【作者】 高晓康；

【导师】 齐从谦；

【作者基本信息】 同济大学 ， 机械制造及其自动化， 2007， 博士

【摘要】 故障诊断系统发展到今天，仍然面临着支撑理论的局限性、规则获取和更新困难、诊断模型更新的不同步等各种问题。为此，本文以粗糙集理论为支撑理论，提出了知识发现模型，并将其分别应用到工程和医学诊断领域，发现了若干有用的诊断知识。针对粗糙集理论核心计算环节的计算复杂性问题，本文提出一种适用于连续和离散条件属性并存的混合条件属性全距离(Mixed Attribute Whole Distance，简称MAWD)空间降维模型，并就MAWD降维方法对于粗糙集理论后续若干重要计算环节的计算复杂度进行了量化分析，提出了MAWD模型对于提高粗糙集核心计算效率的数学评判。针对故障诊断规则获取问题，本文在所研究的论域空间MAWD降维模型、变精度粗糙集模型基础上，结合先进的连续属性IMDV-SOM启发式自动聚类模型、决策矩阵计算模型。本文将四个核心模型进行有机的统一和集成，提出了一种简称为MMS-VPRS(MAWD-MDV-SOM-Variable Precision Rough Set)的知识发现模型，并实现了程序化，为故障诊断规则的发现提供了精准、高效的规则获取途径。本文以工程领域中的旋转机械故障诊断、注射成型质量控制、电子镇流器故障诊断等若干典型故障诊断问题为对象，开展MMS-VPRS知识发现模型的应用性研究，导出了若干有用的工程领域诊断规则。验证了MMS-VPRS知识发现模型的工程适用性。为了研究故障诊断模型的更新问题，本文将MMS-VPRS知识发现模型探索性地应用到医学诊断领域，利用该模型对51例可疑冠心病患者的信息进行全面分析，得出了较理想的诊断规则。经临床验证，该诊断规则有较高的敏感性与特异性，可以为早期诊断冠心病提供一种新的途径。更多还原

【Abstract】 Up to now, fault diagnosis system is still facing various problems such as the limitation of supporting theories, the difficulties in obtaining and renewing the rules, non-synchronization of renewing processes of the diagnostic models. Therefore in this paper, we use the rough set theory as basic supporting theory and put forward a knowledge finding model. This model is then applied to engineering and medical diagnostic areas respectively, to find some meaningful diagnostic knowledge.To simplify the complexity of calculation on rough set theory, we bring forward a spatial-dimensions-reducing model named as mixed attribute whole distance (MAWD). This model is suitable for analysis of the data where both the continuous and discrete attributes are coexisted. We find that MAWD model can be used quantitatively for dealing with the complexity of calculation processes of rough set theory and improve the calculation efficiency of rough set theory.To solve the problem of obtaining fault diagnosis rules, the MAWD model, the variable precision rough set (VPRS) model, the advanced continuous attribute IMDV-SOM heuristic automatic clustering model and the decision matrix calculation model were integrated in this paper and a new knowledge discovery model called MMS-VPRS (MAWD-MDV-SOM-Variable Precision Rough Set) was presented for the first time. This MMS-VPRS model was programmed to provide a concise, high efficiency way for the mining of fault diagnosis rules.To validate the applicability of the MMS-VPRS knowledge discovery model in engineering, and to find out useful diagnosis rules for practical fault diagnosis, we took rotating mechanical devices fault diagnosis, injection molding quality control, and electron ballast fault diagnosis as practical application examples. All these application studies had a good result and some useful diagnostic rules in engineering fields are discovered.For studying the updating of fault diagnosis models, the MMS-VPRS knowledge discovery model was introduced into medical diagnosis field. This model was used to analyze all-around clinical information in 51 patients suspected of having coronary heart disease. The obtained diagnostic results were well matched with that made by medical experts, with high sensitivity and specificity. Thus, this model could be helpful to diagnose coronary heart disease at its early stage.In conclusion, this paper puts forward a spatial-dimensions-reducing model called MAWD, which was then combined with several other models to create MMS-VPRS, a new knowledge discovery model. The validity of MMS-VPRS model has been confirmed in engineering field and in medical diagnostic process.更多还原