Advanced methods for fault diagnosis and fault-

tolerant control Steven X. Ding download

https://textbookfull.com/product/advanced-methods-for-fault-
diagnosis-and-fault-tolerant-control-steven-x-ding/

Download more ebook from https://textbookfull.com

We believe these products will be a great fit for you. Click
the link to download now, or visit textbookfull.com
to discover even more!

Robust Integration of Model-Based Fault Estimation and

Fault-Tolerant Control Jianglin Lan

https://textbookfull.com/product/robust-integration-of-model-
based-fault-estimation-and-fault-tolerant-control-jianglin-lan/

Robust and Fault Tolerant Control Neural Network Based

Solutions Krzysztof Patan

https://textbookfull.com/product/robust-and-fault-tolerant-
control-neural-network-based-solutions-krzysztof-patan/

Adaptive and Fault-Tolerant Control of Underactuated

Nonlinear Systems 1st Edition Jiangshuai Huang

https://textbookfull.com/product/adaptive-and-fault-tolerant-
control-of-underactuated-nonlinear-systems-1st-edition-
jiangshuai-huang/

Intelligent Fault Diagnosis and Accommodation Control

1st Edition Sunan Huang (Author)

https://textbookfull.com/product/intelligent-fault-diagnosis-and-
accommodation-control-1st-edition-sunan-huang-author/
Advances in Gain-Scheduling and Fault Tolerant Control
Techniques 1st Edition Damiano Rotondo (Auth.)

https://textbookfull.com/product/advances-in-gain-scheduling-and-
fault-tolerant-control-techniques-1st-edition-damiano-rotondo-
auth/

Process control system fault diagnosis: a Bayesian

approach First Edition Gonzalez

https://textbookfull.com/product/process-control-system-fault-
diagnosis-a-bayesian-approach-first-edition-gonzalez/

Advanced Automotive Fault Diagnosis Automotive

Technology Vehicle Maintenance and Repair Tom Denton

https://textbookfull.com/product/advanced-automotive-fault-
diagnosis-automotive-technology-vehicle-maintenance-and-repair-
tom-denton/

Artificial Intelligence in Process Fault Diagnosis:

Methods for Plant Surveillance 1st Edition
Fickelscherer

https://textbookfull.com/product/artificial-intelligence-in-
process-fault-diagnosis-methods-for-plant-surveillance-1st-
edition-fickelscherer/

Fault Tolerant Message Passing Distributed Systems An

Algorithmic Approach Michel Raynal

https://textbookfull.com/product/fault-tolerant-message-passing-
distributed-systems-an-algorithmic-approach-michel-raynal/
Steven X. Ding

Advanced
Methods for Fault
Diagnosis and
Fault-tolerant
Control
Advanced Methods for Fault
Diagnosis and Fault-tolerant Control
Steven X. Ding

Advanced Methods for Fault

Diagnosis and Fault-tolerant
Control

13
Steven X. Ding
Universität Duisburg-Essen
Duisburg, Germany

ISBN 978-3-662-62003-8 ISBN 978-3-662-62004-5 (eBook)

https://doi.org/10.1007/978-3-662-62004-5

© Springer-Verlag GmbH Germany, part of Springer Nature 2021

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

Responsible Editor: Michael Kottusch

This Springer imprint is published by the registered company Springer-Verlag GmbH, DE part of
Springer Nature.
The registered company address is: Heidelberger Platz 3, 14197 Berlin, Germany
To My Parents and Eve Limin
Preface

This book is the third one in my book series plan. While the first two are
dedicated to model-based and data-driven fault diagnosis respectively, this one
addresses topics in both model-based and data-driven thematic fields, and increas-
ingly focuses on fault-tolerant control issues and application of machine learning
methods.
The enthusiasm for machine learning and big data technologies has consider-
able influences on the development of fault diagnosis techniques in recent years.
It seems that research efforts in the thematic domain of data-driven fault diagno-
sis gradually become a competition under the Olympic motto, faster transferring
machine learning methods to fault diagnosis applications, preferably adopting
higher actual (most popular) machine learning methods, and stronger publishing.
The main intention of this book is to study basic fault diagnosis and fault-tolerant
control problems, which build a framework for long-term research efforts in the
fault diagnosis and fault-tolerant control domain. In this framework, possibly uni-
fied solutions and methods can be developed for general classes of systems.
This book is composed of six parts. Besides Part I, which serves as a com-
mon basis for the subsequent studies, Parts II–VI are dedicated to five different
thematic areas. In Part II, optimal fault detection and estimation in time-varying
systems, detection and isolation of multiplicative faults in linear time-invariant
systems with uncertainties are addressed. Part III is dedicated to the investigation
on existence conditions of observer-based fault detection systems for a general
type of nonlinear systems, as well as on parameterisation and optimisation issues
of nonlinear observer-based fault detection systems. Part IV deals with statistical
and data-driven fault diagnosis, but is dedicated to different topics, including a
critical review of multivariate analysis based fault detection methods, optimal fault
detection and estimation in large-scale distributed and interconnected systems,
Kullback-Leibler divergence based fault detection schemes, and alternative fault
detection and clustering methods using symmetric positive definite data matrices
and based on Riemannian manifold theory. In Part V, the well-established ran-
domised algorithm theory is applied to the study on assessment and design of fault
diagnosis systems. Finally, fault-tolerant control schemes with a strong focus on

vii
viii Preface

performance degradation monitoring and recovering are studied in Part VI. These
parts are self-contained and so structured that they can also be used for self-study
on the concerned topics.
It should be mentioned that the final work on this book has been done during
the Corona crisis. I was so deeply sorry to hear of Dr. Jie Chen's death due to the
coronavirus. Jie was a good friend, a pioneer and a brilliant researcher of our com-
munity. He will be sadly missed.
This book would not be possible without valuable support from many peo-
ple. I would like to thank Prof. Dr.-Ing. L. Li from the University of Science and
Technology Beijing for the long-term collaboration and for the extensive editorial
corrections of the book. I am very grateful to my Ph.D. students and co-workers
for the valuable discussions and proofreading of the book chapters. They are Ms.
Caroline Charlotte Zhu, Ms. Ting Xue, Ms. Han Yu, Ms. Yuhong Na, Mr. Yannian
Liu and Mr. Jiarui Zhang.
Finally, I would like to express my gratitude to Mrs. Hestermann-Beyerle and
Mrs. Lisa Burato from Springer-Verlag. Mrs. Hestermann-Beyerle has initiated
this book project and Ms. Lisa Burato has perfectly managed the final submission
issues.

Duisburg Steven X. Ding

May 2020
Contents

Part I Introduction, Basic Concepts and Preliminaries

1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Trends and Mainstream in Research. . . . . . . . . . . . . . . . . . . . . . 4
1.1.1 Data-Driven, Statistic and Machine Learning Based
Fault Diagnosis Methods. . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 Model-Based Fault Diagnosis Research . . . . . . . . . . . . 8
1.1.3 Detection of Intermittent and Incipient Faults. . . . . . . . 9
1.1.4 Fault-Tolerant Control. . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Data-Driven and Model-Based Fault Diagnosis . . . . . . 11
1.2.2 Fault-Tolerant Control and Performance Degradation
Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.3 Performance Assessment of Fault Diagnosis and
Fault-Tolerant control Systems . . . . . . . . . . . . . . . . . . . 16
1.3 Outline of the Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.1 Part I: Introduction, Basic Concepts and
Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.2 Part II: Fault Detection, Isolation and
Estimation in Linear dynamic Systems. . . . . . . . . . . . . 18
1.3.3 Part III: Fault Detection in Nonlinear
Dynamic Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3.4 Part IV: Statistical and Data-Driven Fault Diagnosis
Methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.3.5 Part V: Application of Randomised Algorithms to
Assessment And design of Fault Diagnosis Systems. . . 24
1.3.6 Part VI: An Integrated Framework of Control and
Diagnosis, And fault-Tolerant Control Schemes. . . . . . 25
1.4 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

ix
x Contents

2 Basic Requirements on Fault Detection and Estimation. . . . . . . . . . 31

2.1 Fault Detection and Estimation Paradigm. . . . . . . . . . . . . . . . . . 31
2.2 Fault Detection and Estimation in the Probabilistic
Framework. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2.1 Fault Detection Performance Assessment. . . . . . . . . . . 33
2.2.2 Optimal Fault Detection and Estimation Problems. . . . 33
2.3 Fault Detection and Estimation in Deterministic Processes . . . . 34
2.3.1 Performance Assessment. . . . . . . . . . . . . . . . . . . . . . . . 34
2.3.2 Characterisation of Optimal Solutions. . . . . . . . . . . . . . 38
2.3.3 A General Form of Problem Formulation. . . . . . . . . . . 41
2.4 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3 Basic Methods for Fault Detection and Estimation in Static
Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1 A Basic Fault Detection and Estimation Problem. . . . . . . . . . . . 45
3.2 A General Form of Fault Detection and Estimation Problem. . . 47
3.3 Application of Canonical Correlation Analysis to Fault
Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3.1 An Introduction to CCA. . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3.2 Application to Fault Detection and Estimation. . . . . . . 51
3.3.3 CCA and GLR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4 Fault Detection and Estimation with Deterministic
Disturbances. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4.1 A Basic Fault Detection Problem. . . . . . . . . . . . . . . . . . 58
3.4.2 A General Form of Fault Detection and Estimation . . . 60
3.5 The Data-Driven Solutions of the Detection and
Estimation Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5.1 Fault Detection and Estimation in Statistic
Processes with sufficient Training Data. . . . . . . . . . . . . 63
3.5.2 Fault Detection Using Hotelling’s T 2 test statistic. . . . . 63
3.5.3 Fault Detection Using Q Statistic. . . . . . . . . . . . . . . . . . 66
3.5.4 Application of Principal Component Analysis
to Fault Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.5.5 LS, PLS and CCA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.6 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4 Basic Methods for Fault Detection in Dynamic Processes . . . . . . . . 81
4.1 Preliminaries and Review of Model-Based Residual
Generation Schemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.1.1 Nominal System Models. . . . . . . . . . . . . . . . . . . . . . . . 82
4.1.2 Observer-Based Residual Generation Schemes. . . . . . . 84
4.1.3 Parity Space Approach. . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.1.4 Kernel Representation and Parameterisation
of Residual Generators. . . . . . . . . . . . . . . . . . . . . . . . . . 87
Contents xi

