| S700K turnout is the key equipment to realize line conversion in high-speed railways.With the further development of the operation density and speed of high-speed railway,the classification of S700 K turnout under different state levels such as health,sub-health,failure,and serious failure has become the primary premise to ensure the safe operation of the railway.Because of the consistency between S700 K turnout’s action power curve characteristics and its state information in the process of action,this thesis aims to fully extract the power curve characteristic indexes that can represent the state information.Taking the state feature vector of the S700 K turnout power curve as the input object,a fault diagnosis algorithm of S700 K turnout based on a multivariable support vector machine is proposed.Based on the deficiency of fault diagnosis,a fuzzy clustering analysis method is proposed to evaluate the running state of S700 K turnout.The research content of this thesis mainly includes:(1)In the extraction of state information,firstly,using the advantage of variational mode decomposition(VMD)in processing nonlinear and non-stationary time series signals,the power curve of S700 K turnout is decomposed to obtain the detailed components with different frequency characteristics.Then,the complexity of the signal is quantified by improved multiscale permutation entropy(IMPE)to characterize the micro features of different detail components.Finally,to eliminate signal redundancy and fully characterize state information,the kernel principal component analysis algorithm is used to process feature sets,and the eigenvalues with a contribution rate of more than 95% are selected as running state feature vectors.(2)In the fault diagnosis,a fault diagnosis algorithm of S700 K turnout based on multivariable support vector machine(MSVM)is proposed.On the one hand,to prove the effectiveness of the improved multi-scale permutation entropy algorithm in state feature extraction,the fault diagnosis rates of VMD-MPE-MSVM and VMD-IMPE-MSVM algorithms are compared respectively.On the other hand,the stability of fault diagnosis rate under different training sets samples is poor and limited by the difficulty of S700 K turnout sample data collection,which leads to the subsequent state evaluation method without sample training.(3)In the running state evaluation,firstly,the standard set of typical power curves of S700 K turnout under health,sub-health,failure,and serious failure is established.Then,the state feature vectors of power curves under different levels are taken as input vectors,and the fuzzy clustering analysis algorithm is used to establish the S700 K turnout running state evaluation model.Finally,a field example is used to verify that the state evaluation is directly expressed in the form of a dynamic cluster graph when the power curve is input into the state evaluation model. |
