黃孝平臺灣大學：化學工程學研究所李成致Li, Cheng-ChihCheng-ChihLi2007-11-262018-06-282007-11-262018-06-282006http://ntur.lib.ntu.edu.tw//handle/246246/52039本論文主要注重在發展出新的模式方法對於加成性 (additive) 以及乘積性(multiplicative) 的錯誤進行全盤式的診斷。這些建立的錯誤診斷模型具備有相當多樣化的診斷能力，從簡單的錯誤偵測 (detection) 乃至於到細部的錯誤孤立 (isolation) 以及錯誤大小的識別 (identification)。 在一開始，論文中會回顧一些常用於錯誤診斷的多變量統計方法，並且利用一個簡單的範例來說明此類方法在錯誤孤立先天上的限制。然後對於這些常用的統計方法提出一些修正，以便將此類傳統的技術能夠延伸至監控與時間相關 (time-dependent) 的程序，或者能夠在開環的情況之下，正確地孤立出一些事先指定 (specified) 的錯誤型態。 然後，文中將會介紹一些基本的靜態與動態程序模型識別的方法。基於部分最小平方法 (PLS)，本文將提出一個藉由合併部分最小平方法的子模型 (sub-models) 而得到程序的整體模型 (global model) 的效率模型識別方法，並且利用一個簡單數值的範例來說明此方法的實用性。藉由程序的子模型以及合併後的整體模型，文中亦提出一個另類型態的分散式 (decentralized) 錯誤診斷方案，用以孤立可能的錯誤原因。另外，在模式識別之中，對於動態模式參數的變異以及共變結構的估計，亦會作出解析式的推導。 另外，一項新型與全域型的感測器 (sensor) 錯誤診斷方法亦在此論文被提出，此方法可用來診斷任意多維的感測器故障。基於這個提出的感測器錯誤診斷方法，有錯誤的感測器可以被輕易地偵測，孤立，而且錯誤的大小亦可被識別。 基於之前所推導的動態程序參數之變異數，論文中會定義一系列的模式參數的相似度 (similarity)。乘積性的程序錯誤可以利用這些新定義的相似度來偵測與孤立。對於一些特定種類的乘積性錯誤，例如程序增益 (gain) 錯誤以及程序時延 (deadtime) 錯誤，其錯誤的大小可以利用這些相似度來識別。 文中將會利用一些說明用的範例研究來展示上述理論概念的可行性。The focus of this dissertation is on developing novel model-based approaches for additive and multiplicative fault diagnosis (FD). The identified process diagnostic models can be extended to have varying fault diagnostic capabilities, from simple fault detection to detailed fault isolation and identification. Some frequently used multivariate statistical methodologies for FD are reviewed, and their major limitations in fault isolation are demonstrated. Novel modifications of conventional statistical techniques are proposed, and extended to monitor time dependent processes and to isolate some specified faulty types under open-loop conditions. Some basic static and dynamic process model identification approaches are reviewed. An efficient model identification method by merging PLSR sub-models is presented; and a numerical application is used to illustrate the practicality of this method. An alternative decentralized FD scheme is also proposed based on the sub-models and merged global model. Moreover, the parameter variance and covariance structures are investigated analytically for dynamic process representation. A novel and unified sensor FD approach is constructed to arbitrary multiple sensor failure scenarios. Based on the proposed methodology, the faulty sensors can be easily detected, isolated and identified. A variety of parameter similarities for dynamic processes are defined based on the derived parameter variances. With the use of these similarities, the multiplicative faults of processes can be detected and isolated. For some multiplicative faults, e.g. changes in gain and deadtime, the faulty parameter can be specified, and the fault magnitude can be identified. Illustrative case studies are included to demonstrate these theoretical ideas in this thesis.ACKNOWLEDGEMENTS I ABSTRACT III 摘要 V CONTENTS VII