4.2 Fault Detection in Linear Stochastic Processes. . . . . . . . . . . . . . 89

4.3 Fault Detection in Linear Processes with Unknown Inputs. . . . . 91
4.3.1 A Basic Form of the Unified Solution. . . . . . . . . . . . . . 91
4.3.2 Optimality of the Unified Solution. . . . . . . . . . . . . . . . . 92
4.3.3 An Alternative Interpretation and Solution Scheme . . . 95
4.4 A Data-Driven Method for Fault Detection in Dynamic
Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.4.1 An Input-Output Data Model. . . . . . . . . . . . . . . . . . . . . 96
4.4.2 Identification of an SKR-based Residual Generator . . . 97
4.4.3 An Alternative Algorithm of SKR Identification. . . . . . 100
4.4.4 Fault Detection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.4.5 Observer-Based Implementation. . . . . . . . . . . . . . . . . . 103
4.5 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5 Feedback Control, Observer and Residual Generation . . . . . . . . . . 109
5.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.1.1 State Feedback Control, RCF and Image
Representation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.1.2 Parameterisation of all Stabilising Controllers . . . . . . . 111
5.2 On Bezout Identity and Parameterisation of Stabilising
Controllers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.2.1 Observer-Based Realisations of Feedback
Controllers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.2.2 Bezout Identity and Feedback Control Loops. . . . . . . . 113
5.2.3 Parameterisation of Bezout Identity. . . . . . . . . . . . . . . . 115
5.3 An Observer-Based Fault-Tolerant Control Architecture . . . . . . 117
5.3.1 An Output Feedback Controller is an Estimator. . . . . . 117
5.3.2 A Fault-Tolerant Control Architecture. . . . . . . . . . . . . . 118
5.4 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

Part II Fault Detection, Isolation and Estimation in Linear Dynamic

Systems
6 General Solutions of Optimal Fault Detection. . . . . . . . . . . . . . . . . . 125
6.1 Algebraic Solutions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.1.1 An Algebraic Solution for Stochastic Processes . . . . . . 126
6.1.2 An Algebraic Solution for Deterministic Processes. . . . 128
6.1.3 Some Remarks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.2 An Observer-based Optimal Scheme. . . . . . . . . . . . . . . . . . . . . . 134
6.2.1 Problem Formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.2.2 A Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.2.3 Discussions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
xii Contents

6.3 A Data-driven Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

6.4 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
7 Fault Detection in Linear Time-Varying Systems . . . . . . . . . . . . . . . 141
7.1 Formulation of Fault Detection Problems. . . . . . . . . . . . . . . . . . 142
7.1.1 System Model and Assumptions . . . . . . . . . . . . . . . . . . 142
7.1.2 Observer-Based FD Systems. . . . . . . . . . . . . . . . . . . . . 142
7.1.3 Formulation of the Integrated Design of the FD
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.2 Problem Solutions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.2.1 An Alternative Input-Output Model of the
FD System Dynamics. . . . . . . . . . . . . . . . . . . . . . . . . . . 145
7.2.2 The Unified Solution. . . . . . . . . . . . . . . . . . . . . . . . . . . 146
7.3 Algebraic I/O-Model, Co-inner-outer Factorisation
and Unified solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
7.3.1 The Algebraic Input-Output Model for
LDTV Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
7.3.2 Co-inner-outer Factorisation and the Solution. . . . . . . . 155
7.4 Co-inner-outer, Lossless and Unified Solution . . . . . . . . . . . . . . 157
7.5 Operator-Based Co-inner-outer Factorisation and
Interpretations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
7.5.1 Operator-Based Models of the Residual Dynamics. . . . 160
7.5.2 Co-inner-outer Factorisation and the
Unified Solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
7.5.3 Robustness Vs. Sensitivity. . . . . . . . . . . . . . . . . . . . . . . 166
7.6 Examples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
7.7 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
8 Fault Estimation in Linear Dynamic Systems. . . . . . . . . . . . . . . . . . 177
8.1 Regularised Least Squares Estimation. . . . . . . . . . . . . . . . . . . . . 178
8.2 Least Squares Observer and Sensor Fault Estimation. . . . . . . . . 181
8.2.1 Problem Formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 181
8.2.2 Solution Algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
8.2.3 Discussions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
8.3 Fault Estimation: Least Squares Observer Vs.
Unified Solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
8.3.1 Sensor Fault Estimation Using the Unified Solution. . . 194
8.3.2 Comparison Study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
8.4 Least Squares Observer for Process Fault Estimation. . . . . . . . . 197
8.5 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Contents xiii

9 Detection and Isolation of Multiplicative Faults . . . . . . . . . . . . . . . . 205

9.1 System Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
9.1.1 Model Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
9.1.2 Relations Among the Model Forms. . . . . . . . . . . . . . . . 208
9.2 Observer-Based Fault Detection Schemes. . . . . . . . . . . . . . . . . . 210
9.2.1 Basic Ideas and Major Tasks. . . . . . . . . . . . . . . . . . . . . 211
9.2.2 An FD System Design Scheme for Feedback
Control Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
9.2.3 An FD System Design Scheme for
Open-Loop Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
9.3 System Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
9.3.1 Graph and Gap Metrics . . . . . . . . . . . . . . . . . . . . . . . . . 226
9.3.2 The K-gap. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
9.3.3 Residual Dynamics with Respect to Model
Uncertainties in feedback Control Systems. . . . . . . . . . 233
9.3.4 Fault Detection Performance Indicators. . . . . . . . . . . . . 236
9.3.5 Fault-to-uncertainty Ratio and Fault
Detectability in Uncertain systems . . . . . . . . . . . . . . . . 239
9.4 Fault Isolability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
9.4.1 A Motivation Example. . . . . . . . . . . . . . . . . . . . . . . . . . 243
9.4.2 Isolability of Multiplicative Faults. . . . . . . . . . . . . . . . . 244
9.4.3 Formulation of Fault Isolation Problems. . . . . . . . . . . . 247
9.5 Fault Isolation Schemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
9.5.1 Observer-Based Fault Isolation Algorithms. . . . . . . . . . 248
9.5.2 An SKR Identification Based Fault
Isolation Strategy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
9.6 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263

Part III Fault Detection in Nonlinear Dynamic Systems

10 Analysis and Design of Observer-Based Fault
Detection Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
10.1 Preliminaries and Problem Formulation . . . . . . . . . . . . . . . . . . . 268
10.1.1 System Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
10.1.2 Observer-Based Nonlinear Fault
Detection Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
10.1.3 Problem Formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 269
10.1.4 Notation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
10.2 On Observer-Based FD Systems. . . . . . . . . . . . . . . . . . . . . . . . . 271
10.2.1 Two Classes of Observer-Based FD Systems . . . . . . . . 271
10.2.2 On L∞ observer-based FD systems. . . . . . . . . . . . . . . . 271
10.2.3 On L2 observer-based FD systems. . . . . . . . . . . . . . . . . 274
xiv Contents

10.3 Design of Observer-Based FD Systems. . . . . . . . . . . . . . . . . . . . 275

10.3.1 Design of L∞ observer-based FD systems. . . . . . . . . . . 275
10.3.2 Design of L2-NFDF for affine systems . . . . . . . . . . . . . 276
10.3.3 An Extension to L2-RNFDF Design . . . . . . . . . . . . . . . 279
10.3.4 On FD Schemes for L2-stable Affine Systems. . . . . . . . 281
10.4 Examples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
10.5 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
11 Parameterisation of Observer-Based Fault Detection Systems . . . . 295
11.1 Problem Formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
11.2 Parameterisation of Nonlinear Residual Generators. . . . . . . . . . 297
11.3 Parameterisation of Nonlinear Fault Detection Systems. . . . . . . 300
11.3.1 State Space Configuration. . . . . . . . . . . . . . . . . . . . . . . 301
11.3.2 L∞-class FD systems. . . . . . . . . . . . . . . . . . . . . . . . . . . 302
11.3.3 L2-class FD systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
11.4 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
12 Optimal Fault Detection of a Class of Nonlinear Systems . . . . . . . . 309
12.1 System Models and Preliminaries. . . . . . . . . . . . . . . . . . . . . . . . 310
12.1.1 System Model and Hamiltonian Systems . . . . . . . . . . . 310
12.1.2 Inner. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
12.2 Definition of Co-Inner. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
12.3 Co-inner-outer Factorisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
12.3.1 Basic Idea. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
12.3.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
12.4 Application to Fault Detection. . . . . . . . . . . . . . . . . . . . . . . . . . . 322
12.4.1 Threshold Setting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
12.4.2 Design of a Post-filter . . . . . . . . . . . . . . . . . . . . . . . . . . 323
12.5 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326

Part IV Statistical and Data-driven Fault Diagnosis Methods

13 A Critical Review of MVA-based Fault Detection Methods. . . . . . . 329
13.1 On Projection Technique and Its Use in Fault Detection. . . . . . . 330
13.1.1 Problem Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
13.1.2 Discussion: The Pros and Cons. . . . . . . . . . . . . . . . . . . 331
13.2 Data Centering, Time-Varying Mean and Variance. . . . . . . . . . . 336
13.2.1 Problem Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
13.2.2 Data Centering: Conditions and Consequence. . . . . . . . 336
13.2.3 On Handling Time-Varying Mean. . . . . . . . . . . . . . . . . 338
13.3 On Detecting Multiplicative Faults and T 2-test Statistic. . . . . . . 341
13.3.1 Problem Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
Contents xv

13.3.2 Miss Detection of Multiplicative

Faults Using T 2-test Statistic. . . . . . . . . . . . . . . . . . . . . 342
13.4 Assessment of Fault Detection Performance. . . . . . . . . . . . . . . . 343
13.4.1 Problem Formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 343
13.4.2 On FAR and FDR Computation. . . . . . . . . . . . . . . . . . . 344
13.4.3 On Confidential Computations of FAR and FDR. . . . . . 345
13.5 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
14 Data-Driven Fault Detection in Large-Scale and Distributed
Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
14.1 Preliminary Knowledge in Network and Graph Theory . . . . . . . 351
14.1.1 Basic Concepts in Graph Theory. . . . . . . . . . . . . . . . . . 351
14.1.2 An Introduction to Distributed Average Consensus. . . . 352
14.2 An Intuitive Average Consensus Based Fault
Detection Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
14.2.1 System Configuration and Problem Formulation. . . . . . 355
14.2.2 A Basic Average Consensus Algorithm
for Fault Detection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
14.2.3 Performance Analysis and Discussion. . . . . . . . . . . . . . 359
14.3 Practical Algorithms Towards Average Consensus
Aided Fault Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
14.3.1 An Algorithm with Distributed Test Statistic
Computation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
14.3.2 Analysis of Convergence Rate and an
Alternative Fault Detection Algorithm. . . . . . . . . . . . . . 366
14.3.3 Re-visiting Fault Detection Performance. . . . . . . . . . . . 371
14.4 A General Consensus Based Optimal Fault
Detection Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
14.4.1 Problem Formulation, Solution and Algorithm. . . . . . . 373
14.4.2 Variations of the Algorithm. . . . . . . . . . . . . . . . . . . . . . 376
14.5 A Distributed Kalman Filter Based Fault
Detection Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
14.5.1 Problem Formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 382
14.5.2 Modelling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
14.5.3 Kalman Filter Based Optimal Fault
Detection: The Basic Idea And the
Centralised Solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
14.5.4 A Distributed Kalman Filter-Based
Optimal Fault Detection Scheme. . . . . . . . . . . . . . . . . . 388
14.6 Correlation-Based Fault Detection Schemes and Algorithms. . . 392
14.6.1 Problem Formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 392
14.6.2 A Basic Iteration Learning Algorithm. . . . . . . . . . . . . . 396
xvi Contents

