion can ensure the consistency of the fusion target after excluding the evidence with large errors,improving the accuracy of fusion result and the ability to overcome conflicts.On the basis of improved D-S evidence theory,combined with bearing composite fault diagnosis data,this paper continues to improve the practical application of D-S evidence theory.And the SAE feature self-extraction method is used to extract features of every type fault from each sensor,and compressed features of all fault types are obtained to train the classification model.The SFD(Single Fault Detection)results show a higher recognition accuracy.Then,on the basis of single fault detection research,this paper continues to study the compound fault diagnosis framework.The diagnosis of compound fault is realized by adding the results of single fault detection.The experimental results prove the significance and accuracy of the research from single fault detection to compound fault diagnosis,and also prove the applicability of the improved fusion method.Finally,this paper considers the simplification of compound fault diagnosis process,feature extraction methods and fusion methods to reduce the complexity of manual improvement and calculation,to achieve the automation of feature extraction and fusion.Therefore,this paper finally abandons all complex calculations and fusion methods,and designs a two-level CNN bearing compound fault diagnosis method.Through multiple sets of experimental analysis and comparison,this method has a high recognition effect on the compound fault diagnosis of bearings,which proves the effectiveness and advantages of the two-level CNN method.In summary of the research and the verification of the experiment,the multi-source information fusion method can synthesize the advantages of different sensors in the face of complex bearing compound fault signals,achieve the diagnosis goal as much as possible with the actual situation,and improve bearing compound fault diagnosis accuracy.In addition,the use of cutting-edge algorithm technology to automatically extract data features also plays an important role in improving the accuracy of bearing compound fault diagnosis and promoting the application of practice.Finally,it is concluded that the combined research of feature self-extraction and multi-source information fusion has great research potential and value in coping with bearing compound fault diagnosis.