LIST OF TABLES XIII LIST OF FIGURES XV ABBREVIATIONS AND ACRONYMS XVII 1 INTRODUCTION 1 1.1 FAULT DIAGNOSIS (FD) 1 1.2 BASIC TASKS OF FD 2 1.2.1 Detection 2 1.2.2 Isolation 3 1.2.3 Identification 3 1.2.4 Recovery 3 1.3 NECESSITIES FOR FD 4 1.3.1 Safety 4 1.3.2 Performance 4 1.3.3 Cost 5 1.4 METHODOLOGIES FOR FD 5 1.4.1 Diagnostic model-based approach 6 1.4.2 Statistical model-free approach 6 1.4.3 Knowledge-based approach 7 1.5 CLASSIFICATION OF FAULTS 8 1.5.1 Additive faults 8 1.5.2 Multiplicative faults 9 1.6 DISSERTATION ORGANIZATION 11 2 MULTIVARIATE STATISTICAL METHODS FOR FD 13 2.1 OVERVIEW 13 2.2 MULTIVARIATE STATISTICAL METHODS 14 2.2.1 Principal components analysis (PCA) 14 2.2.2 Dynamic PCA (DPCA) 15 2.2.3 Fisher discriminant analysis (FDA) 16 2.3 FAULT DETECTION AND ISOLATION 17 2.3.1 Monitoring statistics 17 2.3.2 Contribution plots 19 2.3.3 Illustrative example 21 2.4 ADAPTIVE PROCESS MONITORING 22 2.4.1 Time series analysis 23 2.4.1.1 Modeling 23 2.4.1.2 Forecasting 24 2.4.2 Dynamic modeling for scores 25 2.4.3 Predictive monitoring model 27 2.4.4 Process change detection 28 2.4.5 Illustrative examples 29 2.4.5.1 Example 1: Predictive monitoring 29 2.4.5.2 Example 2: Process change detection 33 2.4.6 Results and discussions 33 2.5 ISOLATION ENHANCED PCA FOR FD 34 2.5.1 Introduction 34 2.5.2 Filtered signals and their PCA and LPC 34 2.5.3 Isolation of fault type A (FT-A) 36 2.5.3.1 Isolation of fault A1 36 2.5.3.2 Isolation of fault A2 36 2.5.3.3 Isolation of fault A3 37 2.5.4 Isolation of fault type B (FT-B) 37 2.5.4.1 LPC of PCA with input decompositions 38 2.5.4.2 Isolation of fault B1 39 2.5.4.3 Isolation of fault B2 40 2.5.4.4 Isolation of fault B3 40 2.5.5 RPCA for LPC computation 41 2.5.5.1 RPCA algorithm 41 2.5.5.2 Computation of LPC 43 2.5.6 Illustrative example 43 2.5.7 Results and discussions 46 2.6 CONCLUSIONS 46 3 IDENTIFICATION OF STATIC AND DYNAMIC PROCESSES 47 3.1 OVERVIEW 47 3.2 STATIC PROCESS DESCRIPTION 48 3.3 STATIC MODEL IDENTIFICATION 48 3.3.1 Multiple linear regression (MLR) 49 3.3.2 Last principal components (LPC) method 50 3.3.3 Partial least squares regression (PLSR) 50 3.3.4 Goodness of fitting 53 3.4 MODEL MERGING USING PLSR 54 3.4.1 Introduction 54 3.4.2 Regression coefficient matrix for PLSR 54 3.4.3 Merging two sub-models into one global model 56 3.4.4 Multiblock merging procedure 60 3.4.5 Computational complexity and storage capacitance analysis 62 3.4.5.1 Computational complexity analysis 62 3.4.5.2 Storage capacitance analysis 64 3.4.6 Application 1: PLSR model merging 64 3.4.6.1 System descriptions and settings 64 3.4.6.2 Identification of sub-models using PLSR 65 3.4.6.3 Identification of global model using combined data set 66 3.4.6.4 Identification of global model using projected loadings 67 3.4.6.5 Results and discussions 68 3.4.7 Application 2: Decentralized