14.6.3 Online Detection Algorithm. . . . . . . . . . . . . . . . . . . . . . 401

14.7 Analysis and Alternative Algorithms. . . . . . . . . . . . . . . . . . . . . . 402
14.7.1 Couplings, Correlations and Estimation . . . . . . . . . . . . 402
14.7.2 Alternative Algorithms. . . . . . . . . . . . . . . . . . . . . . . . . . 406
14.8 Combined Application of the Consensus and
Correlation Based Schemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
14.8.1 Models and Problem Formulation. . . . . . . . . . . . . . . . . 413
14.8.2 Distributed Kalman Filter Based Fault Detection
Scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
14.8.3 Training and Online Implementation Algorithms . . . . . 419
14.9 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
15 Alternative Test Statistics and Fault Detection Schemes. . . . . . . . . . 425
15.1 A General Formulation and Solution of GLR-based Fault
Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
15.2 KL Divergence Based Fault Detection Schemes. . . . . . . . . . . . . 429
15.2.1 On KL Divergence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
15.2.2 KL Divergence Based Fault Detection. . . . . . . . . . . . . . 430
15.2.3 KL Divergence and GLR Based Methods. . . . . . . . . . . 432
15.3 Asymptotic Behaviour of GLR and KL Divergence as Test
Statistics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
15.4 SPD Matrix Based Test Statistics and
Fault Detection Schemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
15.4.1 Manifold of Symmetric and
Positive-definite Matrices. . . . . . . . . . . . . . . . . . . . . . . . 439
15.4.2 Riemannian Distance Based Fault
Detection Schemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
15.4.3 Clustering and Clustering Based Fault Detection
Schemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
15.4.4 Examples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454
15.5 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465

Part V Application of Randomised Algorithms to Assessment and

Design of Fault Diagnosis Systems
16 Probabilistic Models and Randomised Algorithms. . . . . . . . . . . . . . 469
16.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
16.2 Probabilistic Models for Uncertain Systems. . . . . . . . . . . . . . . . 470
16.2.1 Probabilistic Parameter Model. . . . . . . . . . . . . . . . . . . . 471
16.2.2 Probabilistic Time Function Parameter Model . . . . . . . 472
16.2.3 Probabilistic Uncertainty Mode Model. . . . . . . . . . . . . 473
16.3 Probabilistic Fault and Evaluation Function Models. . . . . . . . . . 473
16.3.1 Probabilistic Fault Models. . . . . . . . . . . . . . . . . . . . . . . 473
Contents xvii

16.3.2 Probabilistic Models of Test Statistics

and Evaluation Functions. . . . . . . . . . . . . . . . . . . . . . . . 475
16.4 Preliminaries of Randomised Algorithms. . . . . . . . . . . . . . . . . . 475
16.5 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478
17 Assessment of Fault Detection Performance and Computation
Algorithms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
17.1 False Alarm Rate and RA-aided Assessment. . . . . . . . . . . . . . . . 479
17.2 Fault Detection Rate and RA-aided Assessment. . . . . . . . . . . . . 483
17.3 Mean Time to Fault Detection and RA-aided Assessment . . . . . 485
17.4 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
18 RA-based Design of Fault Detection Systems. . . . . . . . . . . . . . . . . . . 491
18.1 Randomised Algorithms Based Threshold Settings. . . . . . . . . . . 491
18.1.1 Algorithm I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
18.1.2 Algorithm II. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
18.2 A RA-based Design of Observer-based Fault
Detection Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
18.2.1 Basic Idea and Problem Formulation. . . . . . . . . . . . . . . 495
18.2.2 Optimal Solution and the RA Realisation
Algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
18.3 A Multiple Monitoring Indices Based Fault
Detection Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
18.3.1 Motivation and Review of Fault Detection . . . . . . . . . . 501
18.3.2 SVM- and RA-based Design of Threshold
Hyperplane. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
18.3.3 Randomised Algorithm of Designing the
Threshold Hyperplane. . . . . . . . . . . . . . . . . . . . . . . . . . 506
18.4 Benchmark Study on a Three-tank System. . . . . . . . . . . . . . . . . 507
18.4.1 System Setup and Models . . . . . . . . . . . . . . . . . . . . . . . 507
18.4.2 FAR, FDR and MT2FD Assessment. . . . . . . . . . . . . . . 510
18.4.3 Fault Detection Results with Real Data. . . . . . . . . . . . . 513
18.4.4 Multiple Monitoring Indices Based Fault
Detection using Algorithm. . . . . . . . . . . . . . . . . . . . . . . 514
18.5 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520

Part VI An Integrated Framework of Control and Diagnosis, and

­Fault-tolerant Control Schemes
19 Residual Centered Control and Detection . . . . . . . . . . . . . . . . . . . . . 523
19.1 Residuals and Residual-based Unified System Models. . . . . . . . 523
19.2 Control Performance Degradation, Assessment and
Monitoring. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528
xviii Contents

19.2.1 Loop Performance Degradation. . . . . . . . . . . . . . . . . . . 528

19.2.2 Assessment and Monitoring of Control
Performance Degradation. . . . . . . . . . . . . . . . . . . . . . . . 530
19.2.3 Loop Performance Degradation Assessment and
Monitoring. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
19.3 SIR of Feedback Controller and System Performances. . . . . . . . 536
19.3.1 Stability Margin. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
19.3.2 Residual and Fault Detectability . . . . . . . . . . . . . . . . . . 544
19.3.3 Performance Degradation. . . . . . . . . . . . . . . . . . . . . . . . 547
19.3.4 Summary: A Unified Perspective. . . . . . . . . . . . . . . . . . 548
19.4 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 551
20 Performance Recovery and Fault-Tolerant Control Schemes . . . . . 553
20.1 Reviewing LQ Control Problems. . . . . . . . . . . . . . . . . . . . . . . . . 553
20.1.1 LQG Control Problem. . . . . . . . . . . . . . . . . . . . . . . . . . 553
20.1.2 On Solutions of LQG Control Problem. . . . . . . . . . . . . 554
20.1.3 On Solutions of LQR Control Problem. . . . . . . . . . . . . 558
20.2 An LQ Optimal Observer Design Approach. . . . . . . . . . . . . . . . 562
20.2.1 Problem Formulation and the Basic Idea
of the Solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
20.2.2 A Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 564
20.2.3 The Dual Form. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 567
20.2.4 LQ Observers for Systems with Input Vector . . . . . . . . 570
20.3 LQ Control Performance Monitoring and Recovering . . . . . . . . 571
20.3.1 Problem Formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 571
20.3.2 Reference-Model Based Detection of Performance
Degradation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 572
20.3.3 Performance Residual Based Detection of
Performance Degradation. . . . . . . . . . . . . . . . . . . . . . . . 575
20.3.4 Performance Recovery by Updating the
State Feedback Gain. . . . . . . . . . . . . . . . . . . . . . . . . . . . 578
20.4 Real-Time Monitoring and Optimisation of Observers. . . . . . . . 584
20.4.1 Problem Formulation and Basic Idea. . . . . . . . . . . . . . . 585
20.4.2 Monitoring and Detection of Estimation
Performance Degradation. . . . . . . . . . . . . . . . . . . . . . . . 586
20.4.3 Performance Degradation Recovery . . . . . . . . . . . . . . . 588
20.5 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 596
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 600
21 Performance-Based Fault-Tolerant Control and
Performance Recovery. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601
21.1 Motivation and Basic Ideas for Performance-Based FTC. . . . . . 601
21.2 An Approach of Recovering Stability Margin Degradation . . . . 603
21.2.1 Preliminaries and Problem Formulation . . . . . . . . . . . . 604
Contents xix

21.2.2Performance Degradation Detection and

Recovering Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . 607
21.3 An Approach to Loop Performance Recovery. . . . . . . . . . . . . . . 617
21.3.1 Dual Form of Youla Parameterisation and
Parameterisation of Process Uncertainty. . . . . . . . . . . . 618
21.3.2 Loop Performance Degradation Model and
Problem Formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 620
21.3.3 Loop Performance Monitoring and Degradation
Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 622
21.3.4 Loop Performance Degradation Recovery. . . . . . . . . . . 625
21.4 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 625
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627
22 Data-Driven Fault-Tolerant Control Schemes. . . . . . . . . . . . . . . . . . 629
22.1 Closed-Loop Identification of Data-Driven SIR and SKR . . . . . 629
22.1.1 Data-Driven SIR, SKR, and Problem Formulation. . . . 629
22.1.2 Identification of Is , Ks. . . . . . . . . . . . . . . . . . . . . . . . . . . 632
22.2 Recursive SIR and SKR Based Loop Models. . . . . . . . . . . . . . . 634
22.2.1 A Recursive SIR Based Closed-Loop Model. . . . . . . . . 634
22.2.2 A Recursive SKR Based Closed-Loop Model. . . . . . . . 637
22.3 Performance Monitoring and Performance Degradation
Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 639
22.3.1 Performance Degradation and Its Detection . . . . . . . . . 639
22.3.2 Performance Degradation Recovery . . . . . . . . . . . . . . . 643
22.3.3 Performance Degradation Recovery by
Reduced Order Controllers . . . . . . . . . . . . . . . . . . . . . . 647
22.4 Notes and References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 651
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653

Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 655
Notation

∀ for all
∈ belong to
⊂ subset
∪ union
∩ intersection
≡ identically equal
≈ approximately equal
A := B, B =: A A is defined as B
⇒ implies
⇔ equivalent to
>> (<<) much greater (less) than
max (min) maximum (minimum)
sup (inf) supremum (infimum)
R and C field of real and complex numbers
Rn and Cn space of real and complex n-dimensional vectors
Rn × m and C n×m space of n by m real and complex matrices
H2 H2n signal space of all signals (n-dimensional vectors of signals)


with bounded energy

function space of all transfer functions of stable systems (n by
 n × m
H∞ H∞
 m-dimensional transfer function matrices of stable systems)
RH∞ RHn∞× m space of all rational transfer functions of stable systems (n by


m-dimensional rational transfer function matrices)

