Bayesian Network Based Reliability Modeling Approach Considering Multiple Factors For Spindle System In Machining Center

The technological innovation of domestic CNC machine tools and the independence production of key components is an important task of China's manufacturing industry at present.However,due to existing constraints such as the high failure rate and poor reliability,the supporting capacity of domestic functional components is not good enough during the process of transformation and upgrading of the domestic machine tool industry.To reduce the losses caused by unexpected shutdowns and frequent maintenance,users usually resist to choose machine tools equipped with domestic functional components.Thus,it is of great significance to develop reliability modeling technology for domestic functional components to reflect the actual reliability during operation,provide effective decision-making support for operation and maintenance,and finally reduce users' costs and improve their willingness to use domestic functional components.Achieving these can facilitate the development of the machine tool industry in China.As the key functional component of CNC machine tools,spindle system undertakes the task of power and motion transmission and workpiece formation with complex structure,high maintenance cost.Meanwhile it contains complex and dynamic information during operations,such as working conditions and maintenance history.Thus far,most of the existing research on spindle reliability modeling is carried out based on historical fault information without considering many factors in the operation process of the spindle system.Furthermore,the model analysis results do not reflect the actual state of the spindle,preventing it from providing effective information and decision support for the operation and maintenance of the spindle.In view of the above problems,this thesis takes the spindle system of domestic machining center as the research object and conducts research on the reliability modeling method of the spindle system with the support of the National Natural Science Foundation of China and the National Science and Technology major projects.The main content and innovative achievements of this thesis are summarized as follows:(1)The reliability modeling framework of the spindle system based on Bayesian network is established.Considering that the existing spindle reliability modeling methods are unable to realize both system reliability evaluation and weak link identification simultaneously,we decompose the system fault into unit fault as the idea of spindle reliability modeling.The feasibility of applying the fault tree model and Bayesian network model to the reliability modeling of spindle system is discussed.Then,the realization principles of the two models in system reliability assessment and importance analyses are discussed.After that,the expansibility of the model(i.e.,the ability of the model to integrate multi-source information and factors)are compared,finally the basic framework of reliability modeling based on Bayesian network model is established.(2)The condition monitoring information is integrated into the Bayesian network based reliability modeling method of spindle.To solve the problem in which reliability assessment results based on historical fault information cannot reflect the actual reliability of spindle,the condition monitoring information of spindle is introduced into the Bayesian network model as observation node,and two types of failure modes are distinguished in the Bayesian network with degraded failure treated as multi-state nodes.Then the state space partition of observation node and its relationship with corresponding fault node are discussed.Considering the problem that the state division threshold of vibration signal observation node is not fixed and unclear,we propose a state division method based on Mahalanobis distance,through which the relationship between the state of vibration observation node and bearing health node is established.Finally,the parameter learing problem in Bayesian network containing multi-state nodes is discussed,where the marginal probability distribution(MPD)considering the time factor or not and the uncertain conditional probability distribution(CPD)involved with multi-state nodes are estimated based on Bayesian theory.(3)The dynamic Bayesian network(DBN)modeling method of spindle that considers the time-varying operating conditions is proposed.To determine the impacts of the working conditions on the reliability of each element of spindle,the sensitive working conditions of each node is modeled individually thanks to the factorized representation of Bayesian network,And the relationship between the working condition and the parameters of reliability model is established based on accelerated modeling theory and the acceleration factor constant principle,then the DBN representation of failure and degradation model under time-varying operating condition is proposed.To solve the problem wherein the parameters of the degradation model change dynamically under time-varying conditions,we propose a parameter updating method of the stochastic process model under time-varying conditions based on the recursive Bayesian filtering theorem.Finally,this work discusses the reliability evaluation and prediction method based on dynamic Bayesian network and the degradation prediction method under both known and unknown future working conditions.A simulation based degradation prediction approach is developed to solve the problem that the the analytical form of degradation distribution and residual life distribution under variable working conditions is hard to obtain.(4)The thesis proposes a spindle Bayesian network modeling method that considers the impact of maintenance.In particular,we establish a Bayesian network model that considers maintenance correlation and maintenance history information.By extending decision nodes and utility nodes in the Bayesian network,the model is equipped with decision analysis ability.And the expected cost and return on investment(ROI)of decision are taken as utility indexes to deduce the optimal maintenance decision based on the maximum utility principle.Considering the maintenance history of the spindle as dynamic information,and assuming that the maintenance effect of component faults is as good as new,the action nodes is introduced into the dynamic Bayesian network to establish the quantitative influence of maintenance history.Finally,according to the actual situation of the spindle system,three maintenance strategy schemes of post-maintenance combined with preventive maintenance are proposed,four maintenance evaluation indexes are constructed for optimal maintenance suggestions,and a maintenance decision-making process based on DBN inference is also proposed.Through the above research,the proposed method based on Bayesian network has integrated various factors of spindle during the operation and discussed the key issues of reliability modeling.Thereby the analyzing results of the reliability assessment can provide solid and reliable support for the maintenance policies of spindle.