Research On Application Of Fault Diagnosis Technology For Petrochemical Rotary Machinery Bearings

The diversity and complexity of the working environment of petrochemical units makes some parts prone to failure,which in turn causes concurrent failures of petrochemical units and poses a serious threat to plant and employee safety.This paper takes rolling bearings,an important component in petrochemical rotating machinery,as the research object,and addresses the problems of missing values in the collected data of multi-stage centrifugal fans of petrochemical rotating machinery,incomplete fault feature representation,and inaccurate feature selection,which make fault characterization difficult.The research is oriented to strengthen the fault characterization capability,and the three aspects of interpolation,noise reduction and feature analysis,and feature extraction and feature simplification of the bearing signal are of great engineering significance to prevent accidents in petrochemical industry.The following are the details of the study:A double regression missing value interpolation method based on Lasso regression and model correction is proposed to address the problem of missing values in the collected data due to the short-time failure of sensing equipment.The dual regression model based on Lasso regression and model modification introduces Pearson correlation analysis as an auxiliary prediction feature on the basis of single-layer regression,which formally simulates the process of forward propagation of neural networks and improves the model prediction capability.Finally,the root mean square error and coefficient of determination are used as evaluation indicators to compare with Lasso regression and KNN interpolation method.It is verified that the root-meansquare error is reduced by 10% compared to single-layer regression,and possesses better model generalization ability.To address the problem of incomplete representation of fault characteristics,autocorrelation analysis,partial autocorrelation analysis,and fast Fourier transform are used to analyze the time and frequency domains of each fault signal of the bearing.Based on the frequency domain analysis map,the noise reduction method based on ICEEMDAN and correlation is applied to overcome the adverse effect of noise on the fault characterization for the problem of signal noise.Secondly,the characteristics of the time domain and frequency domain analysis maps are considered comprehensively,and then features are constructed and feature extraction is completed to achieve the purpose of compressing data and improving information value.For the problem of inaccurate feature selection due to the high time complexity of high-dimensional data and data redundancy,a feature selection method with a multidimensional information fusion judging method is proposed.Compared with traditional feature selection methods,such as the recursive feature elimination method based on random forest,the proposed method not only takes into account the feature importance evaluation,but also takes the inter-feature redundancy and feature column interpretability as supplementary parts of the new evaluation mechanism,so as to complete the information fusion of feature evaluation value and obtain a better feature subset.It is verified that the feature selection method of multidimensional information fusion evaluation can find the feature subset with the minimum number of features while maintaining a high classification accuracy.Finally,the human-computer interface is designed using the Py Qt5 program package to visualize the fault characterization process,which achieves a simplified process and improved efficiency.