节点文献

基于支持向量机的动调陀螺仪寿命预测方法研究

Research on Life Forecasting Methods of a DTG Based on Support Vector Machine

【作者】 徐国平

【导师】 田蔚风;

【作者基本信息】 上海交通大学 , 精密仪器及机械, 2008, 博士

【摘要】 随着我国航天技术的发展,航天器越来越向着高可靠、长寿命、高有效性方向发展。作为这些航天器中必不可少的姿态控制执行机构和姿态测量部件——飞轮、陀螺仪等旋转机电部件,由于其成本高、批量小等特点,如何评估这些产品及其部件的可靠性寿命成为一个迫切需要解决的难题。动调陀螺仪作为一种被广泛应用于航天器导航与制导系统中的中高精度陀螺仪,对其进行可靠性寿命预测方法研究具有十分重要的现实意义和理论研究价值。本文根据学校实验室现有的试验条件,在只有单个动调陀螺仪试验子样情况下,给出了一个在极小子样条件下的动调陀螺仪寿命试验和寿命预测方法研究思路、方案,即以支持向量机作为理论基础,采用参数外推法进行动调陀螺仪寿命评估与预测方法研究。通过对动调陀螺仪在极小子样条件下进行1:1完全寿命试验,测试与分析影响动调陀螺仪寿命的性能参数,研究并提取能够表征陀螺可靠性和寿命的关键因子,通过对其的分析与处理建立寿命预测模型,进而对单个动调陀螺仪的寿命进行评估和预测研究。本文主要完成了如下几方面的工作:第一,结合动调陀螺仪的结构特点及试验条件,采用一种新的寿命评估思路(即性能参数外推法)对陀螺仪寿命进行评估与预测研究,并给出了整个动调陀螺仪寿命预测研究方案。同时,根据此方案及陀螺仪自身的特点,确定陀螺振动、漂移、工作和环境温度、陀螺供电电压、功率等作为监测的性能参数,通过相应的测试电路、软件及界面等设计,建立动调陀螺仪寿命测试系统,完成其状态监测与数据采集。第二,由于测试系统及动调陀螺仪自身的结构温度特性,文中结合小波分析优越的多尺度(多分辨率)分析特性,提出了一种基于小波变换和支持向量机的温度建模与补偿模型,对陀螺各参数的长期测试数据进行温度建模补偿与预处理。另外,根据陀螺仪自身的性能状态变化,采用基于支持向量机的启发式搜索策略,以陀螺参数特征集的交叉验证错误率为评价指标进行自学习识别与提取分析,最终选取陀螺振动作为关键特征因子对陀螺仪性能进行评估和预测研究。第三,根据陀螺振动信号的非平稳特性,采用经验模态分解(EMD)的信号分析方法对陀螺振动进行分析。针对EMD分析方法自身的缺陷,文中提出了一种改进EMD方法的混合信号分析策略,并利用其对动调陀螺仪振动信号进行频域能量分析。同时,根据陀螺长期测试期间自身性能的变化,采用基于支持向量机的权重贡献分配方法对陀螺频域能量特征进行自学习提取,并以此建立反映动调陀螺仪性能状态渐进变化趋势的陀螺振动频率能量。第四,结合支持向量机预测及灰色数据生成操作的优势,提出了一种新的灰色支持向量机预测模型(即AGO-SVM预测模型),并利用其对实际陀螺振动能量数据进行建模预测比较研究。预测结果表明所提出的灰色支持向量机模型预测精度优于单一灰色预测和SVM预测方法,具有更好的建模预测性能。另外,根据提取的陀螺寿命指标——振动特征能量趋势,利用灰色支持向量机预测模型对动调陀螺仪进行寿命评估与预测研究。同时,通过对单个动调陀螺仪历史能量数据进行多次分段建模预测统计分析,从而建立多步预测误差模型,以此来评估动调陀螺仪寿命预测的误差及可信度。本论文的研究工作可为我国航天等领域机电旋转部件的寿命预测研究工作提供参考与借鉴。

