【作者】 焦卫东；

【导师】 吴昭同； 杨世锡；

【作者基本信息】 浙江大学 ， 机械制造及其自动化， 2003， 博士

【摘要】 本文以国家自然科学基金项目“基于独立分量分析的旋转机械故障诊断新方法的研究”(编号：50205025)和浙江省自然科学基金项目“基于盲源分离的机械噪声故障诊断技术研究”(编号：5001004)为基础提出，论文题目为“基于独立分量分析的旋转机械故障诊断方法研究”。本文提出了一个基于源分离的故障诊断新架构，探索了独立分量分析(ICA)理论方法在旋转机械故障诊断中的可能应用。全文主要内容如下： 第一章：首先概述了故障诊断研究的多学科交叉发展历程，分析了现有故障诊断理论方法的不足之处。接着，总结了独立分量分析(ICA)理论方法及其应用的国内外研究现状，分析了ICA应用于机械故障诊断的可行性。由此，提出了一个基于源分离的故障诊断新架构。最后，给出了本论文的主要研究内容、技术路线及创新点。 第二章：首先给出ICA模型及其模型估计性质，并从参照函数和优化算法两方面对一般ICA算法进行了概略总结。接着，对后续将用到的重要ICA(BSS)算法，特别是基于神经网络的自适应算法做了详细介绍。另外，还给出了基于带通滤波的改进BSS算法。验证改进算法的实验部分，将在第三章加以详细介绍。 第三章：研究了基于BSS的机械振动、声源分离问题，包括前分析处理、BSS源分离及后分析处理等环节。其中，也包括改善ICA算法收敛性能的某些策略及解决ICA(BSS)自身不确定性问题的某些可能措施。由此，形成了一个机械源分离的整体解决方案。最后，通过实验对该方案的可行性和有效性进行了验证。 第四章：特征提取是故障诊断的关键所在。本章从模式识别角度，研究了ICA应用背景下机械状态特征提取问题，给出了几种新颖有效的ICA基特征提取策略，并对某些传统特征提取方法如小波等进行了适当改进。 第五章：在ICA特征提取基础上，研究了几种典型神经网络如基于误差反传的多层感知器、径向基函数网络和自组织映射网络，以及支持向量机分类器在机械故障模式识别与分类中的应用，并进行了大量的故障分类对比实验。 第六章：在Matlab5.3编程环境下，开发了两个应用软件：BSS基机械源盲分离软件和ICA基故障诊断软件，作为整个论文、特别是第三、第四和第五章相关研究成果的总结。理想地，BSS基干扰消除技术、ICA基特征提取及模式分类方法，可以与通用的数字信号采集器共同整合为一个实用的机械状态监测与故障诊断系统。但是，其中有许多问题需要解决。例如，系统的数据通讯、接口以及实时性要求等。本章最后，利用实验台实测数据及ICA基诊断软件开展了故障诊断综合实验。 第七章：对全文的研究内容进行了总结，并对未来可能的研究方向和内容做了初步探讨与展望。更多还原

【Abstract】 Based on the "Research on New Method for Rotating Machine Faults Diagnosis Based on Independent Component Analysis" (National Nature Science Fund Project, No: 50205025), and the "Research on Technique for Fault Diagnosis with Mechanical Noise Based on Blind Source Separation" (Nature Science Fund Project of Zhejiang Province, No: 5001004), this dissertation with full title "Research on Methods for Fault Diagnosis of Rotating Machines Based on Independent Component Analysis" was written. In this dissertation, a new frame for fault diagnosis based on source separation was proposed, some possible applications of independent component analysis to fault diagnosis of rotating machines were explored. The details were studied as follows:Chapter one recited briefly the intercrossed development of diagnostics with other knowledge firstly. By analysing the existing problems in diagnostics. Then, ICA theory and its worldwide applications were summarized, and its feasibility used for mechanical faults diagnosis was analyzed. Thus, a new frame for mechanical faults diagnosis based on source separation was proposed. At last, the centre, implementation and novelty of this dissertation were brought forth.Chapter two gave out ICA model and its custom implementation from the point of view of contrast function and optimisation algorithm firstly. Then, some important ICA (BSS) algorithms, especially the adaptive algorithms based on artificial neural network that would be used in this dissertation were described in details. Also, the improved BSS algorithm based on band-pass was introduced, which would be tested by some experiments in Chapter three.Chapter three explored the subject of source separation of mechanical vibration and acoustic observations by sensors based on BSS, including such units as forward-analysis, BSS and backward-analysis. In this subject, some strategies for improving ICA (BSS) algorithm were also proposed. Thus, a feasible and integrated solution to mechanical source separation was given out. At last, this scheme was verified by experiments.Chapter four explored feature extraction, as the key to fault diagnosis, on the background of ICA from the viewpoint of pattern recognition. Several novel and effective strategies for feature extraction based on ICA were proposed, and some traditional methods for feature extraction such as wavelet were improved.Chapter five explored the application of several typical ANN including BP, RBF and SOM, and SVM classifier to pattern recognition and classification of mechanical faults. And, a lot of experiments for faults classification were made to test these classifiers.Chapter six developed applicable software for fault diagnosis: BSS based software for mechanical sources separation and ICA based software for fault diagnosis. Ideally, BSS based technique for interference removal, ICA based strategy for feature extraction and some typical methods for pattern classification can be integrated as a practical software-hardware system for fault diagnosis of machines, along with traditional Analogy-Digital Converter (ADC). However, some related problems such as data communication, interface and real-time running, etc. must be solved. At last, an experiment with data from the real world was made by means of the ICA-based software.At last, in the seventh chapter, all of the work in this dissertation was summed up, and the futureresearches on applications of ICA (BSS) were prospected.更多还原