FD using PLSR 68 3.4.7.1 Introduction 68 3.4.7.2 Proposed decentralized FD procedure 69 3.4.7.3 System descriptions and settings 71 3.4.7.4 Fault scenario 1 : raising temperature in cooling water 77 3.4.7.5 Fault scenario 2 : impurity in inlet flow 79 3.4.7.6 Results and discussions 81 3.5 DYNAMIC PROCESS DESCRIPTION 82 3.6 NONPARAMETRIC DYNAMIC MODEL IDENTIFICATION 82 3.6.1 Parameter estimation 82 3.6.2 Estimation of parameter variances 84 3.6.3 Estimation of deadtimes of processes 85 3.6.4 Features of this nonparametric model representation 86 3.6.5 Illustrative example 86 3.7 PARAMETRIC MODEL IDENTIFICATION 89 3.7.1 First order discrete transfer function 89 3.7.2 Second order discrete transfer function 89 3.7.3 Illustrative application 90 3.8 CONCLUSIONS 91 4 MULTIPLE SENSOR FAULT DIAGNOSIS 93 4.1 OVERVIEW 93 4.2 INTRODUCTION 94 4.3 ISF VECTORS AND BSFM 95 4.4 METHODS FOR CONSTRUCTING BSFM 98 4.4.1 Constructing BSFM by perturbing method 98 4.4.2 Constructing BSFM by analytical method 99 4.4.3 Constructing BSFM by hybrid method 103 4.5 FAULT DETECTION USING BSFM 104 4.6 SENSOR FAULT ISOLATION AND IDENTIFICATION USING BSFM 106 4.6.1 Isolation of multiple sensor faults 106 4.6.2 Identification of multiple sensor fault magnitudes 107 4.6.3 Analysis of sensitivity of fault isolation 108 4.7 ILLUSTRATIVE EXAMPLE 110 4.7.1 Process setup 110 4.7.2 Preliminary works 112 4.7.3 Fault detection, isolation and identification 113 4.8 CONCLUSIONS 119 5 MULTIPLE MULTIPLICATIVE FAULT DIAGNOSIS FOR DYNAMIC PROCESSES 121 5.1 OVERVIEW 121 5.2 INTRODUCTION 122 5.3 CONVENTIONAL SIMILARITY MEASURES FOR DATA SETS 124 5.3.1 PCA-based similarity measures 124 5.3.2 Distance-based similarity measure 125 5.3.3 Appraisal of conventional similarity measures 126 5.3.3.1 Illustration 1 126 5.3.3.2 Illustration 2 127 5.3.3.3 Results and discussions 128 5.4 DEFINITIONS OF PARAMETER SIMILARITY FOR STATIC PROCESSES 128 5.4.1 Hypothesis test of significance of parameters 128 5.4.2 Violating number and parameter similarity 129 5.5 DEFINITIONS OF SIMILARITIES FOR DYNAMIC PROCESSES 130 5.5.1 Overall similarity 131 5.5.2 Sub-model similarities 133 5.5.2.1 Similarity for detection of sub-model changes 133 5.5.2.2 Similarity for detection of deadtime changes 133 5.5.2.3 Similarity for detection of gain changes 135 5.6 EXTENSIONS TO ONLINE PROCESS FD 137 5.7 ILLUSTRATIVE EXAMPLES 138 5.7.1 Application 1: Offline process FD 139 5.7.2 Application 2: Online process FD 141 5.8 CONCLUSIONS 146 6 CONCLUSIONS 147 6.1 SUMMARY 147 6.2 CONTRIBUTIONS OF THIS DISSERTATION 148 6.3 RECOMMENDATIONS FOR FUTURE WORKS 149 REFERENCES 1515646651 bytesapplication/pdfen-US模型識別錯誤診斷感知器錯誤乘積性錯誤參數相似度Model identificationFault diagnosisSensor faultMultiplicative faultParameter similaritythesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/52039/1/ntu-95-D89524006-1.pdf