考虑退化和测量不确定因素的射频晶振寿命预测技术研究
发布时间：2018-08-01 08:03
【摘要】：为了解决射频(RF)技术的发展带来的安全隐患,在现有有关射频电路的寿命预测相关研究相对缺乏的情况下,本文总结了国内外有关科研成果,以射频晶体振荡器为对象展开了寿命预测的研究工作。通过相关资料和射频晶振的特性分析,本文提取了相位噪声作为反映射频晶振健康状态的特征量,提出了两种解决思路:从相位噪声的幂律系数出发的贝叶斯框架预测方法和从加速度效应出发的晶振加速灵敏度预测方法,并分别结合减小不确定因素的方法来给出振荡器的剩余使用寿命。针对两种模型的特点,本文分别阐述了剩余寿命预测中存在的各项误差的来源和影响,实验结果分别以概率区间和点估计的形式呈现。主要研究内容分为三部分:1.射频晶体振荡器寿命特征量的提取。详细介绍了射频晶振的关键特性参量以及在加速振动条件下的变化,重点阐述了晶振加速效应原理和相位噪声幂律原理同它们的用于寿命预测现实意义。2.寿命预测方法。加速灵敏度预测方法采集了随时间退化的加速灵敏度值,转化为一组与初始时刻加速灵敏度比较的曲线相似度并通过基于M估计的最优剪枝极限学习机进行预测;相位噪声预测方法根据振荡器相位噪声的Lesson理论构建为不同频偏处的负幂成分线性和,再根据最小二乘估计找出这些负幂成分系数随时间退化的规律,将关键系数退化序列输入相近的退化模型,结合贝叶斯框架在最大似然原则下进行迭代,在线地更新每个监测时刻的模型参数以及下一时刻的预测值。3.不确定因素分析。本文对从数据采集到预测结果生成全部过程中可能产生的不确定性来源进行了介绍,并在两种寿命特征量预测方法中加以体现,作为减小不确定性的主要手段;用基于历史样本与现场样本相似性的思路来修正单一现场样本数据量不足造成的外推预测偶然误差,反映了预测这一概念本身的不确定性以及误差的传播。事实证明结合不确定因素分析的两种特征量进行射频电路寿命预测均得到了良好的效果,具有一定应用价值。
[Abstract]:In order to solve the potential safety problems caused by the development of RF (RF) technology, this paper summarizes the domestic and foreign scientific research achievements under the condition of the relative lack of existing research on life prediction of RF circuits. The research on life prediction of RF crystal oscillator is carried out. Based on the relevant data and the characteristic analysis of the RF crystal oscillator, the phase noise is extracted as the characteristic quantity to reflect the healthy state of the RF crystal oscillator. Two solutions are proposed: the Bayesian frame prediction method based on the power law coefficient of phase noise and the crystal oscillator acceleration sensitivity prediction method based on the acceleration effect. The remaining service life of the oscillator is given by combining with the method of reducing uncertainty. According to the characteristics of the two models, the source and influence of the errors in the residual life prediction are discussed respectively. The experimental results are presented in the form of probabilistic interval and point estimation respectively. The main research content is divided into three parts: 1. Extraction of life characteristic of RF crystal oscillator. The key characteristic parameters of RF crystal oscillator and its changes under accelerated vibration are introduced in detail. The principle of crystal oscillator acceleration effect and phase noise power law principle and their practical significance in life prediction are expounded in detail. Life prediction method. The accelerated sensitivity prediction method gathers the accelerated sensitivity values which degenerate with time, and transforms them into a set of curve similarity compared with the initial acceleration sensitivity, and forecasts them by the optimal pruning extreme learning machine based on M estimation. According to the Lesson theory of phase noise of the oscillator, the phase noise prediction method is constructed as the linear sum of the negative power components at different frequency deviations, and the law of these negative power component coefficients degenerating with time is found according to the least square estimation. The key coefficient degenerate sequence is input into the similar degenerate model and the Bayesian framework is used to iterate under the maximum likelihood principle to update the model parameters of each monitoring time and the prediction value of the next time. Analysis of uncertain factors. In this paper, the possible sources of uncertainty in the whole process from data acquisition to prediction result generation are introduced, which are reflected in two prediction methods of life characteristic quantity, as the main means to reduce uncertainty. Based on the similarity between the historical sample and the field sample, the extrapolation error caused by the shortage of the sample data is corrected, which reflects the uncertainty of the concept of prediction and the propagation of the error. It has been proved that the life prediction of RF circuits based on the two characteristic variables of uncertainty analysis has a good effect and has certain application value.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN752
