Health Management Of Rotating Machinery Incorporating Measuremental Uncertainty

Intelligent health management of rotating machinery is an important means to realize the operation and maintenance of rotating machinery.In the whole life of rotating machinery,the role of health management technology is to work together with the control module to ensure the reliable and stable operation of the rotating machinery,so that the rotating machinery has a long working life.And health management as a multidisciplinary integrated based technology,which involves a large number of basic disciplines,including a wide range of research content.The fault diagnosis,operation condition monitoring and life prediction are three practical technologies in the rotating machinery health management system.In this paper,the author analyzes their characteristics and relationship between them,and gives the method of reducing the data / parameter uncertainty in the application.The main research contents and results of this paper include the following aspects:First of all,the problem of lack of adaptability when facing non-stationary signal processing problems is presented.The method named “Uncetainty extraction” is proposed to obtain a more representative feature set in the feature extraction of fault rotating machinery monitoring signal,which avoids the problems caused by the introduction and selection of the decomposition base in the traditional signal feature extraction.Secondly,a decision tree algorithm based on entropy ratio which is used to achieve feature selection is researched.Compared with the classical decision tree algorithm,the improved decision tree algorithm has better generalization ability and shorter operation time.Combined with the original signal extracted by the uncertainty components and the decision tree algorithm to filter out a better representation of the feature set,the test results show that the selected feature has a higher classification accuracy.Thirdly,in view of the uncertainty of classification data,the possible sources of uncertainty in fault diagnosis are analyzed.Then,a class of uncertain support vector machine algorithm and a class of uncertain neural network algorithms are studied,and they are applied to the classification of data oriented features.A rotor fault diagnosis method considering the uncertainty of data is established by combining the method of uncertainty extraction,feature selection and two feature classification algorithms.The method has a high diagnostic rate on the test data set,and it is proved that the method has good generalization ability.Fourthly,based on the previous diagnosis method proposed by the single fault mode,the numberand mode of the fault source are studied.The blind source separation method considering the uncertainty of source number is studied.Fifthly,a novel detection algorithm based on uncertainty reduction is proposed for the health monitoring of rotating machinery under non-stationary condition.The limit distribution functions and their properties are studied.A health monitoring technology with low false alarm is established based on non stationary data analysis.Finally,the uncertainty model of the monitoring parameters is introduced into the life prediction process of the rotating machinery,and the propagation law of the uncertain parameters in the life prediction model is obtained.The conclusions can be used to achieve the effective control of uncertainty,and on the basis of the process to establish a limited information based,relatively reliable life prediction method.Several algorithms and strategies are proposed in this paper to improve the health management system,which can adapt to the actual conditions of the rotating machinery,and achieve more precised health management.