Diesel Engine Fault Diagnosis Based On Locally Line Embedding Algorithm

Since diesel engines play an important role in our daily production,which promote therapid development of the diesel engine fault diagnosis technology in recent years.Dieselengine complexity of structural lead to the fault in the form presented diversity,coupled withthe impact of poor working conditions and the accuracy of the information collectionsystem,would poorly the diesel fault diagnosis rate.How improve the accuracy and efficiencyof the diesel engine fault diagnosis,which is the main topics researched by scholars in recentyears.Wavelet packet is one of the more effective extraction feature of the original signalmethod.Locally linear embedding algorithm is a dimensionality reduction method that fornon-linear characteristics of the signal,such dimension reduction is not simple optimization innumber,but in the case of the unchanged the original data and map hight-dimensional space toa low-dimensional space that is the feature value of the secondary extract.Information fusiontechnology that the information collected by different position sensor and then dealt byassociation,filters,links and synthetic,and arrive at a precise fault information.In thispaper,combine the wavelet packet,locally linear embedding algorithm and sensor informationfusion method together,Using their advantages to research fault diagnosis of the diesel engineto achieve better diagnostic results.First,by major failure in the form of the diesel engine vibration signal characteristicanalysis,using wavelet packet energy spectrum analysis method to extract the feature value ofeach measuring point and constitute a non-linear hight-dimensional space.Apply locally linearembedding algorithm dimensionality reduction the feature values that each measurementpoint vibration signal of diesel engine. Second,the locally linear embedding algorithm is meaning for high-dimensional featurevector optimization problems and summarized the present problems that the optimizationmethod of feature values,there are some problems in this program,the improve method relatedto is proposed,improved LLE algorithm that can optimization field parameters k and caneffectively distinguish different fault types of data.Finally,combine the low-dimensional vector eigenvalues together,and put it intoSOM-BP neural network to diagnosis,compare with the results of feature values that fuse afterand before,it can shows that the diagnosis system can improve the diagnostic accuracy andefficiency.