非完整数据过程的鲁棒故障检测与故障认知方法

项目名称： 非完整数据过程的鲁棒故障检测与故障认知方法

项目编号： No.61273161

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 杨煜普

作者单位： 上海交通大学

项目金额： 82万元

中文摘要： 数据的非完整性是复杂过程中非常常见的数据特性，该特性会使数据驱动的故障诊断技术表现出鲁棒性变差及误报漏报严重的现象。非负矩阵分解(NMF)是近年来机器学习领域中日益显现特色的数据变换方法，具有正向纯加性、稀疏性、可解释性并体现了一种深刻的人工智能认知机理。本申请针对过程数据的非完整性问题，研究基于NMF方法的鲁棒故障检测与故障认知方法，包括：1）研究一类嵌入变换矩阵NMF（TE-NMF）算法，解决NMF用于故障诊断时首先遇到的测量-特征空间的变换求取问题；2）研究NMF稀疏性强化方法，抑制非完整数据的缺失项影响，提高故障检测鲁棒性；3）研究NMF正向纯加性的增量逼近模式，可在数据结构不完备条件下提高故障检测鲁棒性；4）研究在一定正则性和特征鉴别性约束条件下，NMF基矩阵向量与模糊隶属度函数的等价关系，以及NMF方法在模糊逻辑空间下的多类故障认知方法；5）搭建理论仿真和实际验证平台。

中文关键词： 鲁棒故障诊断；非负矩阵分解；统计特征认知；非完整数据；主动机器学习

英文摘要： Data-driven fault diagnosis methods have become the current hot research topic in the field of process control. One of the key characteristic of complex industrial processes is the incomplete data issue.This characteristic will reduce the robustness and accuracy of the fault diagnosis meodel. Non-negative Matrix Factorization (NMF) is a new dimension reduction technique which can preserve spatial relationships corresponding and retain the intrinsic structure of original data. The non-negativity constraint allows only additions in the synthesis and no cancellations or interference of patterns via subtraction or negative feature vector values. This makes NMF have sparsity,interpretability and learn localized part-based representations. NMF provides a new method to solve the incomplete data issue for fault diagnosis. Therefore, NMF is very potential for the research of fault diagnosis methods. This project has planned to propose a new fault diagnosis method based on NMF algorithms and to research several key problems of NMF-based method. The main content of the work is following:1) In order to obtain the transformation matrix betweem the original data space and the lower dimensional space, this project will propose a new method named transformation embeded NMF (TE-NMF) algorithm for fault diagnosis; 2) this projec

英文关键词： Robust fault diagnosis；Nonnegative matrix factorization；Statistical feature recognition；Incomplete data；Active machine learning