Research On Fault Diagnosis Of Key Components Of Mechanical Equipment Based On Vibration Signal

With the promulgation and issuance of "Made in China 2025" in 2015,the comprehensive deployment of the strategy of manufacturing power has promoted the transformation and upgrading of the manufacturing industry to the direction of intelligence and information technology,and the mechanical equipment has become increasingly large and complicated,which will increase the difficulty of fault maintenance and detection and the maintenance cost.Therefore,it is of great significance to identify the failure mode when the key components of the equipment fail so as to provide a basis for the maintenance strategy.As most of the current researches are focused on a single sensor and a single classification model,multi-source information cannot be effectively utilized,and the reliability and accuracy are low.Therefore,the information fusion theory is introduced into the fault diagnosis research of key components of mechanical equipment in this paper,and the feature-level and decision-level fusion models are constructed.The specific research contents are as follows:(1)The Ensemble Empirical Mode Decomposition(EEMD)method is used to decompose the original signal to obtain several Intrinsic Mode Function(IMF),and then the IMF components are screened according to the IMF screening criteria combined with the correlation coefficient and kurtosis to reduce redundancy and noise.The screened IMF components are reconstructed,and the feature space was obtained by calculating the time domain and frequency domain features of the new signal.(2)Based on the features extracted by the above methods,this paper builds a feature-level fusion model and proposes a two-stage fusion algorithm to fuse the features of multi-source signals,effectively utilizing the vibration signals of multiple sensors to avoid the uncertainty of single information,realize information complementarity,improve fault tolerance,and improve the accuracy of fault diagnosis.Firstly,the same feature in different feature spaces is fused with parallel algorithm to avoid the feature high-dimensional phenomenon in the fusion process.Then,Kernel Principal Components Analysis(KPCA)method is used to remove the correlation operation to extract the principal components and obtain the fusion features.(3)The reliability and stability of a single model for fault diagnosis are low.In this paper,a decision-level fusion model is proposed to fuse the classification results of multiple algorithms,and the Stacking algorithm is used as the fusion algorithm.The advantages and disadvantages of Support Vector Machine(SVM),K-Nearest Neighbor(KNN),Naive Bayesian(NB),AdaBoost,Gradient Boosting Decision Tree(GBDT)and Random Forest(RF)algorithms were analyzed as well as the experimental results to determine the structure of Stacking fusion model.SVM,KNN,NB,AdaBoost,GBDT and RF were used as the first primary classifier of Stacking.RF is used as a meta-classifier to fuse the first-layer prediction results.The XJTU-SY bearing data set provided by Xi’an Jiaotong University and Zhejiang Changxing Shengyang Technology was used to verify the proposed method.And set up a control experiment for comparative analysis.The results show that EEMD and IMF screening criteria can effectively extract features and reduce redundancy and noise.The classification effect of fusion features obtained by the feature-level fusion model and two-stage fusion algorithm is better than that of single sensor,and the classification effect of Stacking fusion model is better than the six single models such as SVM and KNN.