本文编号：2156845
[Abstract]:In order to solve the potential safety problems caused by the development of RF (RF) technology, this paper summarizes the domestic and foreign scientific research achievements under the condition of the relative lack of existing research on life prediction of RF circuits. The research on life prediction of RF crystal oscillator is carried out. Based on the relevant data and the characteristic analysis of the RF crystal oscillator, the phase noise is extracted as the characteristic quantity to reflect the healthy state of the RF crystal oscillator. Two solutions are proposed: the Bayesian frame prediction method based on the power law coefficient of phase noise and the crystal oscillator acceleration sensitivity prediction method based on the acceleration effect. The remaining service life of the oscillator is given by combining with the method of reducing uncertainty. According to the characteristics of the two models, the source and influence of the errors in the residual life prediction are discussed respectively. The experimental results are presented in the form of probabilistic interval and point estimation respectively. The main research content is divided into three parts: 1. Extraction of life characteristic of RF crystal oscillator. The key characteristic parameters of RF crystal oscillator and its changes under accelerated vibration are introduced in detail. The principle of crystal oscillator acceleration effect and phase noise power law principle and their practical significance in life prediction are expounded in detail. Life prediction method. The accelerated sensitivity prediction method gathers the accelerated sensitivity values which degenerate with time, and transforms them into a set of curve similarity compared with the initial acceleration sensitivity, and forecasts them by the optimal pruning extreme learning machine based on M estimation. According to the Lesson theory of phase noise of the oscillator, the phase noise prediction method is constructed as the linear sum of the negative power components at different frequency deviations, and the law of these negative power component coefficients degenerating with time is found according to the least square estimation. The key coefficient degenerate sequence is input into the similar degenerate model and the Bayesian framework is used to iterate under the maximum likelihood principle to update the model parameters of each monitoring time and the prediction value of the next time. Analysis of uncertain factors. In this paper, the possible sources of uncertainty in the whole process from data acquisition to prediction result generation are introduced, which are reflected in two prediction methods of life characteristic quantity, as the main means to reduce uncertainty. Based on the similarity between the historical sample and the field sample, the extrapolation error caused by the shortage of the sample data is corrected, which reflects the uncertainty of the concept of prediction and the propagation of the error. It has been proved that the life prediction of RF circuits based on the two characteristic variables of uncertainty analysis has a good effect and has certain application value.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN752
【参考文献】
相关期刊论文 前7条
1 罗阳;胡昌华;司小胜;张正新;刘豪;;融合随机退化过程与失效率建模的设备剩余寿命预测方法[J];电光与控制;2015年12期
2 司小胜;胡昌华;张琪;何华锋;周涛;;不确定退化测量数据下的剩余寿命估计[J];电子学报;2015年01期
3 冷国俊;陈睿;何著;保宏;张文X;;晶振加速度敏感性矢量计算及耦合机理研究[J];应用数学和力学;2014年S1期
4 何志昆;刘光斌;赵曦晶;王明昊;;高斯过程回归方法综述[J];控制与决策;2013年08期
5 纪龙蛰;单庆晓;唐钱;林明;;加速度对晶体振荡器的影响及补偿技术研究[J];宇航计测技术;2012年02期
6 王玉明;蔡金燕;;基于电子产品板级加速退化数据的可靠性分析[J];测试技术学报;2008年04期
7 张池平;刘宗尧;;一种改进的自适应模糊卡尔曼滤波算法[J];计算机工程与应用;2007年28期
相关博士学位论文 前3条
1 尤明懿;基于状态监测数据的产品寿命预测与预测维护规划方法研究[D];上海交通大学;2012年
2 杨成林;模拟故障字典技术测点选择问题研究[D];电子科技大学;2011年
3 刘震;智能BIT诊断方法研究及其在多电飞机电源系统中的应用[D];西北工业大学;2007年
相关硕士学位论文 前3条
1 曾现萍;含有奇异信号的电子系统性能退化趋势预测方法研究[D];电子科技大学;2016年
2 李志博;射频电路多参数健康评估技术研究[D];电子科技大学;2015年
3 宋辰亮;电子系统可靠性与剩余寿命的实时预测算法设计与实现[D];电子科技大学;2014年
，本文编号：2156845
本文链接：https://www.wllwen.com/kejilunwen/dianzigongchenglunwen/2156845.html