L∞ Lebesgue space of all functions essentially bounded on the
imaginary axis

For the definitions of H2, H∞, RH∞ and L∞, and the reader is referred to.

xxi
xxii Notation

XT transpose of matrix X
X* conjugate transpose of (complex) matrix X
X⊥ orthogonal complement of matrix X
X−1 inverse of matrix X
X+ pseudo-inverse of matrix X
X− left or right inverse of matrix X
diag(X1, ···, Xn) block diagonal matrix formed with X1, ···, Xn
X(i:j, p:q) submatrix consisting of the i-th to the j-th rows and the p-th to
the q-th columns of matrx X
col(X)  
x1
 .. 
vectorise X, col(X) =  .  ∈ Rnm , for
xm
X = [x1 · · · xm ] ∈ Rn×m , xi ∈ Rn , i = 1, · · · , m
rank(X) rank of matrix X
tr(X) trace of matrix X
det(X) determinant of matrix X
(X) eigenvalue of matrix X
σ (X) (σ max (X)) largest (maximum) singular value of matrix X
least (minimum) singular value of matrix X
 
σ (X) σmin (X)
σi (X) the i-th singular value of matrix X
�·� Euclidean norm of a vector
�·�F Frobenius norm of a matrix
l 2-bounded discrete-time signals with bounded energy
L2-bounded continuous-time signals with bounded energy
G(p) transfer matrix, p is either s for continuous-time systems or z
for discrete-time systems
G*(jω), G∗ ejθ conjugate of G(jω), G ejθ
   

(A,B,C,D) shorthand for the state space representation

rank (G(p)) normal rank of G(p)
G2 H2 norm of (stable) transfer function matrix G
�G�∞ H∞ norm of (stable) transfer function matrix G

For the definitions of G2 and �G�∞, the reader is referred to.
Notation xxiii

Pr(a < b) probability that a < b

N (a, ) Gaussian distribution  with mean vector a and
covariance matrix
x ~ (a, ) x is distributed as (a, )
 
χ 2 (m) chi-squared distribution with m degrees of freedom
E (x) expectation of x
var(x) variance of x
cov(x,y) covariance (matrix) of x and y
var(x) = cov(x) covariance (variance-covariance) matrix of vector x
i. i. d. independent and identically distributed
 Part I
Introduction, Basic Concepts and
Preliminaries
Chapter 1
Introduction

The past two decades have witnessed tremendous development in the field of fault
diagnosis and fault-tolerant control. This trend is the logic consequence of the ever
increasing demands for highly economic and ecological operations of technical sys-
tems, processes and assets in all industrial sectors. The speed, at which new concepts,
schemes and methods are developed, is rapid. In the course of this development, in-
numerable methods and successful applications have been reported.
Fault diagnosis technique is an engineering thematic area and its applications can
be found across all technical fields. Consequently, in its history, the development
of fault diagnosis technique and methods was obviously formed by individual tech-
nical demands and characteristics of systems and assets under monitoring. There
was no uniform and standardised theory and framework. The situation has changed
dramatically since this decade. Like all technical and scientific disciplines, in the
era of information, digitalisation and big-data, the impact of artificial intelligence,
computer science, information theory and communication technology on the de-
velopment of fault diagnosis technique is enormous and everywhere. A uniform
technological framework towards intelligent and data-driven fault diagnosis is being
newly established.
Generally, fault-tolerant control (FTC) deals with feedback control systems. From
the methodological viewpoint, fault-tolerant control is a thematic field of control
theory and engineering. Logically, the development of FTC technique can be well
characterised by the application of advanced control theoretical methods and asso-
ciated mathematical tools. Robust control theory, adaptive control algorithms and
newly optimisation methods including model predictive control (MPC) are the ma-
jor methodological tools for the design and implementation of fault-tolerant control
systems. The impact of the main technologies in the era of information and big-data
on the FTC technique is reflected by the recent research efforts on FTC in networked
control systems (NCSs) and cyber-physical systems (CPSs).

© Springer-Verlag GmbH Germany, part of Springer Nature 2021 3

S. X. Ding, Advanced methods for fault diagnosis and fault-tolerant control,
https://doi.org/10.1007/978-3-662-62004-5_1
4 1 Introduction

1.1 Trends and Mainstream in Research

In the current decade, the following new trends can be observed in comparison with
the research efforts in the past decades,

• data-driven, multivariate analysis (MVA) and machine learning based methods are
dominate in the field of process monitoring and fault diagnosis,
• research efforts on model-based fault diagnosis pounce on fault detection and
estimation methods for special classes of systems like networked and distributed
systems, event-triggered or switched systems,
• detection of two special types of faults, the so-called intermittent and incipient
faults, is receiving remarkable attention, and
• in the research area of fault-tolerant control, efforts are strongly focused on

– fault estimation and, based on it, fault compensation strategy, as well as

– application of real-time optimisation techniques like MPC technique.

1.1.1 Data-Driven, Statistic and Machine Learning Based

Fault Diagnosis Methods

The enthusiasm for machine learning (ML) and big data technologies significantly
influences the developments in all engineering and scientific areas. The key issues in
the diagnosis framework, like feature generation and analysis, (fault) classification
and decision making, are also basic tasks in machine learning. It is a logic conse-
quence that most of the existing ML methods and concepts have been introduced
into the fault diagnosis framework, often in combination with MVA methods. At the
very beginning of this development, the process of transferring an existing MVA-
ML method to the fault diagnosis application was built with sophisticated research
efforts, which resulted in certain time lag, but allowed sufficient time for necessary
validations. In the course of this development, the time lag of the transfer processes
has become shorter and shorter. Recently, it seems that it is becoming a competition
of publishing applications of newly developed ML methods and algorithms to fault
diagnosis. The most recent example is the application with deep learning technique.
The consequence of this copy-and-paste style of research is that it is hard to have a
solid overview of all published MVA-ML based fault diagnosis methods. Below, we
would like to shortly review and analyse the basic ideas and common working prin-
ciples of the existing MVA-ML based fault diagnosis methods without addressing
the methods in detail.
Which MVA-ML methods are really useful for a reliable and efficient fault di-
agnosis? To answer this question, let us recall the basic tasks of a fault detection
system as an example. Note that fault detection is the first and most importance task
of fault diagnosis, and most of the fault isolation and classification problems can be
1.1 Trends and Mainstream in Research 5

re-formulated as a number of fault detection problems. A fault detection system or

scheme is composed of three essential components and steps:
• data collection and pre-processing,
• feature generation and extraction, and
• decision making and classification algorithms.

Most of the published MVA-ML based fault diagnosis methods and algorithms are
dedicated to the second and third steps on the assumption that the collected data
have been well pre-processed and satisfy the required conditions. The basic idea
behind the step with feature generation and extraction is to solve the core problem in
fault detection: faults vs. uncertainties. Without the existence of uncertainties, fault
detection is straightforward. Regrettably, in most of technical systems and assets,
uncertainties are inevitable and exist in different forms. Measurement and process
noises are the simplest form of uncertainties, which can be, for example, modelled as
random variables with certain distributions. Variations in system and asset parame-
ters caused by, for instance, varying system operation conditions, ageing in machine
and asset components or changing environmental conditions around the system and
asset, result in uncertainties which are hard to be described analytically. A feature is
a function of all reliable (possibly pre-processed) measurement variables. It should
be formed (generated) in such a way that it depends on the faults to be detected.
Unfortunately, it is unavoidable that a feature is, more or less, corrupted with uncer-
tainties. For the fault detection purpose, a good feature is a function that is sensitive
to the faults and less affected by the uncertainties. In the step of decision making, a
decision is made based on an analysis of the feature. The mostly adopted strategy is
to compare the feature values in the fault-free and faulty operations, and a decision
is made on account of the difference between these values. This requires that the
feature (as a function) should be a metric or a good measure or an indicator for the
influences of the faults and uncertainties. In an abstract form, we can summarise
the fault detection problem as finding a metric for assessing the measurement data.
Roughly speaking, a defined fault detection problem is optimally solved, when the
ratio,
the metric value corresponding to the faulty operation
,
the metric value corresponding to the fault-free operation

is at the largest.
It should be emphasised that both data-driven and model-based fault detection
schemes follow the above-described strategy. Their differences lie in the realisation
and implementation. In the data-driven fault diagnosis framework, there are two
types of strategies for the steps of features generation and decision making:

• modelling of a designed feature by learning,

• automated learning of the two steps in a single learning process.

In the first strategy, on account of certain assumptions or a priori knowledge of the

system under consideration or applying some statistic and mathematical methods, a
6 1 Introduction

model mapping the process measurement variables to the feature is first established.
By means of collected operation data, the model parameters and even the model
structure are then identified. This process is called training or learning, and runs
typically offline. Based on such a model, an online detection of faults can be realised
by checking and analysing the feature value delivered by the model. Below are some
examples. For the technical details, the reader is referred to the study in the subsequent
chapters.
Example 1.1 PCA based fault detection (referred to Sub-Sec. 3.5.4). Principal com-
ponent analysis (PCA) is a simple fault detection method, in which two features, the
so-called T 2 and SPE (squared prediction error) test statistics, are defined. Re-
grettably, in the literature the PCA method is generally introduced as a projection
algorithm without describing (i) the assumptions for its use in fault detection and
(ii) its statistic interpretations. In order to apply T 2 and SPE test statistics success-
fully to fault detection, (i) the process measurement variables are assumed to be
normally distributed, (ii) a data pre-processing is needed, and (iii) the faults to be
detected can be modelled as the so-called additive faults which cause changes in the
mean (vector) of the process measurement variables. Moreover, T 2 test statistic is
the so-called Mahalanobis distance which is a dissimilarity measure between two
random vectors of the same distribution with the same covariance matrix. Finally,
the detection logic is established based on the statistic distributions of both features.
The model parameters of the T 2 and SPE test statistics are the mean (vector) and the
covariance matrix of the process measurement variables, which should be estimated
(learned) using the collected process data.
Example 1.2 KL divergence based fault detection (referred to Sect. 15.2). Kullback-
Leibler (KL) divergence is a dissimilarity measure between distributions. When it
is assumed that the fault to be detected would cause changes in the distribution of
the measurement variables, KL divergence is a good feature for detecting such a
fault. Since for the computation of the KL divergence model the distribution of the
(random) measurement variables during the fault-free operation is needed, in the
data-driven framework an estimation of the distribution is to be performed during
the training phase using the collected process data. To this end, there exist numerous
algorithms, among them the so-called kernel density estimation (KDE) method is
very popular.
Example 1.3 SVM based fault detection (referred to Sub-section 18.3.2). Support
vector machine (SVM) is a popular ML method. One of its application fields is
classification, which can be directly adopted for the use of fault detection. In order
to solve the problem of measuring the distance between two data sets, the faulty
and fault-free data sets, a (threshold) hyperplane is introduced, which separates the
faulty data set from the fault-free one, and thus guarantees a defined (Euclidean)
distance between these two data sets. During the training, the hyperplane model is
to be determined.
Artificial neural networks (ANNs) are the most popular technique to realise feature
generation and decision making in a single step. For example, for the fault detection
1.1 Trends and Mainstream in Research 7

