Applied Research On Relieff Weighted Feature Selection Algorithm In Fault Diagnosis Of Rotating Machinery

Rotating machinery is one kind of mechanical equipments that is widely used in theindustrial device, such as steam turbine, compressor, fan and mill. Through the work ofthese devices’ condition monitoring and fault diagnosis, to ensure the device’s safe andreliable operation, it will obtain huge economic benefit and social benefit. Taking ReliefFweighted feature selection algorithm into rotating machinery’s fault diagnosis, it willrealize the dimension reduction of the original high-dimensional feature vector that has aplurality of information.Feature selection is a process of reducing features’ space dimension through choosingsome most effective features from a group of features. It will lower the training time ofbuilding the model and improve resolver’s correct rate. It is one of the key problems ofpattern recognition. ReliefF weighted feature selection algorithm is currently recognizedfeature validity assessment method that has superior performance. This method caneffective evaluate feature’s ability of classification, but it do not consider features’correlation, then it can’t remove the redundant feature. For this reason, the text integratesfeature correlation method to remove the redundant feature that have close classificationcapability.In order to verify the validity and superiority of this method, the text analyze thesignal of swashplate axial piston pump vibration in the hydraulic system of materialtesting machine and the signal of bearing vibration in the integrated laboratory mechanicalfault simulation test bench (MFS-MG). This method can determine the resonancefrequency range of hydraulic pump and rolling bearing in every states. Then it providesbasis for wavelet packet bandpass filter. It uses the wavelet packet theory and the Hilberttransform envelope demodulation method to analyze the signal. Extracting envelopesignal’s feature indicators of the amplitude domain and The time-frequency domain as theoriginal feature vector, using ReliefF weighted feature selection algorithm,choosing thefeatures that have strong ability of classification, integrating Feature correlation algorithmto remove redundant features that have similar ability of classification, to form t he featurevectors that have lower dimension. In the end, it deal the samples that is after featureselection with pattern recognition using K-means clustering algorithm. The text usesMATLAB sofeware to program, proves the effectiveness of ReliefF weighted feature selection method in the fault diagnosis of rotating machinery, and compares with featuredimension reduction method of principal component analysis (PCA) that have widelyapplication today. Then it proves that ReliefF weighted feature selection algorithm keepsthe selected features’ original physical meaning and ensures the accuracy of classifierbetter. It reflects the superiority of the ReliefF weighted feature selection algorithm in thefault diagnosis of rotating machinery.