电子系统可靠性与剩余寿命的实时预测算法设计与实现
发布时间：2018-10-08 09:40
【摘要】：目前,故障预测与健康管理(PHM)领域面临着不断提高的技术要求和不断增长的应用需求,对于PHM中的可靠性预测、性能退化趋势预测和剩余寿命预测的研究和探索也越来越受到重视。本论文以电子产品的可靠性、性能退化趋势和剩余寿命的实时预测为核心研究课题,在文中重点探索了基于贝叶斯方法和滑动窗口样本分组法的实时可靠性预测方法、基于可靠性试验数据的实时性能退化趋势预测方法、基于差异分析和相似性的实时剩余寿命预测方法,核心内容分为四个组成部分。第一部分将研究分析基于贝叶斯方法和滑动窗口样本分组法的可靠性实时预测方法。通过使用可靠性试验获取的历史退化数据和实时采集测量获取到的现场退化数据,基于贝叶斯方法,将现场退化数据融入与历史退化数据中,利用滑动窗口样本分组法,更新性能参数变量分布的时变参数,计算出伪失效寿命,由此进一步得到产品的可靠性的实时预测结果。这种方法适用于历史数据的数量有限但并不缺乏的情况下,可以在最大程度上利用到有限的现场数据信息,得到准确有效的实时可靠性预测信息。第二部分将研究分析基于可靠性试验退化数据的性能退化趋势实时预测方法。通过利用现场退化数据和可靠性试验退化数据之间的关系,运用差异分析理论,分别获得由现场退化数据以及融合了现场退化数据和可靠性试验退化数据的数据集运算得到的趋势预测结果,然后根据预测结果的曲线拟合误差计算两者的权值并进行数据融合,最终得到产品的性能退化趋势实时预测结果。这种方法相较于基于现场数据时间序列的性能退化趋势的预测方法,适用范围更广,可以提供更加准确、更加稳定的预测结果。第三部分将研究分析基于差异分析和相似性的剩余寿命实时预测方法。将可靠性试验退化数据分成若干组,运用差异分析理论,将每一组可靠性试验退化数据分别与现场退化数据进行比较分析,并得到若干个剩余寿命预测结果,根据每一组可靠性试验退化数据与现场退化数据之间的相似度分配权重值,将若干个剩余寿命预测结果融合成最终的剩余寿命实时预测结果。这种方法不需要针对退化数据进行数学建模,对退化数据的轨迹类型和统计分布特性没有依赖,对有着很强的适用性。同时又能弥补基于相似性的方法所存在的缺陷,能够进一步提升剩余寿命的预测效果。第四部分将展示一个利用VC6.0和MATCOM编程的软件,主要用于验证前三部分中提到的基于贝叶斯方法和滑动窗口样本分组法的可靠性实时预测方法、基于可靠性试验退化数据的性能退化趋势实时预测方法和基于差异分析和相似性的剩余寿命实时预测方法。
[Abstract]:At present, the field of fault prediction and health management (PHM) is faced with increasing technical requirements and increasing application requirements, for reliability prediction in PHM, More and more attention has been paid to the research and exploration of performance degradation trend prediction and residual life prediction. This paper focuses on the real-time prediction of reliability, performance degradation trend and residual life of electronic products. In this paper, the real-time reliability prediction method based on Bayesian method and sliding window sample grouping method is explored. The real-time performance degradation trend prediction method based on reliability test data and the real-time residual life prediction method based on difference analysis and similarity are divided into four parts. In the first part, the reliability real-time prediction method based on Bayesian method and sliding window sample grouping method is studied. By using the historical degradation data obtained from reliability test and the field degradation data obtained by real-time acquisition and measurement, based on Bayesian method, the field degradation data is integrated into the historical degradation data and the sliding window sample grouping method is used. The time-varying parameters of the distribution of performance parameters are updated, and the pseudo-failure life is calculated, and the real-time prediction results of the reliability of the products are obtained. This method can be used to obtain accurate and effective real-time reliability prediction information by using the limited field data to the maximum extent when the number of historical data is limited but not lacking. In the second part, the real-time prediction method of performance degradation trend based on reliability test degradation data is studied. By using the relationship between field degradation data and reliability test degradation data, the difference analysis theory is used. The trend prediction results obtained from field degradation data and data set operations that combine field degradation data and reliability test degradation data are obtained, respectively, Then according to the curve fitting error of the prediction results, the weights of the two are calculated and data fusion is carried out, and finally the real-time prediction results of the performance degradation trend of the products are obtained. Compared with the performance degradation trend prediction method based on field data time series, this method can provide more accurate and stable prediction results. In the third part, we analyze the real-time prediction method of residual life based on difference analysis and similarity. The degradation data of reliability test are divided into several groups, and each group of degradation data of reliability test is compared with field degradation data by using the theory of difference analysis, and a number of residual life prediction results are obtained. According to the similarity between each set of reliability test degradation data and the field degradation data, several residual life prediction results are fused into the final residual life real-time prediction results. This method does not need to do mathematical modeling for degenerate data and has no dependence on the trace type and statistical distribution characteristics of degraded data. It has strong applicability to the degenerate data. At the same time, it can make up for the defects of the similarity based method, and can further improve the prediction effect of residual life. The fourth part will show a software which is programmed by VC6.0 and MATCOM, which is mainly used to verify the reliability real-time prediction method based on Bayesian method and sliding window sample grouping method mentioned in the previous three parts. Performance degradation trend real-time prediction method based on reliability test degradation data and residual life real-time prediction method based on difference analysis and similarity.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TB114.3