purpose, an ANN is trained using the collected operation data in such a way that the
ANN directly delivers the decision result for faulty (alarm) or fault-free. That is, an
ANN-based fault diagnosis system is a mathematical model that describes relations
between the measurement variables and the faults. During the training, this model
is identified using the measurement data. In other words, training an ANN is in fact
to perform a model identification. In application, the model is driven by the system
measurement variables (as the input) and delivers information about the possible
faults (as output).
The unbelievable media effects of recent remarkable successes of deep learning
technique in some technical fields have initiated the enthusiasm for applying different
forms of ANN schemes, in particular the deep learning technique, to fault diagnosis.
Thanks to their self-learning capacity, ANNs are able to extract features of the faults
to be detected from the training data automatically. And these features enable us to
distinguish the faults from the uncertainties in the measurement data efficiently. On
the other hand, its extreme dependence on the training data also limits the application
of the ANN technique in fault diagnosis. One obvious problem is the availability of
the so-called labelled data. Labelled data are data which have been recorded under
known operation conditions. In order to train an ANN, the expected output of the
model (to be identified) with respect to the given training data (as input) should
be known. In fault detection applications, this means, whether the data have been
recorded by faulty operation and thus labelled by faulty or by fault-free operations
and labelled by fault-free should be known. In a reliable technical system, faulty
operation is an event with (very) low probability, while most of the data are collected
during fault-free operations (and thus labelled as fault-free). As a result, the number
of the data samples for the faulty operation is in general much smaller than the
number of the data labelled by fault-free. A further concern is the transferability.
For instance, an ANN-based fault diagnosis system for an industrial asset has been
well trained and constructed using a huge number of data collected as the asset is
in operation. Even for all these efforts at great expense, it cannot be guaranteed that
this diagnosis system can be used for another asset of the same type but located in a
different operation environment.
A potential and promising solution to the above problems is the application of the
so-called transfer learning technique. Transfer learning is a research field in machine
learning and deals with, roughly speaking, development of methods which enable
to apply learned knowledge of the solution of a defined problem to solving different
but relevant problems. Recall that data-driven fault diagnosis can be schematically
defined as a model identification problem and solved using the system data (as
model inputs) and labels (knowledge) of the faults (as model outputs). To achieve
fault diagnosis of two different systems, two different models are to be identified
using two different system data sets and two labelled data sets. Assume that

• two systems, say S1 and S2, under consideration are similar,

• for one system, say S1, the operation data are well labelled, and thus its (diagnosis)
model M1 is well identified,
• for system S2, there exist few data and they are poorly labelled.
8 1 Introduction

Consequently, it is hard to identify a reliable diagnosis model M2 for system S2 using

the available data. This problem can be addressed using transfer learning methods,
which help us to use the well-labelled operation data and model of system S1 to
improve the identification of model M2 based on probabilistic relations and using
optimisation algorithms.

Example 1.4 Application of transfer learning to machine diagnosis. It is well known

that machine diagnosis becomes a challenging task, once a machine is integrated
in a large-scale and complex system. Differently, under laboratory conditions, the
same machine can be well tested and sufficient operation data can be collected. This
allows us to build a reliable machine diagnosis model for the machine (under labo-
ratory conditions). Now, transfer learning methods may support system engineers to
apply machine diagnosis knowledge, including the machine diagnosis model and the
operation data under laboratory conditions to diagnosing the machine integrated in
the large-scale system. In the literature given at the end of this chapter, successful
cases of this application have been reported.

In their recent survey report (see the reference given at the end of this chapter),
Lei et al. have pointed out that the transfer learning technique would be the major
technological tool for data-driven fault diagnosis of the next decade.

1.1.2 Model-Based Fault Diagnosis Research

Model-based fault diagnosis techniques played once a very dominant role in the
research domain of fault diagnosis. Even today in the era of information and digital-
isation, model-based fault diagnosis methods are widely accepted as an efficient and
powerful technique in dealing with fault diagnosis issues for dynamic systems.
After a dynamic development in the 1980s and 1990s, in which the framework for
the model-based fault diagnosis techniques was established with three main research
areas,

• observer-based fault detection, isolation and estimation,

• parity-space based fault detection and isolation, and
• parameter identification based fault detection and estimation,
all the three areas of the model-based fault diagnosis technique were well equipped
with basic concepts, methods and algorithms. It is remarkable that most of these
methods and algorithms are the well-established results in control theory with slight
modifications. In the following decade, the research focus was mainly on robustness
issues, which, due to the use of mathematical models, is a natural and necessary key
step. In the course of this development, the foundations for system design techniques
with the associated (mathematical) tools, for instance, H∞ robust theory, handling
of special classes of nonlinear systems like systems satisfying Lipschitz conditions,
Takagi-Sugeno (T-S) fuzzy systems etc., system adaptive technique, sliding mode
1.1 Trends and Mainstream in Research 9

control methods, have been laid for the research efforts in the recent decade. In a
certain sense, the most valuable contributions in this time period have been the “trans-
lation and standardisation works”, which have resulted in formalisation of common
fault diagnosis problems by means of problem formulations known in control theory
and engineering. A representative example is the so-called H− /H∞ design scheme
for observer-based fault detection systems with LMI (linear matrix inequality) algo-
rithms as the solution tool. It is noteworthy that the common basis of these techniques
is Lyapunov theory. Reviewing the research results published in the current decade,
it can be recognised that

• there are no significant contributions (i) to the model-based fault diagnosis frame-
work, and (ii) to the existing essential design schemes, as established and developed
in the past three decades,
• the main research efforts have been focused on addressing fault diagnosis issues,
on the basis of the well-established formalisation framework, for special kinds
of dynamic systems like systems with different types of time-delays, NCSs, dis-
tributed large-scale systems, switching systems, and newly multi-agent systems
and CPSs, and
• in this regard, a great number of publications have been dedicated to the design
algorithms with skilled application of existing mathematical and control theoretical
tools.

Recently, a further trend can be observed. More and more reported research efforts
have been devoted to the (robust) fault estimation issues with the argument that, once a
fault is estimated, fault detection and isolation problems are solved as well. The real
reason behind this handling is in fact a simplification of the problem formulation
and handling. As mentioned, a fault detection problem is, in its core, a trade-off
between the sensitivity to the fault and robustness against uncertainties or simply
fault detection rate vs. false alarm rate. From the mathematical point of view, this is
a multi-objective optimisation problem, and its solution is often a challenging task.
In the framework of (robust) fault estimation, the problem is generally formulated
as minimising the estimation error of the fault with respect to uncertainties. It is
well-known in control theory that such a problem can be efficiently addressed in the
framework of Lyapunov theory.
As a summary, it can be concluded that the research on the major model-based
fault diagnosis techniques has been strongly driven and formed by the development
of control theory. And, this trend will be reinforced rather than weakened.

1.1.3 Detection of Intermittent and Incipient Faults

Driven by industrial applications and continuously increasing demands for reliability

and safety, recent research efforts have been devoted to two special types of faults,
the so-called intermittent and incipient faults. Intermittent faults are malfunctions
 Another Random Document on
Scribd Without Any Related Topics
remained British subjects in the eyes of the law, and Sir Peregrine Maitland's
Proclamation of August 21, 1845, at the Cape, expressly reserved the rights
of the Crown in this connection.

Moshesh the
Meanwhile, however, two other communities had
Basuto developed in their neighbourhood. East of what afterwards
became the Orange Free State and in territory which the
emigrant farmers, or Boers as they were beginning to be called, claimed for
themselves, an exceedingly able native chief, in the person of Moshesh the
Basuto, had risen into power and had welded together the scattered
fragments of tribes which had been crushed by the raids of the Matabele and
Zulus. From the rugged heights of Thaba Bosigo he dominated a large extent
of country, an increasing native population and much spoil in cattle and
slaves. To the south and west of the Boers two half-breed leaders named
Adam Kok and Waterboer had established themselves respectively with
strong, armed bands of Griquas—the name given to the offspring of Dutch
farmers and Hottentot women—and had become a recognized force. With
Moshesh they constituted the elements of a new British policy which was
inaugurated in 1843. The Colonial Office did not want at this time to extend
its territories. South Africa, indeed, appeared during the first portion of this
century as the least promising, and the most turbulent and troublesome, of all
British possessions. The soil was supposed to be arid and without fertility or
minerals, the population seemed hostile and the net result of colonization
and administration had been a series of costly Kaffir wars. In dealing with
the Kaffirs, or Kosas, on the eastern frontier of the Colony the British
Government had shown this disinclination with quite sufficient clearness.
But to allow the emigrant Boers to repudiate their allegiance was another
matter, and even to the not very far-seeing statesmen of the Colonial Office
of that day it presented possibilities deserving of consideration. With Sir
Harry Smith's arrival and the termination of the Kaffir War of 1846-47 came
another development of the situation. The new Governor of Cape Colony,
who for the first time had also been appointed High Commissioner with
power of control over native matters outside of the bounds of the Colony,
visited the Orange River region, looked into the results of the Treaty State
policy, came to the conclusion that agreements with native chiefs were like
arrangements made with little children, and determined to suppress these
creations of missionary statecraft as soon as might be possible.
 Orange River
Meanwhile the High Commissioner was well received at
Sovereignty Bloemfontein, and soon made arrangements with Adam
Kok and Moshesh which greatly curtailed their authority
and independence. On February 3, 1848, he announced the annexation to
British dominions of the whole territory between the Vaal and Orange Rivers
and the Drakensberg mountains under the name of the Orange River
Sovereignty. The colored population was left under the control of its chiefs,
and their land was carefully reserved for their own use. All relations between
tribes, however, or with Europeans, were to be guided by British authorities.
Major Warden was continued at Bloemfontein as the Governor, or Resident,
and Sir Harry Smith returned to Cape Town after having carried out a policy
which should have been effected long before. And it was now too late.
Although without any definite government amongst themselves, or any
allegiance to the little republics which had sprung up over the Vaal, a certain
number of Dutch farmers in the new Sovereignty would not accept British
rule, and they were speedily aided by the Transvaal Boers under Pretorius in
a direct attack upon Bloemfontein. Major Warden was compelled to
surrender, and the British officials were speedily driven out of the country.
Sir Harry Smith, however, was too vigorous and able a commander to stand
this sort of thing, and he hastily got some troops together, crossed the
Orange River, attacked Pretorius in a strong position at a place called
Boomplaatz, defeated him and re-established the Sovereignty Government.
Those of the Boers who were inveterately opposed to British rule at once
crossed the Vaal and were not interfered with by British officials. Their
places, to some extent, were taken by fresh emigrants from Cape Colony,
many of them English, and from this time forward the Orange River State
was populated by white settlers more or less passively friendly toward
England and composed of the least hostile amongst the emigrant farmers
with a certain proportion of Englishmen.

