【作者】 刘勤明；

【导师】 董明；

【作者基本信息】 上海交通大学 ， 机械工程， 2014， 博士

【摘要】 随着现代工业技术的发展，设备健康预测和维护都直接影响着企业的生产经营和经济效益。设备的可靠性和维护效果保证了其正常运行，是企业生存的必要条件。因此，设备维护在企业生产经营中的作用和地位日益突出，是企业降低生产成本和保证生产效率的基础。半个多世纪以来，不少学者针对设备的维护进行了许多研究，但是，很少研究健康预测对维护策略的影响。本文立足于过去几十年国内外的设备健康预测和设备维护相关研究的基础上，分析了设备的运行状态，描述了设备的衰退趋势，实现了有效的在线健康预测，为设备的维护提供了决策依据。并且，维护模型考虑了维护资源约束与维护风险。本文首先应用隐式半马尔可夫模型(HSMM, Hidden semi-Markov Model)对设备在线运行过程中的健康进行识别与预测，提出了单监测信息在线健康预测方法。其次，基于单监测信息在线健康预测方法，建立了多监测信息在线健康预测模型和剩余寿命预测模型。然后，基于设备在线健康预测和衰退性能的预测，建立了设备的集成动态维护模型和多部件设备的维护模型。基于前人的研究成果和大量实际的工程经验，本论文在基于设备衰退机制的在线健康预测及设备维护集成研究中主要做了以下三方面的研究工作：(1)针对单监测信息离线健康预测的不足，建立了单监测信息在线健康预测模型基于HSMM和SMC的理论，建立了设备的单监测信息在线健康预测模型。提出了一种联合多步向前健康识别算法，用于在线识别设备健康状态。并且，将在线线识别的健康状态用作寿命预测的分析基础和依据，提出了在线剩余寿命预测模型。本文提出的联合在线健康模型，是基于HSMM丰富的数学结构和SMC的在线特征而来的，利用健康预测对设备的性能状态进行量化，描述设备的实际运行状况。为了说明方法的有效性，通过案例分析对提出的在线健康预测方法进行验证。结果显示提出的方法很好的描述了设备的健康变化情况。并且，与一步向前识别算法、多步向前识别算法和HSMM进行了比较分析，说明了提出的单监测信息在线健康预测模型的有效性和准确性，为后续内容的发展提供了分析依据和基础。(2)针对单监测信息健康预测的不足，建立了多监测信息在线健康预测模型为了提高在线健康预测的精确度，基于单监测信息在线健康预测方法，提出了多监测信息在线健康预测模型。为了降低模型的计算复杂性，对HSMM的基本算法进行了修正，提出了修正的前向-后向算法、Viterbi算法和Baum-Welch算法，计算复杂性从O((MD+M2)T)降低到O((D+M2)T)。在此基础上，建立了自适应HSMM，来处理多监测信息的在线健康预测问题。对于自适应HSMM，利用最大似然线性回归训练对输出概率分布和驻留概率分布进行自适应训练，处理多监测信息之间的差异性，进行有效的多监测信息在线健康预测。结合失效率理论，提出了多监测信息在线剩余寿命预测模型。为了说明方法的有效性，通过案例分析对提出的多监测信息在线健康预测方法进行验证。在案例分析中，与单监测信息在线健康预测相比，多监测信息在线健康预测在设备健康状态识别、诊断、预测和计算复杂性方面，具有更好的有效性和准确性。也为后续设备维护的研究，提供了基础。(3)针对传统维护模型的不足，建立了基于在线健康预测的集成动态维护模型首先，针对传统的设备维护模型的不足，集成维护模型考虑了设备的衰退性能(退化和老化信息)，集成了设备的诊断信息和预测信息，并且，以总的维护成本(故障成本、维护成本和资源成本)和总维护时间为目标，建立了基于在线健康预测的两层集成动态维护模型。在模型中，同时考虑了备件和维修人员的双约束条件。并且，针对小修和大修的维护方式，引入了维护风险因子。其次，基于集成动态维护模型，针对多部件设备的特点，建立了多部件设备的维护模型。多部件设备的维护决策包括性能衰退、维护方式和维护费用三部分内容，在性能衰退方面，通过在线诊断信息和预测信息得到设备故障率变化趋势；在维护方式方面，定义小修、大修和更换三种维护方式；在维护费用方面，考虑了故障成本、维护成本、资源成本和停机成本四部分，在每次维护活动的费用模型基础上，建立了多阶段的总费用率模型。最后，为了说明模型的有效性，通过案例分析对集成维护模型进行验证。根据案例分析可知，与定期维护模型和纯动态维护策略相比，集成动态维护模型具有更好的有效性和准确性。并且，通过对多部件设备的维护模型分析可知，与周期性维护模型相比，提出的多部件设备维护模型在总费用率、生命周期维护活动次数和设备利用率方面，具有很好的有效性。本文的三个研究内容相互之间联系紧密，形成了一个系统性的设备维护决策框架。基于单监测信息在线健康预测思想，提出了多监测信息在线健康预测方法；基于在线健康预测，提出了考虑设备衰退性能和维护资源的集成动态维护模型；基于集成动态维护模型，建立了多部件设备的维护模型。本文所做的研究内容有助于提高企业的维护水平和设备可靠性、降低维护成本、提高设备利用率，最终提高企业的竞争力。拓展了制造系统的维护管理领域，为制造企业维护策略的制定提供决策支持和科学有效的指导。更多还原