本文编号：2256301
[Abstract]:At present, the field of fault prediction and health management (PHM) is faced with increasing technical requirements and increasing application requirements, for reliability prediction in PHM, More and more attention has been paid to the research and exploration of performance degradation trend prediction and residual life prediction. This paper focuses on the real-time prediction of reliability, performance degradation trend and residual life of electronic products. In this paper, the real-time reliability prediction method based on Bayesian method and sliding window sample grouping method is explored. The real-time performance degradation trend prediction method based on reliability test data and the real-time residual life prediction method based on difference analysis and similarity are divided into four parts. In the first part, the reliability real-time prediction method based on Bayesian method and sliding window sample grouping method is studied. By using the historical degradation data obtained from reliability test and the field degradation data obtained by real-time acquisition and measurement, based on Bayesian method, the field degradation data is integrated into the historical degradation data and the sliding window sample grouping method is used. The time-varying parameters of the distribution of performance parameters are updated, and the pseudo-failure life is calculated, and the real-time prediction results of the reliability of the products are obtained. This method can be used to obtain accurate and effective real-time reliability prediction information by using the limited field data to the maximum extent when the number of historical data is limited but not lacking. In the second part, the real-time prediction method of performance degradation trend based on reliability test degradation data is studied. By using the relationship between field degradation data and reliability test degradation data, the difference analysis theory is used. The trend prediction results obtained from field degradation data and data set operations that combine field degradation data and reliability test degradation data are obtained, respectively, Then according to the curve fitting error of the prediction results, the weights of the two are calculated and data fusion is carried out, and finally the real-time prediction results of the performance degradation trend of the products are obtained. Compared with the performance degradation trend prediction method based on field data time series, this method can provide more accurate and stable prediction results. In the third part, we analyze the real-time prediction method of residual life based on difference analysis and similarity. The degradation data of reliability test are divided into several groups, and each group of degradation data of reliability test is compared with field degradation data by using the theory of difference analysis, and a number of residual life prediction results are obtained. According to the similarity between each set of reliability test degradation data and the field degradation data, several residual life prediction results are fused into the final residual life real-time prediction results. This method does not need to do mathematical modeling for degenerate data and has no dependence on the trace type and statistical distribution characteristics of degraded data. It has strong applicability to the degenerate data. At the same time, it can make up for the defects of the similarity based method, and can further improve the prediction effect of residual life. The fourth part will show a software which is programmed by VC6.0 and MATCOM, which is mainly used to verify the reliability real-time prediction method based on Bayesian method and sliding window sample grouping method mentioned in the previous three parts. Performance degradation trend real-time prediction method based on reliability test degradation data and residual life real-time prediction method based on difference analysis and similarity.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TB114.3
【参考文献】
相关期刊论文 前1条
1 尤明懿;;一个拓展的基于相似性的剩余寿命预测框架[J];电子产品可靠性与环境试验;2012年03期
，本文编号：2256301
本文链接：https://www.wllwen.com/guanlilunwen/gongchengguanli/2256301.html
