【作者】 任丽娜；

【导师】 芮执元；

【作者基本信息】 兰州理工大学 ， 机械制造及其自动化， 2016， 博士

【摘要】 传统的可靠性评估方法通常不考虑维修等因素对机床可靠性的影响,在“修复如新”的假设下,基于威布尔分布、指数分布和对数正态分布等寿命分布模型对机床可靠性进行评估。而实际上,数控机床在其服役周期内是典型的可修复系统,维修行为贯穿于其使用的全过程,且不同的维修行为对系统的可靠性有不同程度的影响。通常情况下,系统经维修可恢复至“如旧”或“好于旧而次于新”的状态,即“最小维修”和“不完全维修”假设更符合实际情况,因此,在建模过程中,应考虑维修的影响,根据不同服役阶段机床的故障及维修特性,对机床可靠性进行准确有效的评估。本文针对考虑维修影响的数控机床可靠性评估中存在的问题和难点,以随机点过程理论为技术核心,以免疫克隆选择算法和贝叶斯方法为辅助技术手段,构建了数控机床不同服役阶段的可靠性评估体系框架,重点研究了早期故障阶段、耗损故障阶段以及整个服役阶段的机床可靠性评估技术,为合理制定维修决策、提高数控机床的可靠性提供理论参考和支撑。主要研究内容如下:(1)通过对机床整个服役阶段进行划分及对各服役阶段的故障及维修特性进行分析,得出不同维修条件下不同服役阶段机床可靠性评估中存在的关键问题。针对这些问题和难点,构建数控机床服役阶段可靠性评估框架。(2)最小维修条件下,基于改进的幂律过程模型I和模型II对处于耗损故障阶段的数控机床可靠性进行评估,并针对模型参数和可靠性指标的求解问题,提出了可靠性模型参数的免疫克隆极大似然估计法。将参数估计问题转化为带有约束的优化问题,以负对数似然函数最小为目标函数,通过模拟生物免疫系统的克隆选择过程对目标函数进行优化,进而得到模型参数和可靠性指标的最优极大似然点估计和区间估计。通过实例验证表明,该算法是可行和有效的。其同样适用于4参数及以上可靠性模型参数的求解,为可靠性模型的参数估计提供了一个有效的新方法。(3)基于边界强度过程理论,提出了一种新的具有封闭形式解的连续比例强度模型。讨论了模型的特性,推导了模型参数及诸如给定时刻的条件可靠度、期望故障数、累积平均故障间隔时间等重要可靠性指标的极大似然点估计的计算公式,利用Fisher信息矩阵法和delta法给出了模型参数及上述可靠性指标的区间估计,并基于Akaike信息准则(AIC)和拟合优度检验指标R,给出了模型评价准则。以数控机床现场故障数据为例对所提方法进行验证,结果表明本文所建模型优于最小维修模型,且利用本文所提方法可以获得数控机床可靠性指标封闭形式的解及置信区间,可以很好地应用于工程实际。(4)最小维修情况下,为描述经历早期故障和耗损故障且故障强度随工作时间增加而趋近于某一常数的数控机床故障过程,提出一种新的4参数非齐次泊松过程模型。讨论了模型的特性,给出了模型中各参数的物理意义,推导了计算模型参数及诸如最小故障强度、最小故障强度对应的故障时刻等重要可靠性指标点估计的计算公式;基于似然比检验理论,给出了两台数控机床可靠性评估的模型选择方法,并基于拟合优度检验指标R,给出了模型评价准则。最后对两台数控机床故障数据进行分析,结果表明,该模型可定量评估机床的早期故障期,适合描述具有边界浴盆形状故障趋势的数控机床故障过程,为优化设计可靠性试验,尽可能在机床企业内部排除早期故障提供了一定的理论依据。(5)基于DIC信息准则、BGR诊断原理、蒙特卡洛仿真误差及模型参数和可靠性指标后验估计的区间长度,提出了数控机床贝叶斯可靠性模型的综合评价方法。给出了不同先验下用于Gibbs抽样的幂律过程模型参数的后验分布,并利用马尔科夫链蒙特卡洛法获得了模型参数和可靠性指标的贝叶斯点估计和区间估计。最后,结合两个工程实例,分别分析了数控机床在处于早期故障阶段和耗损故障阶段时的贝叶斯可靠性,结果表明,幂律过程模型各项评价指标均优于Weibull分布模型,且在其形状参数β的先验分布为贝塔分布或伽马分布时,得到的分析结果更加接近实际情况。更多还原