For a time all went apparently well. Then, in 1851,

Rebellion of
Molitsane Moshesh, finding his power had been restricted by the new
arrangements, and knowing that he was much stronger in a
military sense than the British authorities had any conception of, began to
foment disturbances between his own people and native clans in the
Sovereignty. He did not appear publicly in the matter, but his policy was
none the less effective in drawing both Major Warden and the Cape
Governor into a determination to punish Molitsane—a vassal of Moshesh—
who was a distinct offender. With 162 soldiers, 120 Boers and some fifteen
hundred natives, Major Warden marched out from Bloemfontein, and at
Viervoet was drawn into a trap and suffered a disastrous defeat. It is said that
Moshesh himself was surprised at the easy result. At any rate, he at once
threw off the mask and joined forces with his vassal. A section of the Boers
also repudiated the Sovereignty Government, so far, at least, as to promise
Moshesh absolute neutrality if he would leave their cattle and property
unharmed. This he promised and fulfilled by plundering without mercy the
Boers who remained loyal. Major Warden was now helpless at
Bloemfontein, as Cape Colony was in the throes of another Kaffir war. and
not a soldier could be spared—a fact of which Moshesh and the disloyal
Dutch were perfectly aware. The latter added to the difficulties of the
situation by suggesting to Pretorius that now was his time to avenge
Boomplaatz. He was not unwilling, but thought a primary duty lay to his
own adherents beyond the Vaal; so he wrote Warden that if the independence
of the Boers of that region were definitely acknowledged he would refrain
from participation in the struggle.

The Sand River

Major Warden reported to Sir Harry Smith that the
Convention safety of the Sovereignty for the time lay in assenting to
this proposal, as he could not hold it against the Basutos
and the Transvaal Boers combined. The result was the appointment of
Commissioners and the negotiation in 1852 of the Sand River Convention
"with the Commandant and Delegates of the Boers living beyond the Vaal,"
by which the British Government "guaranteed to the emigrant farmers
beyond the Vaal River the right to manage their own affairs and to govern
themselves according to their own laws without any interference on the part
of the British Government." Provisions were included by which the British
authorities disclaimed all alliances with colored peoples north of the Vaal,
and the Boers accepted the declaration (on paper) that "no slavery is or shall
be permitted or practiced" in the country under their control. This
arrangement finally severed the two communities, carried across the Vaal
another migration of the anti-British element, and in time consolidated the
bitterly hostile and prejudiced sections of population into the present
Transvaal Republic. Meanwhile, peace had been made with the Kaffirs, and
Sir George Cathcart, who was now Governor at the Cape, invaded
Basutoland with a considerable force of regulars for the purpose of
punishing Moshesh. As usual in South African warfare, he under-estimated
the numbers and fighting skill of his opponents as well as the natural
strength of this Switzerland of the Veldt. Thaba Bosigo was too hard a nut
for his force to crack, and he was, besides, drawn into an ambush and
defeated. Moshesh, however, was wise enough not to press his advantage too
far, and with statecraft which was worthy of a greater sphere, asked and
received peace on terms very beneficial to himself.

But the Colonial Office was now in the hands of the

Changed
Policy Manchester School party, England was living in the
exhilaration of a period of great and growing commercial
prosperity, and her politicians were sick of the prolonged succession of petty
and costly wars which had marked South African history. It was decided that
all further responsibility must be avoided, that existing boundaries must be
drawn back wherever possible, and that extension of territory must be
imperatively resisted. The first point of contact with this feeling was the
Sovereignty, and the Duke of Newcastle, who was then acting as Colonial
Secretary, sent Sir George Russell Clerk out in 1853, as a Special
Commissioner: "To ascertain whether it was practicable to make
arrangements for the abandonment of the whole of that territory." Then
followed the most extraordinary and perhaps regrettable incident in all the
turbulent and troubled history of South Africa. The Commissioner had called
a Convention of European Delegates for the purpose of taking over the
government of the Sovereignty. But these twenty-four men sounded public
opinion, and they had soon found that the feeling was clear and
unmistakable that from every standpoint of right, honor and expediency
Great Britain should retain its authority and continue its protection. Sir
George Clerk, however, was under definite instructions, and any protests
from the Delegates, or from the public meetings which were hastily held,
were simply regarded as so much unnecessary obstruction to the fulfilment
of his mission. The Convention refused to accept in any way his proposition,
Formation of
and was promptly dissolved. A small body of men were
the Orange found, however, to favor independence, and with these
Free State representatives of a distinct minority Sir George concluded
an agreement on February 23, 1854, by which the country
was practically handed over to them as the Orange Free State. This precious
document "guarantees on the part of Her Majesty's Government the future
independence of that country and Government"—although it also provides
"that this independence shall, without unnecessary delay, be confirmed and
ratified by an instrument promulgated in such form and substance as Her
Majesty shall approve, finally freeing them from their allegiance to the
British Crown, and declaring them, to all intents and purposes, an
independent people." So far as can be ascertained this instrument was never
actually promulgated, and it may be a delicate technical point as to whether
the Free State people have ever been legally freed from their allegiance to
Great Britain.[1]

[1] Westminster Review. April, 1869.

Large popular gatherings were held to protest against the policy of

dismemberment, and the Chairman and another member of the late
Convention were sent to England to bring the whole case before the Queen's
Government. But it was all in vain. Hardly any notice had been taken in
Great Britain of the Sand River Convention, and even less concern was
exhibited over this new development of weak and nerveless Colonial
administration. A motion upon the subject in the House of Commons had to
be withdrawn for lack of a seconder, and Parliament voted $240,000 as a
compensation to loyal settlers—presumably as a solace for having forced
them to give up their allegiance. By the terms of the Bloemfontein
Convention—already quoted from—no slavery or trade in slaves was to be
permitted and the Government was made free to levy import duties and to
buy ammunition in the British Colonies. In this way were two Boer
Republics founded in South Africa, and the evils which might naturally have
been expected from the intense isolation and ignorance of the emigrant
farmers crystallized into constitutional shape, and finally into military form.
These Conventions of 1852 and 1854 legalized a lasting and bitter schism in
the small European population of South Africa, and even the conditions and
interests of the Free State and the Transvaal were not, for many years
afterwards, considered identical by the Boers themselves.
 CHAPTER V.

Development of Dutch Rule

From 1854 to 1877 the two Republics developed along

Development
of the Two very different lines. Their general principle of government
Republics was the same, but it was not administrated in the same way.
In form their constitutions were nominally republican; in
practice they became essentially arbitrary and absolutely antagonistic to
British and Colonial ideas of government. The coloured people who, in
hundreds of thousands, were established around the Dutch, had few civil
rights and no political ones. They were the prey of small military bodies, the
source of an enforced labour which could not in practice be distinguished
from slavery, the object of personal contempt and with little protection from
public law or private conscience. Citizenship was practically limited to the
Boer, in the Transvaal; and in the Orange Free State, through the stringent
military conditions connected with the privilege, the same result followed
for some years. The right of participating in the Government of the country
was thus confined to one class, the burghers or native-born Dutch citizens.
These alone could elect the President, the Executive Council and the
Volksraad, or popular Assembly.

There were important differences, however, in the

Important
Differences further evolution of the Republics. Something of this was
due to the modified feeling of the Orange River Boers
towards England, to their proximity to the Cape and to the fact of English
settlers being scattered amongst them with the natural result of friendly
association and occasional intermarriage. They, therefore, approximated in
character and type to the Dutchmen of Cape Colony. The Boer of the
Transvaal, on the other hand, was entirely isolated, of unmixed stock and
with sentiments of hostility toward everything British as strong and stern as
they were when he first left Colonial territory. Both Republics were allowed
to develop their own institutions in their own way and were, as the
Bloemfontein Convention of 1854 declared, "to all intents and purposes a
free and independent people." No slavery, or trade in slaves, was to be
permitted, however, and what might be termed Imperial rights of control
over native questions was retained along lines enunciated as follows, by Sir
M. E. Hicks-Beach, in a despatch dated November 20, 1879: "Neither by the
Sand River Convention of 1852, nor at any other time, did Her Majesty's
Government surrender the right and duty of requiring that the Transvaal
should be governed with a view to the common safety of the various
European communities." The same principle, of course, covered the Free
State position and, later on, was applied in connection with Moshesh and the
Basuto question.

Early
Without roads and bridges, churches and schools, or the
Organization ordinary machinery of government, the Dutch of the Free
State commenced the work of organization in 1854, and the
ultimate result reflects considerable credit upon the ignorant burghers of
those scattered communities. As in the Cape Colony and the Transvaal the
fundamental law was the old Roman system as modified by the Legislature
of Holland prior to 1652. The official language was Dutch, and the Courts
were constituted after the Dutch fashion. For a short period Josias Hoffman
was President, and then Jacobus Nicolaus Boshof was elected to the
position. Relations with Moshesh and the Basuto tribe constituted the chief
trouble of this early period. The continuous object of this ambitious ruler
was to recover certain territory which had once belonged to tribes of which
the remnants now acknowledged his rule. The Boers wished to retain regions
which had in great part appeared as wild and empty wastes when they had
settled there. Apart from the general question, both sides were aggressive
and warlike. Each hated the other, and the intermittent struggles which
ensued were of the usually merciless character. But Moshesh was too much
for the Boers in skill and craft, and, in 1858, the Free State President, after
appealing in vain to his Transvaal brethren for aid, turned to Sir George
Grey, who was then Governor of the Cape. Sir George accepted the position
of mediator, studied the situation closely, and came to the apparent
conclusion that the claims of Moshesh were in a measure just. To him,
therefore, he gave a piece of territory which the Boers believed to be theirs,
and handed over to the latter an outlying mission station which had hitherto
acknowledged Basuto authority. Mr. Boshof promptly resigned the
Presidency, and was succeeded by Marthinus Wessel Pretorius, a son of the
famous general. He devoted himself to effecting a union with the Transvaal
republics of the time, but was unsuccessful, owing to conflicting interests
and jealousies and to the declaration from Cape Town that such action would
dissolve the Conventions with Great Britain.
 Chronic
Meantime, and during the greater part of the years from
Condition of 1854 to 1868, the Boers of the Free State were in a chronic
War condition of war with the Basutos. There were few direct
conflicts, and the troubles consisted mainly in raids, the
burning of houses or kraals, the stealing of cattle, or the kidnapping of
children. The Basutos fought in much the same Fabian manner that the
Boers themselves practiced, and met invaders concealed behind rocks or
cairns or the ever-present kopje. The region ruled by Moshesh was a
compact and round-shaped territory lying between Natal, Cape Colony and
the Free State. Its surface was broken by steep hills or mountains with more
or less flat summits admirably fitted for villages or kraals, and with every
requisite for defence in the form of perpendicular wall-like sides. Between
The
these natural fortresses were the sweeping and fertile
Basutoland valleys where the Basutos grew their corn and raised their
cattle, and which for years it was the delight of the Boers to
raid; as it was the primal pleasure of the Basutos to pour down in sudden
forays from their rocky fastnesses upon Dutch territory. This constant
interchange of robbery and pillage embittered the character of both peoples,
but naturally had the most degrading effect upon that of the Boer. For a
presumably civilized and Christian race to be engaged year in and year out
in the seizure of cattle from a savage enemy and in the occasional
enslavement of children or the shooting down of stray individuals and small
parties of a mobile enemy could not but have an evil influence upon a
character so peculiar as was that of even the best and most enlightened of the
emigrant farmers.

