Accelerated Life Testing (ALT) is an important technique to assess the reliability and lifetime of components with long life and high reliability. To some expensive products, it is difficult to supply a large of samples and long test time to carry out the life testing, so this paper studies accelerated model and its statistical method based on hybrid Weibull distribution under variable synthetic stresses. Based on the cumulative exposure model aforementioned, the parameters of accelerated model can be estimated with ALT data and its accelerated stress profile. However, the development of multi-stresses ALT under variable stress profile leads to a bottleneck problem on parameter estimation when evaluated parameters are greater than three. Direct toward the optimum difficulties with multi-parameters in Maximum Likelihood Estimation (MLE), this paper presents an Improved Float Genetic Algorithm (IFGA), which adopts the fuzzy algorithm in the float genetic algorithm to approach the global optimum solution. Application of hydraulic pump indicates that this method can reduce test time and test samples greatly.

Hydraulic actuation system is typical displacement control system whose performance is subjected to random inputs from noise, random disturbance from pump vibration or parameter changes. Through importing the integral item of residual error in different rank, this paper establishes the Proportion-Integral Observer (PIO) that can eliminate the influence of random input, parameter excursion and inherent disturbance and realize the fault diagnosis robustly. In order to reach the high fault coverage and low false alarm, this paper provides twin-threshold and corresponding fault decision-making strategy that not only considers the inherent noise existed in the system but also can manifest the fault information effectively. Application and simulation of hydraulic actuator system indicates that the adaptive fault diagnosis can detect the failure in high precision at the stochastic condition.

提出了以系统脉冲响应序列作为闭环系统模型的建模方法，并且给出了这种建模方法的计算方法以及模型输出的计算方法。这种闭环系统建模方法保留了脉冲响应序列中的高频成分，避免了从脉冲响应序列向传递晒数变换过程中丢失脉冲响应函数高频信息的缺点；同时，也避免了神经网络每次学习结果获得的网络模型不确定的缺点。采用该方法建立的闭环系统模型精度非常高并且可以应用在故障诊断的系统中。文章最后以数值仿真和实际带随机干扰的系统为例，证明该方法是有效的。

在复杂系统可用度分析中，采用深度和广度遍历算法将复杂动态故障树分解为独立子树，推导和证明了独立子树等效时变可用度参数模型，建立了各独立子树同构的马尔可夫状态转移链。通过求解降维微分方程获得故障子树的时变可用度参数，采用由下至上的递归处理算法快速实现复杂系统的可用性定量分析，既解决了复杂系统可用度计算的“组合爆炸”问题，又通过考虑时变参数保证了可用度解算的精度。实例分析表明，采用独立子树等效时变可用度分析方法在有效的分解下，其误差可以很小。

冷储备系统的动态结构变化使得系统可用度分析必须基于瞬态求解，针对频域两部件冷储备系统马尔可夫更新方程解算困难的问题，提出基于时域伏尔特拉方程的迭代算法，证明了该迭代方程的唯一性和收敛性，并培出选代算法的误差分析。理论和实例分析表明，当冷储备系统部件的工作寿命和维修时间分布是有界函数时，利用迭代算法可以容易实现时域马尔可夫更新方程的求解。