【Abstract】 In traditional assessment methods of reliability,the influences of repair and other factors on the reliability of machine tools are usually not considered,and the Weibull distribution,exponential distribution and lognormal distribution are usually used to estimate the reliability of machine tools under the assumption of “as good as new”.In fact,numerical control(NC)machine tools is a typical repairable system in its service cycle,repair actions run through the whole process of use,and the influence of different repair actions on system reliability are different.Repair actions often can bring the system to one of the following states: “as bad as old”,or “better than old,but worse than new”,that is “minimal repair” and “imperfect repair” assumptions are more realistically,therefore,according to the specific characteristic of the fault and repair of NC machine tools in different service stage,the effect of repair actions should be considered during the course of modeling,thus estimating the reliability of NC machine tools accurately and effectively.Aim at the problems and difficulties in the assessment of reliability of NC machine tools when considering the effect of repair actions,the assessment framework of reliability for NC machine tools in different service stage is constructed by taking the theory of stochastic point process as the core of technology and taking the immune clone selection algorithm and Bayesian theory as the supplementary means,and the assessment technology of reliability of NC machine tools in early failures stage,deterioration stage and the whole service stage are mainly studied,which can provide a theoretical reference and support for making a reasonable maintenance decision and improving the reliability of NC machine tools.The main contents of this paper are as follows:(1)The whole in-service stage of NC machine tools is divided and the specific characteristic of the fault and repair in each stage are analyzed.Then the critical problems in reliability assessment of NC machine tools under different maintenance conditions and in different in-service stage are proposed.To solve these problems and difficulties,the assessment framework of reliability for NC machine tools in the service stage is constructed.(2)Based on improved power law process model I and II,the assessment method of reliability of NC machine tools experiencing degradation phenomena in minimal repair is given,and aiming at the solving problem of model parameters and reliability indices,the immune clone maximum likelihood estimation method is proposed for solving the reliability model parameters.Convert the parameter estimation into an optimization issue with constraints,Perform optimization through setting the minimum negative logarithmic likelihood function as the objective function with the clone selection process by simulating biological immune system to obtain the optimal maximum likelihood point estimation and interval estimation of model parameters and reliability indices.The example verification indicates that the algorithm is feasible and effective.This algorithm also applies to solve reliability model parameters with four or more parameters and provides an effective new method for parameters estimation of reliability model.(3)A new continuous proportional intensity model with closed-form solutions is proposed base on the theory of bounded intensity process.The characteristics of the model are discussed,the formula of point maximum likelihood estimators for model parameters,as well as the important reliability indices such as the conditional reliability at given time,the s-expected number of failures,and the cumulative mean time between failure are all derived,the interval estimators of model parameters and reliability indices are given by using the Fisher information matrix method and delta method,then the assessment criterion of model is provided based on Akaike information criterion and the index of goodness-of-fit test.An example of real failure data from NC machine tools is taken to prove the proposed method,the results show that the closed-form solutions and confidence intervals of reliability indices for NC machine tools are obtained and the method is found very well uses in practical applications.(4)In minimal repair,in order to describe the failure process of NC machine tools experiencing both early failures and deterioration phenomena,and operating so long that the intensity function approaches a finite asymptote as the system operating time grows,a new four-parameter non-homogeneous Poisson process model is proposed.The characteristics of the model are discussed,and the physical meaning of its parameters is given,calculating formula of point estimation for model parameters and reliability indices such as the minimum intensity value and the time of minimum intensity are all derived,then a method of model selection for reliability assessment of two NC machine tools is provided based on the likelihood ratio statistic,and the model evaluation criterion is given based on the index of goodness-of-fit test.Finally,the real failure data of two NC machine tools are analyzed by using the proposed method,the results show that this method can quantitatively evaluate the machine tools’ lasting time of early failure period,it is suitable for describing the failure process of NC machine tools with bounded and bathtub shaped failure trend,and can provide a theoretical basis for optimizing reliability test and eliminating as much early failures as possible before the NC machine tools leave the factory.(5)Based on deviance information criterion,Brooks-Gelman-Rubin diagnosis principle,Monte Carlo simulation error and the interval length of posterior estimate for model parameters and reliability indices,the comprehensive evaluation method for Bayesian reliability model of NC machine tools was proposed.The posterior distributions of power law process model used for Gibbs sampling under different priors were given,and the Bayesian point and interval estimate of model parameters and reliability indices were obtained by using Markov Chain Monte Carlo simulation.Finally,Bayesian reliability of NC machine tools in early failures stage and deterioration stage are analyzed respectively by using two real engineering examples,the results of each evaluation index show that the power law process model is better than Weibull model,and when the prior distribution of shape parameter β is Beta distribution or Gamma distribution,the obtained analysis result are more close to the practical situation.更多还原