Application Of Sparse Optimization And Dimensionless Parameters To Fault Diagnosis For Rotating Machinery

Nowadays,rotating machinery has been widely used in a range of pillar industries,however,unexpected failure in such machinery always cause breakdowns and lead to significant economic under the working condition of high-speed and heavy-load.Hence,to monitor the health condition of rotating machinery,the machine fault diagnosis is of great industrial significance.This thesis aims to present the methods of sparse optimization and dimensionless parameters to condition monitoring and trend prediction of roller bearings.The followings are the descriptions of each developed effective method and its tested results.1.A novel combination of compressive sensing for vibration signals is presented to overcome the problems such as mechanical vibration signals high speed transmission and long-term storage online are limited by hardware.This combination relies on the strategy which acquires and processes signal based on its sparsity.The vibration signals could keep good reconstruction accuracy with high compressive ratios by using DCT(Discrete Cosine Transformation)for sparse,Gaussian random matrix for compressive and Lasso-LSQR via ADMM for reconstructing which takes priority over Basis Pursuit and Lasso.According to the results,the effectiveness of method can be validated for vibration signals.2.In order to extract and recognize the fault features,a novel method based on the improved local mean decomposition(LMD),permutation entropy(PE)and the optimized K-means clustering algorithm is put forward in this paper.The improved LMD method is proposed based on the self-similarity of vibration signal extending the right and left side of the original signal to suppress the end effect of the original one.After decomposing the roller bearing vibration signal into a set of product functions(PFs),the PE is utilized todisplay the complexity of the PF component.Then,the optimized K-means algorithm is used to cluster analysis as a new pattern recognition approach,which has the priority of recognition accuracy compared with the classic.Finally,the experiment results show the proposed method is effectively to fault extraction and recognition for roller bearing.3.The last method is aimed to predict the remaining useful life of roller bearings by models.We extract the hop factors as the eigenvalue and forecast the trend by short-range dependence model(ARIMA)and long-range dependence model(FARIMA)which can be mixed by MIX-ARMA.The results show that the hop factors with long-range dependence which tested by R/S method is more close to the FARIMA model and its predictions provide the practical significance in the decision making process.