针对滚动轴承不同故障位置、不同损伤程度的振动加速度信号的智能分类,提出一种基于随机搜索与长短时记忆(long short term memory,LSTM)神经网络的滚动轴承故障状态识别算法。该算法直接利用原始数据作为非线性输入,避免因人工提取特征值造成的原始信息缺失;使用LSTM与深度神经网络的混合网络提高模型性能;引入随机搜索算法自动优化超参数得到最优的网络配置;使用不同量纲、不同来源、不同损伤结构的两类数据集对模型进行试验验证。试验结果表明,在两类单一数据集及随机混合数据集均可达到99.8%以上的诊断准确度,表明本算法具有较高的泛化能力和鲁棒性。与BP、支持向量机、粒子群算法最小二乘支持向量机、LSSVM、浅层LSTM等方法在同等试验条件下的诊断结果进行比较,本文算法具有更高的识别准确度。
The current research of reliability allocation of CNC lathes always treat CNC lathes as independent series systems. However, CNC lathes are complex systems in the actual situation. Failure correlation is rarely considered when reliabil?ity allocation is conducted. In this paper, drawbacks of reliability model based on failure independence assumption are illustrated, after which, reliability model of CNC lathes considering failure correlation of subsystems is established based on Copula theory, which is an improvement of traditional reliability model of series systems. As the failure time of CNC lathes often obeys Weibull or exponential distribution, Gumbel Copula is selected to build correlation model. After that, a reliability allocation method considering failure correlation is analyzed based on the model established before. Reliability goal is set first and then failure rates are allocated to subsystems according to the allocation vector through solving the correlation model. Reliability allocation is conducted for t = 1. A real case of a CNC lathe and a numerical case are presented together to illustrate the advantages of the reliability model established consider?ing failure correlation and the corresponding allocation method. It shows that the model accords to facts and real working condition more, and failure rates allocated to all the subsystems are increased to some extent. This research proposes a reliability allocation method which takes failure correlation among subsystems of CNC lathes into consid?eration, and costs for design and manufacture could be decreased.
In a structural system reliability analysis that lacks probabilistic information, calculating the numerical characteristics of the state functions, especially the first four moments of the state functions, is necessary. Based on that, the structural system reliability is analyzed with a fourth-order moment method. The reliability sensitivity is required to conduct the differential operation of the numerical characteristic functions. A reliability sensitivity analysis formula is then derived in combination with the relation of the differential operation. Based on the matrix theory and Kronecker algebra, this paper systematically derives a matrix expression of the first four moments of the state functions, and establishes the matrix relation between the first four moments of the state functions and those of the basic random variables. On this basis, a differential operation formula of the first four moments of the state functions is further derived against the first four moments of the basic random variables. The vector relation between the state functions and the multidimensional basic random variables is described by means of the matrix operation to extend the operation method. Finally, a concise and intuitive formula is obtained to explore the inherent essential relation between the numerical characteristics of the state functions and those of the basic random variables, leading to a universal equation for the two kinds of numerical characteristics.
