由于對生產(chǎn)質(zhì)量體系性能和經(jīng)濟(jì)運(yùn)行的要求越來越高,現(xiàn)代工業(yè)體系日益龐大,復(fù)雜性也越來越高。為了解決這些問題,數(shù)據(jù)驅(qū)動技術(shù),如主成分分析（PCA）、偏最小二乘（PLS）和典型相關(guān)分析（CCA）用于系統(tǒng)故障診斷和過程監(jiān)控。它們假設(shè)要研究的數(shù)據(jù)不是自相關(guān)的。然而,大多數(shù)大型化學(xué)工業(yè)工廠本質(zhì)上都是非線性的,所以這些技術(shù)不能試用于它們,本質(zhì)上是無效的。為了彌補(bǔ)這個缺陷,需要開發(fā)一種能夠管理這些過程異常的算法。工業(yè)產(chǎn)品的需求正在迅速增長,因此提出了不同的適應(yīng)性技術(shù)。典型相關(guān)分析（CCA）是多元數(shù)據(jù)驅(qū)動的方法,它將同時考慮輸入輸出過程數(shù)據(jù)。本文討論了數(shù)據(jù)驅(qū)動技術(shù)的實(shí)現(xiàn),如用于田納西伊士曼（TE）過程監(jiān)控的主成分分析（PCA）偏最小二乘（PLS）和典型相關(guān)分析。主成分分析（PCA）是用于檢測和診斷化學(xué)過程中的故障的最常用的降維技術(shù)。盡管PCA在故障檢測方面具有一定的最優(yōu)性,并且已被廣泛應(yīng)用于故障診斷,但它不是最適合的用于故障診斷。與PCA和PLS相比,典型相關(guān)分析（CCA）已被證明可改善化學(xué)過程中的故障診斷。使用T²統(tǒng)計和Q統(tǒng)計（SPE）同時檢測多個故障。在這項(xiàng)工作中,我們比較了這... 

【文章來源】：哈爾濱工業(yè)大學(xué)黑龍江省 211工程院校 985工程院校

【文章頁數(shù)】：68 頁

【學(xué)位級別】：碩士

【文章目錄】：
摘要
Abstract
Chapter 1 Introduction
    1.1 Research background
    1.2 Source of the project
    1.3 Literature review
        1.3.1 Principal component analysis
        1.3.2 Partial Least Square
        1.3.3 Canonical Correlation Analysis
    1.4 Literature analysis
    1.5 Main content of research
        1.5.1 Introduction
        1.5.2 The canonical correlation Analysis
        1.5.3 Comparison techniques with CCA
        1.5.4 Industrial Benchmark
        1.5.5 Results
    1.6 Outline of thesis
    1.7 Concluding Remarks
Chapter 2 Basics of Fault Diagnosis
    2.1 Data-Driven based fault diagnosis
        2.1.1 Signal processing fault diagnosis
        2.1.2 Hardware redundancy fault diagnosis
        2.1.3 Plausibility test
        2.1.4 Model-based fault diagnosis
    2.2 Information infrastructure
        2.2.1 Data acquisition
        2.2.2 Networking
        2.2.3 High computation power availability
        2.2.4 The storage capacity
        2.2.5 Fundamentals for data-driven techniques requirement
    2.3 Process monitoring procedures
    2.4 Process monitoring measures
        2.4.1 Data-driven
        2.4.2 Analytical approach
        2.4.3 Knowledge base
    2.5 Process monitoring methods
        2.5.1 Parameter estimation
        2.5.2 Observers
        2.5.3 Parity relations
    2.6 Description of technical systems
    2.7 Basic principle of fault detection
    2.8 Basic statistical fault detection methods
    2.9 Test statistics
        2.9.1 Hotelling's T statistics
        2.9.2 Q statistics
    2.10 Concluding remarks
Chapter 3 Data-Driven Techniques
    3.1 Multivariate Statistics
        3.1.1 Data Pretreatment
        3.1.2 Outliers
    3.2 Univariate process monitoring
    3.3 Principal component analysis
        3.3.1 PCA based fault detection
        3.3.2 Fault Detection with PCA
    3.4 Partial Least Square
        3.4.1 Introduction
        3.4.2 Partial Least Square based fault diagnosis:
        3.4.3 Partial Least Square Algorithms
        3.4.4 PLS Fault detection algorithm
    3.5 Canonical Correlation Analysis
        3.5.1 Introduction
        3.5.2 The Basis of CCA Technique
        3.5.3 CCA based FD method
    3.6 Performance Evaluation
        3.6.1 False Alarm Rate
        3.6.2 Fault Detection Rate
    3.7 Numerical Study Example
    3.8 Concluding remarks
Chapter 4 Benchmark Case
    4.1 Case Study on TE Benchmark Process
    4.2 Background
    4.3 Process flow sheet
    4.4 Main Process Variables
    4.5 Simulated Faults in TEP
    4.6 Results and discussion
        4.6.1 Application to TEP
    4.7 Concluding remarks
Conclusions and Future work
References
Acknowledgement



本文編號：2917376