Research On Key Support Technologies For Prognostics And Health Management Of Rotating Machinery

Equipment in the petrochemical industry often requires long-term operation,but its stability is low and the risks are high.As the "power heart" of petrochemical equipment,rotating machinery often causes huge production losses,serious economic losses,and even severe safety and environmental accidents if it malfunctions and causes unplanned shutdown.Accordingly,ensuring the safety and reliability of refining equipment operation,completing potential fault warning and accurate diagnosis,thereby minimizing unplanned downtime,and achieving the prognostics and health management(PHM)of rotating machinery are crucial for making scientific maintenance decisions.The research content of PHM includes warning,diagnosis,evaluation,and prediction.Among them,fault warning and diagnosis are the primary focus fields in PHM engineering and the main supporting technologies for PHM.This paper focuses on two key technologies of PHM for typical components of rotating machinery,providing effective solutions and methods for intelligent fault detection and diagnosis technology to further move towards practical engineering applications and meet the needs of equipment predictive maintenance.Summarize the main research contents of the paper as follows:(1)Research on incipient fault warning method for rotating machinery.An online fault warning model for rotating machinery has been constructed,which self-learns alarm reference lines based on historical normal operating conditions data.Firstly,the complete ensemble empirical mode decomposition with adaptive noise is used to denoise the raw vibration signal,improving the feature differentiation between normal state data and incipient fault data.Whereafter,the characteristic indicators of the signal are extracted and smoothing filtering to enhance the sensitivity of the model,while avoiding the impact of occasional outliers.Finally,the incremental support vector data description algorithm is applied to alleviate the situation of excessive matching of training samples and adapt the model based on new data samples,which makes it suitable for different device performance degradation scenarios.A fault warning method based on signal adaptive decomposition and incremental support vector data description is proposed.The real-time performance and robustness of the proposed method are verified based on rolling bearing full life cycle data and rotor system engineering case data of "Run to failure".(2)Research on fault pattern recognition method for rolling bearing.An online fault identification model for rolling bearing has been constructed,which utilizes the raw vibration signals of four types of health states under different operating conditions as training data.Firstly,the maximum kurtosis value method is applied to determine the parameter preset range of improved variational mode decomposition,then the sensitive components are screened for reconstruction.Whereafter,the refined composite multiscale dispersion entropy of the reconstructed sensitive components is calculated,and the entropy values of the first five scales are screened as feature value.At the same time,a fault knowledge base with one-to-one correspondence between feature values and health states is structured.Finally,by comparing the computed outcomes of the optimized support vector machine with the knowledge base,the fault identification conclusion is obtained.A fault pattern recognition method based on refined composite multiscale dispersion entropy is proposed.The practicability and reliability of the proposed method are verified based on open experimental data and actual engineering case data.(3)Research on typical fault diagnosis method for rotor system.An online fault diagnosis model for rotor system has been constructed,which utilizes five typical fault data from five different rotor equipment as training data.Firstly,a multi-source domain feature space is constructed based on non-homologous data from different devices and operating conditions,and the commonality diagnostic knowledge is learned.Next,the joint screening indicator is constructed to adaptively extract and screen sensitive features to highlight fault feature expression.Whereafter,narrowing the differences in feature distribution of cross domain data through weighted semi-supervised transfer component analysis,and obtaining the optimal feature transfer matrix.Finally,the strong classifier constructed by the ensemble feature classification recognition algorithm is applied to diagnosis fault types.A fault diagnosis method based on weighted feature transfer and ensembled classification recognition is proposed.The effectiveness and generalization of the proposed method are verified based on typical engineering fault cases data under across equipment and operating conditions.