基于增量式最大方差展开的水下控制系统故障诊断

Fault diagnosis of the subsea control system based on incremental maximum variance unfolding

中文关键词  流形学习  最大方差展开  增量式学习  故障诊断  水下控制系统

英文关键词  manifold learning  maximum variance unfolding  incremental learning  fault diagnosis  subsea control system

中文摘要

      对于复杂工业系统的故障诊断，由于非线性的存在，使得利用核函数的多元统计方法存在因核函数选择不同导致诊断结果不同的问题。本文采用最大方差展开的方法，作为一种流行学习方法，该方法在处理非线性数据时通过学习确定核矩阵，因而无需人为选择核函数。针对该方法难以对新增数据进行处理，本文提出了最大方差展开的增量式改进方法，利用正常样本进行学习建模，对检测样本通过增量的方式降维构造出低维空间，在该空间中构造监控统计量来完成故障的检测。最后，本文将该方法应用在水下控制系统的故障诊断中，通过仿真分析验证了该方法应用的有效性。

英文摘要

      For the fault diagnosis of complex industrial systems, because of the existence of nonlinearity, the multivariate statistical method using kernel function has the problem that the diagnosis results are different due to the different selection of kernel function. In this paper, a manifold learning method called the maximum variance unfolding is used which can find the kernel matrix for non-linear data by self learning, so it does not need to choose the kernel function artificially. But this method has difficulty processing the new data, this paper proposes an incremental improvement method of maximum variance unfolding. The normal samples are used for modeling, and then the low-dimensional space is constructed by incremental method for dimension reduction of the detected samples, in which the monitoring statistics are utilized to complete the fault detection. Finally, this method is applied to the fault diagnosis of subsea control system, and the feasibility of this method is verified by simulation analysis.