【Abstract】 With the development of modern industrial technologies, the health management,and maintenance directly affect the production operation and the economic benefits ofenterprises. The reliability and maintenance status of equipment ensure the normaloperation of the system and are the important survival condition of enterprises. Thus,the equipment maintenance has increasingly prominent role and status in theproduction and operation of enterprises, and is the basis to reduce production costs andensure the production efficiency. Some maintenance strategies are studyed, but theyignore the influence of health prognosis.Based on the available research on equipment health prognosis and maintenance,it studies the equipment operational state, describes the equipment degradation andobtains the equipment health prognosis, which provides the making decision basis forthe integrated dynamic maintenance planning. The resources and maintenance risk canbe integrated into the dynamic maintenance model. The single-sensor online healthprognosis is studied first. Hidden semi-Markov model (HSMM) is applied into theonline health states recognition and prognosis of equipment. Then, based on the singlecondition information online health prognosis, the multi-sensor health prognosis isstudied, and the residual useful lifetime prediction model can be proposed. Based onthe online health prognosis, the integrated dynamic maintenance model can beproposed, and the maintenance area can be improved. Finally, based on the integratedmaintenance model, the maintenance of multi-component equipment is described.Based on the above previous research and literature, the disadvantage and advantage of this research can be analyzed. This paper studies the health prognosis andmaintenance strategy. The following parts can be focused.(1) The single-senor online health prognostic model can be proposedThis paper addresses prognostic methods based on HSMM by using sequentialMonte Carlo (SMC) method. HSMM is applied to obtain the transition probabilitiesamong health states and the state durations. The SMC method is adopted to describe theprobability relationships between health states and the monitored observations ofequipment. This paper proposes a novel multi-step-ahead health recognition algorithmbased on joint probability distribution to recognize the health states of equipment andits health state change point. A new online health prognostic method is also developedto estimate the residual useful lifetime (RUL) values of equipment. At the end of thepaper, a real case study is used to demonstrate the performance and potentialapplications of the proposed methods for online health prognosis of equipment.(2) The multi-sensor health prognostic model can be proposedFirst, based on the single-senor online health prognostic model, the basicalgorithms of HSMM are modified in order for decreasing computation and spacecomplexity, and the modified HSMM with multi-sensor information is applied, inwhich the hidden degradation process can be seen as the system state. Then, theadaptive HSMM (AHSMM) is proposed for hidden degradation state identification,while the maximum likelihood linear regression transformations method is used totrain the output and duration distributions to re-estimate all unknown parameters. TheAHSMM can also be used to obtain the transition probabilities among health states andhealth state durations of a complex system. Finally, the main results are verified by acase study, and the results show that the AHSMM with multi-sensor information has abetter performance for health prognosis than HSMM.(3) Based on online health prognosis, the integrated dynamic maintenancemodel can be proposedThe integrated dynamic maintenance includes the deterioration and aginginformation into the maintenance for improving the overall decisions. This paperpresents an integrated decision model which considers both health prognosis and theresource planning. Based on online health prognosis, the system multi-failure states canbe classified and the transition probabilities among the multi-failure states can be generated. The upper triangular transition probability matrix is used to describe thesystem deterioration and the changing of transition probability is used to denote thesystem aging process. And the resource planning is integrated into the maintenancemodel for different failure states. Finally, a bi-level dynamic programmingmaintenance model is proposed to obtain the optimal maintenance strategy and therisks of maintenance actions are analyzed.Based on the integrated maintenance, the multi-component maintenance model isestablished, which includes degradation, maintenance cost and maintenance action. Forthe degradation, the failure changing trend can be obtained by the diagnostic andprognostic information. The minor maintenance action, the imperfect maintenanceaction and the replacement maintenance action are defined, and the impaction on thefailure is described. For maintenance cost, the failure cost, maintenance cost, resourcecost and downtime cost can be considered. Based on the cost model of eachmaintenance activity, the multi-stage total cost rate model can be proposed.A case study is used to demonstrate the implementation and potential applicationsof the proposed methods.The three parts of research contact closely with each other, and forms a systematicframework for equipment maintenance scheduling policy. Based on the online healthprognosis, this paper develops the integrated dynamic maintenance strategy consideringthe degradation and maintenance resources. The research can improve the maintenancelevel and reliability, reduce the maintenance cost, raise utilization, and ultimatelyenhance the competitiveness for enterprises. It can also expand the field of maintenancemanagement of manufacturing systems, and provide effective decision support andscientific guidance for system maintenance strategies of manufacturing enterprises.更多还原