For lack of research on degradation test（DT） for molecular sieve oxygen concentrator（MSOC） and clear account of the factors causing degradation failure of molecular sieve bed（MSB）, a DT design method of MSOC based on orthogonal test was proposed. Based on the working principle of MSOC, the main factors causing degradation failure of MSB were determined, and a DT system of the MSOC was built. The optimal test height for the DT was 9 km, as was determined and verified by several controlled tests. A DT scheme of MSOC was designed based on the orthogonal test idea. The influence of various factors could be obtained through 9 typical tests, which reduced the test times and costs, and improved the efficiency of coupling tests. The test design can simulate the real working environment, and can effectively reduce the impact of test factors on the initial value of the test results.

针对机载燃油泵性能退化过程呈现的多阶段、非线性的特点以及对寿命预测实时性的要求,提出了一种基于失效物理与数据驱动融合的燃油泵在线退化建模与寿命预测方法.通过开关卡尔曼滤波器对燃油泵退化阶段进行在线识别,并对快速退化阶段建立失效物理与数据驱动融合的退化模型,然后基于无迹卡尔曼滤波器对建立的退化模型不断进行模型参数更新,并使用更新后的模型对失效寿命进行预测.将所提方法分别与纯数据驱动的方法、不进行退化阶段识别以及不进行参数更新的融合方法进行比较,整个参数更新过程中其均方根误差不超过0.3,寿命预测百分比误差不超过2％,均小于对比方法,验证了本文方法的有效性与优越性.

针对机载燃油泵故障诊断过程中存在需要先验知识、专家经验、特征解释及信号特征提取困难等问题,提出基于模拟退火遗传优化CNN-SVM的诊断方法.该方法利用CNN具有自动提取故障特征的特性实现特征快速提取.并针对传统CNN诊断方法所存在网络结构不确定、计算效率低等问题,采用模拟退火遗传算法对其模型结构和参数进行优化,使用SVM取代全连接层的Softmax分类器优化CNN分类效果,最后应用t分布随机近邻嵌入使其故障特征学习过程可视化,评估其特征提取能力.试验结果表明:与SA-GA-CNN、1DCNN、ANN、GA-SVM及GA-BP诊断方法相比,基于改进CNN-SVM的方法能更有效实现机载燃油泵的故障诊断.

针对机载燃油泵可靠性高、寿命长,实际工作环境复杂,而试验应力因素单一等特点,设计搭建了复杂应力条件下燃油泵退化试验平台,并对应力因素影响进行了分析.首先通过对燃油泵失效机理的分析,选取了影响轴承磨损的电应力和机械振动作为主要应力,开展其性能退化研究;然后基于以上选取的应力搭建了燃油泵退化试验平台和振动试验装置,并对压力传感器和流量传感器进行了选型,介绍了信号采集控制系统,设计了燃油泵夹持装置并对其动态性能进行了分析;最后基于正交试验思想设计了试验方案,并采用极差分析和方差分析2种方法对试验结果进行了分析.所提方法缩减了2/3的试验次数,节省了试验时间,并得出电压对燃油泵可靠性的影响更为显著,其置信度可达99％.

针对实际复杂系统诊断与测试过程中普遍存在的不确定性问题,提出测试不可靠条件下基于人工免疫克隆选择算法(artificial immune clone selection algorithm,AICS)的测试点优化选择方法.通过综合考虑故障检测率、隔离率、虚警率以及测试总费用等性能指标,构造了反映测试点集性能的适应度函数,并设计了基于AICS的不可靠测试点优化方案,有效地降低了算法复杂度,时间开销缩减到0.496 s,提高了运行效率.最后用燃油耗量测量系统的耗量组件进行实例验证,结果表明该方法能够获得在满足故障检测率、隔离率、虚警率等性能指标要求下,使得测试总费用最少的测试点集合,并且其综合性能指标优于遗传算法和模拟退火粒子群算法.

As an important part of radar receiver, radar gain control module determines whether the radar system can work normally and reliably. Therefore, it is necessary to estimate its life reliably. Whereas, there are few designs for radar degradation tests and the traditional accelerated life test method is no longer suitable for long life analog integrated circuit. In this paper, an environment test system for multi-stress loading is built, and the power control board and the fixed board of the radar gain control module are designed and made based on the idea of accelerated degradation test. The test data are obtained through the temperature cycle accelerated degradation test, and the eigenvectors are extracted which can reflect the performance degradation trajectory of the radar gain control module. The extreme learning machine combining a kernel based on grey wolf optimization is used to fit the degradation trajectory, and it is compared with the grid search method, the extreme learning machine with hidden layers and several other regression methods. The results show that, the accelerated degradation test and the extreme learning machine combining a kernel based on grey wolf optimization proposed in this paper can satisfy the performance evaluation of high reliability electronic products and have high precision and great application value.

The chip is a core functional component. Its reliability plays a vital role in electronic equipment normal operation. As the typical cause for chip malfunction, the solder joint degradation is selected to study chip reliability. The degradation models of solder joint in different failure modes are established through data-driven and failure physical model, and chip reliability model is constructed based on mutually competing failures of multiple solder joints. First, the chip reliability degradation test and finite element modeling(FEM) are carried out under coupled environment stress. The solder joint failure modes and degradation processes are studied through the analysis of test data, microstructure and mechanical simulation. Then, solder joint degradation models are established based on Coffin-Manson and Paris functions that have been modified by a data-driven method. Taking the solder joint failure time of degradation model as the characteristic parameter of Weibull distribution, the solder joint reliability function is obtained. Finally, mutually dependent competing failure theory is cited to describe the correlation about solder joint reliability of different failure modes, then the chip reliability model is established. The parameter estimation is realized by the inference function for margins (IFM) method. From verification tests, results show models are highly consistent with the actual reliability, indicting our reliability modeling method achieves the transition from underlying solder joint level to integrated component level. The joint application of different models can make up for the deficiencies of the single model and obtain more accurate results. In addition, we determined that intermetallic compounds (IMC) are main source of model error.