After a decade of this sort of intermittent struggle,

Basutoland
Overrun however, the Boers were encouraged by familiarity with
that part of the Basuto country which lay in the valleys and
fields to try the task of storming some of the strongholds of the enemy. With
the aid of a few small cannon, the first attempts were successful and
surprisingly easy. Thus encouraged, within the three years following 1865,
the greater part of Basutoland was overrun and the best cornfields captured.
They were promptly "annexed" to the Free State, and then attention was
devoted to the French missionaries, who had, meanwhile, been doing a
splendid work amongst the natives. They were turned out of the country in
which half a million of dollars had been expended upon their stations; their
homes were plundered and the private property of men who had, in some
cases, been laboring for thirty years in the region was confiscated; furniture,
books and other items of value were destroyed, and all redress was refused.
Permission was afterwards given to re-occupy their stations, not as such, but
as farms for which $500 was in each case to be paid the Boer Government.
Much of the conquered territory was also surveyed and sold. But the power
of the Boers was a very fitful one. With a weak Government at home they
were unable to hold the regions which they captured from time to time, and
the result was a re-occupation by the Basutos, an attempt to cultivate their
fields, further reprisals, and more attacks upon the mountain strongholds.
Upon one occasion the Boers destroyed all the growing crops of an extensive
section. But Thaba Bosigo, the central fortress of the country, could not be
subdued by any force available.

AN ARMORED TRAIN SHELLING A BOER BATTERY AT NIGHT.

 BOERS CROSSING THE MALMANI FORD NEAR MAFEKING

In 1867 one last struggle occurred, and then Moshesh,

Basutoland
under British weakened by age and realizing that his sons were much as
Rule other natives were, and did not possess the ability to hold
the country together when his own end had come, turned to
Sir Philip Wodehouse, the Governor and High Commissioner at Cape Town,
and asked that his people be proclaimed British subjects. This was done,
partly from a wise unwillingness to have the Free State so immensely
strengthened as it would have been by the possession of Basutoland, partly
by a natural objection to have so large a number of natives dispersed over
the country without home or special object, and partly by dislike of the
policy which the Boers had been for years pursuing in regard to savages
generally and missionaries in particular. The Free Staters were intensely
annoyed. They had lost the opportunity for a lasting revenge upon their
enemy and the possibility of possessing the Switzerland of South Africa. In
the light of after events the action of Sir Philip Wodehouse seems almost
Providential, and is certainly one of the few instances where British
statecraft was really brought into play in this part of the world. Were the
Basuto strongholds in possession of Dutch sharpshooters and fortified by
German science and artillery, the struggle of 1899-1900 would be infinitely
more serious than it is at the time of writing.

"The
The Boers of the Free State bitterly resented this
Hollanders" annexation. Although now governed by the wisest
Dutchman who has come to the front in South Africa—Jan
Hendrik Brand—(afterwards better known as Sir John Brand) who had
succeeded Pretorius as President in 1865—they were also greatly influenced
by a small and compact body of men, known as Hollanders, who had
obtained possession of nearly all the offices of emolument in the State.
These Hollanders afterwards drifted largely into the Transvaal where they
had fuller and freer scope for anti-British sentiment and policy; and for
isolation from the British ideas and principles which gradually and, in the
end, powerfully, controlled the policy of President Brand. Meantime,
however, these adventurers from Holland had much influence in the Free
State. In 1858, when the Basutos had driven back the farmers and were
threatening their homes and cattle during one of the ups and downs of the
long struggle, a number of the Boers, and even some of the Hollanders, were
in favor of seeking annexation to Cape Colony, and actually a resolution to
that effect went through the Volksraad. But five years later, when fifteen
hundred and fifty signers of a memorial asked the Volksraad to press an
agitation to this end, the situation in regard to the Basutos had meanwhile
changed, and the Hollanders opposed the proposition strongly. The
movement was never seriously revived. Speaking in this connection at the
prorogation of the Cape Parliament in September, 1868, Sir Philip
Wodehouse declared that: "Entirely on my own responsibility, giving
expression only to my own opinions, I may say that I regard the measures
which severed from their allegiance the European communities in those
regions to have been founded in error."
 The Boers
This Hollander party refused to enter into any
Protest negotiation with the High Commissioner concerning the
Basutoland annexation, indulged in much talk about French
and Russian intervention, and finally despatched two Commissioners to
London armed with a long and emphatic protest. Fortunately for all
concerned, the British Government approved of the policy pursued by Sir
Philip Wodehouse and authorized him to take such further action as, to his
knowledge of local conditions, might seem desirable. This wisdom of this
course was so unusual and striking in connection with South African affairs
that a tribute of respect seems due to the Colonial Secretary of that period—
the Duke of Buckingham and Chandos. The annexation was, in fact, in the
immediate interest of the Free State as well as in the future interests of Great
Britain. It gave the exhausted republic a rest from protracted and injurious
conflict. It afforded an opportunity for the statesmanship of the new
President to assert and express itself. It facilitated the development of a
friendliness between Cape Colony and the Free State which, so long as
President Brand lived and ruled (1865-88), did much for the general good of
South Africa and something for the improvement of individual character
amongst the less implacable farmers of the little republic. There was indeed
much for a statesman to do. Ideals of Government amongst the best of the
Boers were still so crude as to be almost laughable. Masses of useless paper
money were in existence. Farms or ranches had been neglected, many cattle
destroyed and heavy debts incurred.

Just at this moment the discovery of diamonds effected a

Discovery of
Diamonds revolution in South African affairs. As this incident is
variously described by many writers, and as its importance
is so great from an historical point of view, I propose to pin my faith upon
the record given by Dr. George M. Theal. His position as a civil servant and
Historiographer to the Cape Government would, perhaps, lay the most
impartial of historians open to occasional allegations of favoritism in dealing
with annals so permeated with Dutch and English rivalry as are those of
South Africa. But there can be no question as to his accuracy in treating of
such questions of fact as this.[1] He states that: "One day, in 1867, a child on
a farm in the north of Cape Colony was observed to be playing with a
remarkably brilliant pebble, which a trader, to whom it was shown as a
curiosity, suspected to be a gem of value. It was sent for examination to a
qualified person in Grahamstown, who reported that it was a diamond of
twenty-one carats weight and that its value was £500. Search was
immediately commenced in the neighborhood by several persons in odd
hours, and soon another, though much smaller, was found. Then a third was
picked up on the bank of the Vaal River, and attention was directed to that
locality. During 1868 several were found, though as yet no one was applying
himself solely to looking for them. In March, 1869, the 'Star of South Africa'
was obtained from a Korana Hottentot, who had been in possession of it for
a long time without the least idea of its value except as a powerful charm. It
was a magnificent brilliant of eighty-three carats weight when uncut, and
was readily sold for £11,000."

[1] The Story of the Nations Series. South Africa, p. 322.

Ownership and
The lower Vaal then became the scene of a bustling,
Territorial Rulerestless and struggling population of miners and
speculators. Wealth and diamonds go together, and with
them naturally came questions of ownership and territorial rule. The latter
was and had been in dispute for many years. The southern bank of the river
was probably Free State territory, but the ownership of the northern bank
was in grave doubt. No actual government had been established there,
although the Transvaal, the Free State, the Batlapin tribe of natives, and the
Griqua captain—Waterboer—all claimed portions of the ground. There was
naturally much disorder at the mines, both north and south of the River,
under such conditions, and, finally, as the bulk of the miners were British
subjects, the High Commissioner at Cape Town decided to interfere, and
proposed a general arbitration. President Brand declined the suggestion, but
President Pretorius of the Transvaal acceded, and a Court was established at
Bloemhof, on the northern bank of the Vaal, with Mr. Keate, Governor of
Natal, as final Umpire. From the information then available there seems no
doubt that the Award issued by Mr. Keate in October, 1871, was just. He
acted, and could only act, upon the evidence presented to the Court, and, as
the Free State refused to work up or present its case, and as Waterboer was
enabled by the use of a clever advocate to prepare a fairly strong one, the
region in dispute was finally awarded to him. He had already offered his
claim to the territory to the British authorities, and, as soon as the legal
decision was announced, Sir Henry Barkly, as High Commissioner,
proclaimed the Diamond Mines and what had long been familiarly known as
Griqualand West, to be a British dependency. Afterwards, during the holding
of a special Court for the settlement of individual ground-claims, a minute
search into the history of the region south of the Vaal revealed an
unsuspected flimsiness in Waterboer's title, and the judgment of the Court
thereupon threw out all titles based upon Griqua grants. This very impartial
verdict—under all the circumstances of the case—at once gave President
Brand a position in the matter which he did not hesitate to use. He went to
London and laid his case before the British Government, which replied that
the possession of the country in question was a necessity to the paramount
Power in South Africa, but that he would be given $450,000 as a settlement
of the Free State claims. This he accepted.

