基于退化数据的产品可靠性建模与剩余寿命预测方法研究

发布时间：2018-06-27 04:31

本文选题：退化数据 + 剩余寿命　；参考：《华中科技大学》2014年博士论文

【摘要】：高可靠长寿命产品广泛存在于航空航天、电子工业、军事等可靠性和安全性要求较高的领域,由于其成本高、批量小以及失效机理复杂等特点,如何评估这类产品的可靠性和剩余寿命成为一个亟待解决的难题。随着传感器和信息技术的发展,对于具有退化型失效的高可靠长寿命产品,可以通过监测其性能退化过程中的关键特征参数获得退化数据,进而利用退化数据来实现产品的可靠性建模和剩余寿命预测的目的。退化数据不仅能够提供丰富的寿命信息,弥补了高可靠长寿命产品可靠性信息量不足的问题,而且其变化往往能够反映动态环境对产品的影响。因此,开展基于退化数据的可靠性建模和剩余寿命预测技术研究具有十分重要的理论意义和实际应用价值。本文主要采用随机过程与稀疏贝叶斯学习理论展开研究,主要研究内容如下： (1)提出了一种基于期望最大化算法和Wiener过程的退化建模及寿命估计方法。该方法通过将Wiener过程中的漂移参数视为服从正态分布的随机变量来刻画一批产品个体之间的异质性,进而在首达时意义下给出了随机效应约束下对应的可靠度函数的表达式,然后采用期望最大化算法有效地解决了隐变量情况下的最大似然估计问题,最后通过数值仿真和在卫星用动量轮上的应用研究验证了所提出方法的有效性。 (2)针对具有线性单调退化过程的产品,提出了一种基于逆高斯过程的退化建模和剩余寿命预测方法。考虑到实时监测的退化数据不断变化的特点,设计了一种有效的基于Bayesian更新与期望最大化算法的参数估计方法,其中采用Bayesian方法来更新随机参数,进而利用期望最大化算法来估计模型中的非随机参数。该方法能够利用实时监测的退化数据通过不断更新来追踪产品的最新状态,并且可以获得剩余寿命分布函数的表达式,最后数值仿真和实例研究表明了该方法能够有效地对单个产品的退化过程进行建模,获得了较精确的剩余寿命估计结果。 (3)针对具有非线性退化过程的产品,从时间序列分析的角度提出了一种基于均熵和稀疏贝叶斯学习的剩余寿命预测方法。该方法首先利用小波阈值降噪的方法对监测的退化数据进行预处理,从而降低了噪声对预测结果的影响,其次为了实现正确的时间序列重构的目的,采用基于均熵的方法来确定最优的嵌入维,然后利用相关向量机预测未来的退化数据,通过结合失效阈值获得了剩余寿命的概率分布函数,有效地量化了剩余寿命预测的不确定性。最后通过在锂离子电池上的应用研究说明了所提出方法的有效性,取得了较精确的剩余寿命预测结果。
[Abstract]:High reliability and long life products are widely distributed in the areas of high reliability and safety requirements in aerospace, electronic industry and military. Due to its high cost, small batch and complex failure mechanism, it is an urgent problem to evaluate the reliability and residual life of such products. Development, for high reliability and long life products with degenerative failure, degenerate data can be obtained by monitoring the key characteristic parameters in the process of performance degradation, and then using degraded data to achieve the purpose of product reliability modeling and residual life prediction. Degenerated data can not only provide rich life information and make up for high availability. The reliability information of long life products is insufficient, and its changes often reflect the impact of dynamic environment on products. Therefore, it is of great theoretical significance and practical value to carry out the research on Reliability Modeling and residual life prediction technology based on degenerate data. The main contents of the study are as follows:
(1) a degenerate modeling and life estimation method based on the expectation maximization algorithm and the Wiener process is proposed. The method describes the heterogeneity of a batch of products by taking the drift parameters in the Wiener process as a random variable that obeys the normal distribution, and then gives the corresponding corresponding in the sense of random effect in the sense of the first arrival. By using the expression of the degree function, the maximum likelihood estimation problem in the case of hidden variables is effectively solved by the expectation maximization algorithm. Finally, the effectiveness of the proposed method is verified by numerical simulation and the application of the momentum wheel on the satellite.
(2) for products with linear monotone degradation process, a degenerate modeling and residual life prediction method based on inverse Gauss process is proposed. Considering the changing characteristics of real-time monitoring, an effective method of parameter estimation based on Bayesian updating and expectation maximization is designed, in which Bayesian square is used. The method is used to update the random parameters and then estimate the non random parameters in the model by using the expectation maximization algorithm. This method can track the latest state of the product by constantly updating the degraded data of the real-time monitoring, and can obtain the expression of the residual life distribution function. Finally, the numerical simulation and the case study show the method. It can effectively model the degradation process of a single product and obtain more accurate residual life estimation results.
(3) for products with nonlinear degradation process, a method of residual life prediction based on entropy and sparse Bayesian learning is proposed from the point of view of time series analysis. Firstly, the method of wavelet threshold denoising is used to preprocess the degraded data of monitoring. In order to realize the correct time series reconstruction, the optimal embedding dimension is determined by the entropy based method. Then the correlation vector machine is used to predict the future degradation data. The probability distribution function of residual life is obtained by combining the failure threshold, and the uncertainty of the residual life prediction is quantized effectively. Finally, the lithium ion is used in the lithium ion. The applied research on the battery shows the effectiveness of the proposed method and obtains a more accurate residual life prediction result.
【学位授予单位】：华中科技大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TB114.3;TP274

【相似文献】