Intelligent Fault Diagnosis Based On Empirical Mode Decomposition And Random Forest

The tendency of machinery equipment development in the modern production process is constantly to be large-scale,complex,high-speed,and intelligent.If the machinery equipment cannot be timely and effectively detected and diagnosed at the early stage,the failure will be exacerbated.It may cause shutdown,even worse,economic loss or personal injury.Therefore,an intelligent,rapid and accurate fault diagnosis of machinery equipment is an essential guarantee for safe production and economic improvement,which shows great practical significance.This paper has studied the core algorithm based on empirical mode decomposition(EMD)and random forest,which is successfully used in fault diagnosis.The main research results of the paper are as follows:(1)A fault feature extraction method based on EMD has been studied.When a mechanical device fails,it is often reflected in the vibration process.The vibration signal contains a lot of rich working condition information,but the vibration signal is typically nonlinear and non-stationary,which makes it difficult to extract features and identify faults.In response to the above problems,it can firstly use EMD to decompose the vibration signal into several stationary intrinsic mode functions(IMF).The dimensionless index and sample entropy of the IMF component are used as the fault characteristic parameters.Then these characteristic parameters are selected to construct feature vectors for fault classification.(2)A fault classification algorithm based on EMD and random forest has been studied.Based on the EMD dimensionless index of the vibration signal as the characteristic parameter,the decision tree algorithm is used to classify the fault,and the decision tree model is optimized by constructing an evaluation function that takes into account both the classification accuracy and the model complexity.The experiment shows that the decision tree is not ideal for fault diagnosis because it is the base learner of the random forest.Therefore,this paper will further research on the fault diagnosis method based on EMD dimensionless indicators and random forest and verify the method's effectiveness and superiority,which shows that the fault classification algorithm based on EMD and random forest is effective.(3)An improved method of DS evidence theory based on Cosine similarity has been studied.In view of the defects of traditional DS evidence theory that evidence conflicts can easily cause misjudgment,poor robustness,and one-vote veto.In the core step of improving DS evidence theory,Cosine similarity is used to directly calculate the similarity between evidence bodies,which overcomes the defeats in using distance to calculate the similarity between the evidence bodies.And the effectiveness of this improved method is verified by practical examples in this paper.(4)An information fusion method based on the voting ratio of diagnosis unit fault classification has been studied.In order to solve the problem of obtaining basic probability assignment(BPA)in the process of information fusion using DS evidence theory,the voting results of the random forest classifier are used as reference.The ratio of the number of votes which is used to identify fault types to the total number of decision trees is used as the BPA of DS evidence theory.And the intelligent fault diagnosis based on EMD sample entropy and random forest is used as the diagnostic unit.The improved DS evidence theory is used to fuse the diagnostic information from the diagnostic units of different sensors to form the fault diagnosis of multi-sensor information fusion.Example verification shows that this method can not only overcome the problem of incomplete information of single sensor and improve the accuracy of fault diagnosis,but also be more robust than single sensor fault diagnosis.