【Abstract】 With the development of aerospace technology, aerospace products are evolving towards high reliability, long lifetime and great effectiveness. However, these necessary gesture controlling and measuring components, such as flywheel and gyroscope, are high cost and small batch, so how to estimate their reliability and lifetime has become an urgent challenge. As a result, this research on lifetime evaluation and forecasting of a high precision dynamically tuned gyroscope (DTG), which is widely applied in the navigation systems of these aerospace products, is significant.Due to the limitation of the lab’s testing conditions, this thesis presents a kind of research scheme for single gyro lifetime testing and forecasting based on the factor that only one DTG can be employed. That is, this scheme adopts trend modeling and extrapolation to estimate and forecast the DTG lifetime. In this scheme, the 1:1 testing system of the single DTG is first built and executed to collect the testing data of different performance parameters, and then the key factor is studied and selected from the different performance parameters based on the corresponding analysis methods. Through the analysis and processing of the key factor, the corresponding DTG lifetime forecasting model based on SVM is built and investigated, which is employed to evaluate and forecast the DTG lifetime. The main research contents of the thesis are listed as follows:Firstly, owing to the configuration characteristic of the DTG and the testing conditions, a new lifetime evaluation scheme, the trend forecasting and extrapolation, is given to evaluate and forecast the lifetime of the single DTG. According to the given scheme, the different performance parameters of the DTG, such as gyro vibration, gyro drift (X- and Y-axis), gyro temperature, environmental temperature, gyro volt and power, are selected as the testing parameters, and the related testing circuits are designed, respectively. By designing of system soft and interface with Visual C++, the corresponding DTG lifetime testing system is built to accomplish the state measurement and data collection.Secondly, due to the superior ability of multi-resolution analysis, wavelet transform (WT) is introduced into the SVM, and thus a novel WT-SVM temperature compensation model is proposed and applied in the temperature modeling and compensation to reduce the influence of temperature variation on the long-term testing data of the different performance parameters. The corresponding modeling and compensating results indicate that the proposed WT-SVM model is feasible and effective. In addition, according to the change of long-term performance state of the DTG, the SVM-based recursive feature elimination strategy and the evaluation criterion of cross-validating error rate are adopted to conduct the self-learning identification and gyro feature selection. Consequently, gyro vibration is selected as the key performance parameter for the DTG’s lifetime evaluating and forecasting investigation.Thirdly, according to the nonlinear and non-stationary characteristic of the gyro vibration, empirical mode decomposition (EMD) method is employed to analyze the gyro vibration signal. Aiming at the problems of intrinsic mode function (IMF) criterion in single EMD method, neural network (NN) prediction model and wavelet packet transform (WPT) technology are introduced into the EMD method, and an improved hybrid EMD-based analysis strategy is thus proposed and applied in the frequency-domain energy analysis of the gyro vibration. Simultaneously, owing to the change of long-term performance state of the DTG, the distribution approach of weight contribution based on SVM is employed to extract the frequency-domain energy features of the DTG, with which the energy trend of the gyro vibration denoting the gradual performance tendency of the DTG is built.Fourthly, combining the superiority of SVM forecasting and grey accumulated generating operation (AGO), a kind of new grey support vector machine (i.e. AGO-SVM model) forecasting model is proposed and applied in the forecasting analysis of gyro vibration energy. The modeling and forecasting results of the real energy data of the DTG show that the proposed strategy of the combination between AGO and SVM is effective and superior to the traditional grey model and single SVM method. In addition, according to the extracted gyro lifetime index (the vibration energy trend), the grey SVM model is exploited to forecast and analyze the residual lifetime of the DTG. Moreover, based on the statistical analysis of multiple subsection modeling and forecasting of the single DTG’s energy data, the multi-step forecasting error model of the grey SVM model is built, which can be used to estimate the forecasting error and reliability of the DTG lifetime.In summary, the research of the thesis may be used as reference for lifetime forecasting research of other electromechanical rotating components in our aerospace or other fields.

节点文献中: