Research On Fault Diagnosis Of Rolling Bearing Based On Online Incremental Dimensionality Reduction And Optimization Support Vector Machine

In rotating machinery and equipment,rolling bearings undertake the task of connecting and fixing,so they can be considered as the core component of this type of machinery.However,the actual operating environment of rolling bearings often contains many unfavorable factors,and its performance will be affected to a certain extent,and even damage or failure may occur.Once such a situation occurs without timely treatment,it will directly affect the normal and stable operation of mechanical equipment,cause equipment damage,and even cause safety accidents.Therefore,the fault diagnosis of this part has important practical significance.In this subject,a set of novel diagnostic techniques have been developed for the problems of rolling bearing fault feature extraction,fault feature online dimensionality reduction screening and fault classification.The technology consists of three steps: First,apply refined time-shift multi-scale fuzzy entropy(RTSMFE)to comprehensively mine the fault information of bearing signals,and construct a fault feature set with higher dimensions.Then,use the online incremental dimensionality reduction method—topology learning and out-of-sample embedding(TLOE)to perform online dimensionality reduction screening on the high-dimensional fault feature set.In this way,a low-dimensional,easy-to-identify feature set can be obtained.Finally,the marine predators algorithm-based support vector machine(MPA-SVM)is used to perform fault classification on the feature set after dimensionality reduction,and the effectiveness of the method is verified through experiments.The main tasks of the subject are as follows:(1)In order to fully mine the fault information of the bearing signal,the multiscale fuzzy entropy(MFE)is combined with the ideas of time shifting and refinement,and the refine time-shifting multiscale fuzzy entropy(RTSMFE)is developed to extract the bearing fault characteristics.Simulation and bearing data prove that the RTSMFE method is more stable than multi-scale fuzzy entropy(MFE)and time-shift multi-scale fuzzy entropy(TSMFE).(2)In order to overcome the defect that common dimensionality reduction methods cannot perform online dimensionality reduction on incremental data,the TLOE algorithm is proposed for high-dimensional fault feature dimensionality reduction screening to remove redundant information such as noise in high-dimensional features.Thereby,more sensitive and effective low-dimensional fault characteristics can be obtained.The validity of the method is verified by bearing experimental data and compared with the commonly used dimensionality reduction methods.(3)In order to achieve higher-precision bearing fault diagnosis,the best SVM parameters are found through the marine predator algorithm,and then the low-dimensional fault features are identified and classified.The superiority of this method is verified through two sets of simulation experiments and bearing experimental data,and it is compared with several commonly used failure mode recognition methods.(4)Based on the aforementioned methods,a fault diagnosis model for rolling bearings based on RTSMFE,TLOE and MPA-SVM is developed.In order to verify the effectiveness of the proposed model,fault diagnosis experiments are carried out using rolling bearing data,and the results prove the effectiveness and superiority of the method.At the same time,this model is used to diagnose the planetary gearbox fault,which verifies the generalization performance of the model.