Study On Residual Life Model Of MOV In Railway SPD

With the development of intelligent operation and maintenance of railway system,the demand for on-line condition monitoring and residual life prediction of SPD(Surge Protection Device)has become increasingly urgent.Railway SPD is a kind of lightning protection equipment with a large number of uses,and plays the role of absorbing surges and protecting equipment in railway surge sensitive places.MOV(Metal Oxide Varistor)is the core component of SPD,which is accompanied by its own deterioration in the process of surge protection.The key point of the research on the life prediction of railway SPD is the description of MOV deterioration process.It has become very necessary to carry out research on the residual life model of MOV.Firstly,the deterioration exploratory experiment of MOV was carried out in this paper.Surges with the type of 8/20μs was used in the experiment to impact MOV,until MOV was completely damaged.Fourteen kinds of electrical parameters,clamping voltage waveform and surface temperature of the MOV were collected over the MOV entire life cycle,and the trends of these parameters were analyzed.The analysis results show that 14 electrical parameters can characterize the deterioration of the MOV,but these parameters are not independent and can be merged,so they need to be screened when they are used.Then,based on the results of the deterioration exploratory experiment,five kinds of deterioration sensitive parameters were screened.Taking the five parameters as input,the remaining life model of the MOV based on machine learning algorithms was constructed with the method of k nearest neighbor and linear regression.The k nearest neighbor algorithm is used to discriminate whether the MOV is damaged or not,and the linear regression algorithm outputs the remaining lifetime value of the MOV.Next,the model was tested using experimental data.The test results show that the model can well characterize the deterioration of MOV,but the machine learning method has the disadvantages of serious dependence on data.Finally,based on the historical surges,self-characteristic parameters and the selected deterioration sensitive parameters of MOV,this paper constructed the deterioration kernel to represent the health degree of MOV.According to the theory of stochastic process,taking the deterioration kernel as the unified input,two residual life models of MOV were constructed by using Markov Chain and Gamma Process respectively,and the influence of key parameters in the models on the output results was analyzed.Two models were tested and compared by using 50 random surges with the type of 8/20μs.The results show that both models can describe the deterioration process of MOV.The model based on Markov Chain has a clear physical concept,a small amount of computation,and a consistent deterioration rate in the middle and later stages.However,the model based on Gamma Process deteriorates slowly in the middle stage and sharply in the later stage,and the change trend is more close to the actual situation,but the calculation amount is large.Considering the limitation of computing resources of field monitoring devices,it is an economical and safe choice to adopt the life model based on Markov Chain.