基于流形学习的航空发动机故障诊断技术研究

项目名称： 基于流形学习的航空发动机故障诊断技术研究

项目编号： No.51505492

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 机械、仪表工业

项目作者： 张赟

作者单位： 中国人民解放军海军航空大学

项目金额： 21万元

中文摘要： 航空发动机故障诊断中一个有挑战性的难题是如何处理具有高维数、非线性化特点的采样数据。针对这一问题，基于振动采样信号位于嵌入于高维信号空间中低维非线性流形上的假设，本项目利用流形学习理论，研究新型的航空发动机故障诊断技术，对需要解决的若干关键问题进行深入研究，主要包括：构建一种以保持数据邻域信息不变，并最大化子流形间距为目标的多流形学习模型，以实现发动机故障特征的有效提取；以加权局部主元分析为基础，结合贝叶斯估计理论，研究流形内蕴维数的鲁棒自适应优化估计方法；研究基于矩阵分解理论的流形学习增量式泛化算法，以解决新增故障的映射识别问题；研究一种基于局部流形几何信息的新型分类器，使之适合流形特征提取背景下的故障分类。以上研究内容将推动流形学习理论在航空发动机故障诊断领域中的应用，为如何对高维非线性振动采样数据进行故障特征提取与识别的难题提供有效的解决途径，提高发动机故障诊断的准确性。

中文关键词： 流形学习；故障诊断；航空发动机；非线性降维；振动信号

英文摘要： How to deal with the high-dimensional and nonlinear data is a challenging problem for aero-engine fault diagnosis. Following the intuition that the measured signal samples usually distribute on or near the nonlinear low-dimensional manifolds embedded in the high-dimensional signal space, a novel aero-engine fault diagnosis based on manifold learning is proposed. Several crucial problems are researched. A multiple manifold model which keeps the local data information and maximizes the distance of manifolds is proposed for fault feature extraction. The optional and robust estimation of reduced dimension is researched by local principal component analysis and Bayesian methods. The generalization of manifold learning based on maxtirx decomposition is proposed for mapping the new fault data. The new classifier based on local manifold geometry is researched for fault classification. The research could promote the application of manifold learning to aero-engine fault diagnosis, and provide an effective approach to sovle the fault recognition with high-dimensional and non-linear samples, which could improve the accuracy of aero-engine fault diagnosis.

英文关键词： manifold learning;fault diagnosis;aero-engine;nonlinear dimensionality reduction;vibration signal