Gear Fault Diagnosis Based On Mixture Model Of Scale Mixtures Of Skew Normal

The gearbox of large metallurgical equipment is characterized by heavy load,high speed and bad working environment.Its vibration signal has the characteristics of few fault samples,high noise and overlapping fault features.According to classical fault diagnosis theory,the vibration signals of gears in normal state obey Gauss distribution,but the probability distribution of vibration signals in fault state will change.Therefore,as long as the probability distribution of the gear state signal is accurately described and captured,the fault diagnosis of the gear can be carried out.However,in the industrial environment,the vibration signal characteristics of the gearbox do not strictly obey the Gauss distribution,but have high-dimensional,multivariate and multiple long-tail characteristics,which makes it difficult to diagnose the fault state of the gears in metallurgical equipment accurately.In this paper,mixture model of scale mixtures of skew normal(MMSN)is proposed to describe the probability distribution of the characteristics of gear vibration signals.Compared with the classical Gaussian mixture model,this model adds a family of parameters reflecting skewness,it can reflect the distribution direction and tightness of fault data,so it is more flexible.The feasibility and practicability of the MMSN in the application of gear fault diagnosis engineering are verified by applying it to the research of gear fault early warning by cluster and identification by classification in mechanical field.The main work of this paper is as follows:(1)A method about gear vibration signal feature description based on the MMSN is studied,and the parameters of the MMSN are estimated based on the maximum expectation iteration algorithm.(2)A method about gear fault clustering to early warning based on MMSN is proposed.First,the feature sample set of gear fault measured data was established,and then the results about early warning were compared with Gaussian mixture model from the view of univariate and multivariate,the feasibility and practicability of MMSN compared with the Gaussian mixture model in the gear fault clustering to early warning engineering are verified.(3)A method about gear fault classification to identification accurately based on MMSN is proposed.Firstly,the feature sample set of gear fault open data is established,and combined with the feature sample set of gear fault measured data,the gear fault is identified by MMSN.And the identification results were compared with those of probabilistic neural networks,support vector machine and Gaussian mixture model,the comparison results of the two data sets shows the feasibility and practicability of MMSN in the engineering application of gear fault identification accurately.