Elevator Fault Diagnosis Based On High Order Spectrum And Support Vector Machine

With the application of elevator, people put forward higher demand on thereliability of the elevator and the recognition of fault. Considering the complexity andthe fault of the elevator system concealment, relying on manpower to troubleshootingis not enough. The contribution of this paper is using the time series of high ordercumulant and support vector machine based on time sequence of higher ordercumulants coefficients.We analyzed the collected signal with this two methods toidentify the fault state of elevator.Vibration signals of elevator are nonlinear, non Gaussian. High order cumulantwhich can restrain Gauss noise and symmetrical distributed noise is a powerful toolfor processing the non-stationary, non-Gaussian, non-minimum phase and noncausal signal. Applying its research on elevator, trying to find out the elevator signalspectrum characteristics under different conditions, and find out the differencebetween the spectral features are significant. At the same time, coefficient of ARmodel established by higher order cumulant are used as inputs of support vectormachine, the normal and fault condition of the elevator are used as the output, whichare learned and predicted with support vector machine to achieve a result ofautomatic fault recognition. We can conclude the following conclusions:(1)Because of the nonlinear and non Gaussian signals, whose frequency andfrequency components two phase coupling occur between each other, the distributionof spectral peaks are different from each other no matter what the condition of theelevator is. So we can identify which status the elevator is at from the amplitudeinformation reflected in spectrum.(2)In the slice spectrum, the number of spectral peak and the frequencycorresponding to spectral peak are vary from the elevator’s working condition.Spectral peak appear at both dominant frequency and multiple frequency, whichshows nonlinear phase coupling occur at these frequencies.(3)We combine the AR model coefficients with support vector machine applying to signal analysis of elevator. The kernel principal component analysis is introduced tocalculate the coefficients of the contribution rate in order to reduce the inputdimension of support vector machine, which enhances the generalization ability of themodel.(4)The penalty factor and kernel parameter values would affect the predictionaccuracy in the process of modeling. For this matter, we choose cross validationoptimized parameters to select the better parameters combination. Then build supportvector machine model of training samples with the selected parameters to forecast thetraining samples, based on which the fault recognition rate reached up to86.11%.