The decision was as momentous in its results as the

A Momentous
Decision annexation of Basutoland. Without the possession of
Griqualand West, the British Government and settlers, and
Cape Colony itself, would have been shut off from expansion to the north.
The unclaimed country from the Limpopo to the Zambesi would have been
open to the raids and eventual occupation of the Boers of the two Republics.
The diamond mines of South Africa—with their hundreds of millions' worth
of precious stones—would have been in the hands of England's enemies as
well as the gold mines. Matabeland and Mashonaland and the empire created
by Cecil Rhodes to the north and west of the republics would have been
alien ground. The development of British South Africa would, in a word,
have been effectually confined to the limited region south of the Orange
River and the Drakensberg Mountains. The Keate Award, therefore, and the
dispute between the two Dutch Governments and that of Great Britain,
turned upon more important issues than the discovery of diamonds. The
Boers did not really want the latter, but it is fairly evident now that they fully
appreciated the importance of holding the only route to the north which still
remained open to British acquisition. Had President Brand shared in the
hostile sentiments of many of his own people and of his compatriots over the
Vaal toward Great Britain, he would never have sold his claim even for the
sum which did so much to place the finances of the Free State upon a sound
footing. From this time forward to the end of the century, however, the
Orange Free State enjoyed a condition of progressive prosperity. Roads,
public buildings and bridges were constructed. A fairly good system of
Dutch public schools was established in the villages, though it did not
Railway from
greatly affect the farmers on their wide ranches. A railway
Cape Town was run through the country from Cape Town to Pretoria,
largely at the expense of the Cape Government, while
branch lines in time connected the Free State system with Durban, in Natal,
and with Port Elizabeth and East London, on the southeast coast of Cape
Colony. President Brand was re-elected to his position until he died in 1888,
leaving the highest of reputations as a wise administrator, a warm friend of
Great Britain, and a sincere admirer of British institutions. After his time
other influences predominated, and the first evidence of this was in the
election of Mr. F. W. Reitz—previously Chief Justice of the State—as his
successor.

Condition of
Meanwhile, the Transvaal State, or South African
the Transvaal Republic as it called itself, was passing through an infinite
variety of more or less painful experiences. The region
possessed by the Boers north of the Vaal is a great tract of fairly fertile and
level land broken here and there by rugged hills. The climate is varied, but
upon the whole pleasant and healthful. Its wheat-producing capabilities are
famed throughout South Africa. Coffee and tobacco also thrive. But cattle-
raising was and is the primary pursuit of almost the entire white or Dutch
population. The Boers of this region did not arrive there all at once, or found
their State upon conditions of mutual interest and a basis of common
principles. Their one tie of union, their single basis of co-operation, was
hatred of the English. Whether trekking north from Cape Colony under
Potgieter and fighting the Matabele for a country to live in; or leaving Natal
in utter disgust at the proposed free institutions of the new British
administration; or crossing the Vaal from the Orange River Sovereignty to
escape from even friendly relations with British communities; they were,
and remained, the most implacable, the most ignorant, the most isolated and
unmanageable of the emigrant farmers. At first the Boer population
numbered only some sixteen thousand, and in 1837, after the destruction of
Moselkatze and the Matabele power on the south side of the Limpopo, an
unsuccessful attempt was made to form a common government. A little later
four republics—Pochefstroom, Zoutpansberg, Lydenburg and Utrecht—were
established, but without much effect so far as practical government was
concerned. A period of wild license followed, and was marked by much
cruelty towards the natives as well as anarchy and strife amongst the farmers
themselves.

Transvaal
In all the great region between the Orange River and the
Under Limpopo these conditions, however, prevailed between
Pretorius 1836 and 1850 to a greater or lesser degree. South of the
Vaal a check came through the vicinity of British power
and population; but north of that historic river there was little ameliorative
influence until about 1864. Marthinus Wessel Pretorius became President of
one of the Transvaal sections, or republics, in 1857, and by 1860 had united
the entire region under his control. Even then, however, there was a further
period of civil war until, in 1864, Pretorius succeeded in obtaining general
acceptance by the people and a legal election, with S. J. P. Kruger as Vice-
President. He at once resigned the Presidency of the Orange Free State,
which he had also held since 1858—but without success to his efforts at
uniting the northern and southern republics—and devoted himself to
breaking the power of the Baramapulana tribe which had established itself,
in great and growing strength, upon the southern banks of the Limpopo and
in territory which the Boers thought they should control. During more than
three succeeding years the Transvaal tried in vain to subjugate this tribe. The
State, however, had no money, and could not even pay for the transport of
ammunition from Durban, on one occasion, while its people were not united
in the prosecution of the war. The result was a practical withdrawal from the
Zoutpansberg region; a recognition of the independence of the
Baramapulana under the nominal form of a small annual tribute; and the
creation of difficulties amongst other tribes which realized the check thus
given to a people who had often oppressed them and frequently attacked
their kraals. Wars followed with the Baralong and other clans, and the
Republic presently found itself unable to assert its authority over the natives
within its claimed sphere of supremacy, or to even hold its own territory
intact. By 1870, when the Transvaal became mixed up in the Diamond Fields
controversy and entered into the arbitration resulting in the Keate Award, the
condition of the people was deplorable. The generation
Ignorance and
Isolation which was now grown up had absolutely no knowledge of
anything beyond their own family circle, and had no
acquaintance whatever with books, or history, or external affairs. The rivers
were unbridged, the Treasury was empty, the salaries of the officials were
only occasionally paid and trade was carried on by barter in the absence of
gold or silver. The natives around them could not be more densely ignorant,
or more completely isolated, than were these farmers on the veldt with all
their thriving flocks and herds and stores of grain and vegetables and fruit.
Whatever the poverty of intellect, or knowledge, or the primitive nature of
their government, there was never any lack of food and wealth of cattle
amongst the Dutch of the Transvaal. Like the Matabele and Zulu in their
days of power, the Boers always possessed these requisites of life. Yet they
would not pay taxes, or support their government, or educate their children.

Discontent and
President Pretorius was compelled to resign as a result
Disintegration of his participation in the Diamond Fields' arbitration, and
the Reverend Thomas Francois Burgers, a clergyman of
unorthodox views, who had distinguished himself as a lawyer, was elected,
in 1872, to the position. He was an able man, but somewhat visionary for the
strained situation which required his attention. He had to deal with a few
thousand ignorant men of seventeenth century views who were unable to
govern themselves, or to control the surrounding natives, and be expected
within a few years to mould out of this unpromising material a prosperous
Republic with colleges, railways, telegraphs and a great name amongst the
nations of the world. That his dreams were afterwards in a measure realized
reflects credit upon his patriotism and perspicacity; but his policy broke
down before the obstacles of the immediate present. Money to the extent of
$450,000 was obtained from Holland, which the President visited in 1874,
under authority from the Volkraad. With this sum railway material was
purchased for a proposed line from Lorenzo Marques to Pretoria, and a
Superintendent of Education was brought back to manage a system which
was not yet in existence and for the creation of which there was neither
money nor popular desire. When Mr. Burgers arrived home again he found
discontent and disintegration everywhere visible, and his educational scheme
was put aside; while his railway material was sent to rot at the Portuguese
port for want of more money to carry on the enterprise. Then the strong
Bapedi tribe under Sekukuni rose in rebellion; many of the Boers refused to
fight under an agnostic President; and a large commando which he
succeeded in getting together failed to accomplish anything and in the end
stampeded homeward. The first result of this failure was anarchy, and the
secondary consequence was the development of a situation, through the
menacing attitude of the Zulu forces upon the frontier, which brought about
annexation to the British Crown and the creation of the strictly modern phase
of the South African question.

CHAPTER VI.

Development of Cape Colony.

The dismemberment of South Africa, which commenced

Gradual
Growth of in the days of the Great Trek, which was made more
Cape Colony distinct by the Conventions of 1852-4, and was destined to
culminate in the Conventions of 1881-4, was at first
somewhat of a boon to Cape Colony. It removed about ten thousand of the
most discontented, restless and ignorant portion of its population and left
plenty of land and room for the occupation of future immigrants. They came
slowly, however, as the Kaffir wars had given the country a bad name and
the reputation of its climate was not particularly good. But, between 1845
and 1850, some five thousand British settlers were brought in under aid from
the Government, and a little later a number of Germans who had fought for
England in the Crimean war migrated to the Cape. In 1858, two thousand
German peasants were settled on lands near the southern coast of the Colony
which had once belonged to the Kaffirs. They made excellent settlers, and in
time merged with the British population, which came to predominate in the
eastern part of the country, as the Dutch did in the western section.

The Climate
The climate was found to be reasonably healthful. To
newcomers the sudden change from heat to cold, owing to
the south-east winds, was found unpleasant, and in cases of
weak constitutions somewhat dangerous. But with proper care in clothing
and gradual acclimatization this difficulty soon moderated, and the peculiar
dryness of the climate was found to make strongly for health. Sunstrokes
were rare, and the only serious evil arising from the heat was the drying up
of the rivers in the interior of the country. In most parts of the continent
malarial fever was then an admitted and serious danger, as it is to-day in the
great lake region of Central Africa and in the valley of the Nile. In German
East-Africa, in parts of the Transvaal and in the Delagoa Bay region there is
still a similar state of affairs. But Cape Colony, the Orange Free State and
Natal were then, and are at the present time, almost entirely free of this
dreaded disease. For weak lungs it was discovered that no finer country
exists in the world than the Cape, and for the development of general
healthfulness and vigour the settlers of the Colony soon found themselves in
an ideal region.

Natural
Natural resources were not quite so apparent. A wealth
Resources, etc. of brilliant flowers and tropical plants existed, but forests
were few, timber was scarce and costly, and it was years
before the introduction of the Australian Eucalyptus embowered many a
village from the Cape to Kimberley and from Buluwayo to Pretoria in groves
of that useful tree. The land in some cases was fertile, but, on the whole, was
perhaps more suited to the raising of sheep and cattle than to agriculture in
the American or Canadian sense. Farming of the latter kind involves severe
labour, and neither the original slaves, the coloured labourers of an after-
time, nor the Dutch farmers, were fitted by disposition or nature for the
work. But, as the population increased from 26,000 Europeans in 1805 to
182,000 in 1865, and to 237,000 ten years later, the country assumed a more
civilized and prosperous appearance. Sheep and cattle were literally
scattered over a thousand hills, while various collateral industries were
developed by English settlers which the slow-moving Dutch would never
have dreamed of. Between 1812 and 1820 the Merino sheep was introduced,
and its wool soon became a source of profit and wealth. In 1865 ostrich
farming was commenced, and speedily developed great importance through
the process of artificial incubation. Roads were made, churches and schools
were built, municipal government in the towns and villages was introduced,
and the Colonial finances were put into shape despite the expenses of Kaffir
wars and native troubles—which were mainly charged to the Imperial
exchequer. The first railway was constructed in 1859, and wagon roads were
carried over various mountain passes and through much of the settled part of
the country.

An Executive
In 1834 an Executive Council had been created
Council composed of members nominated by the Governor, and
therefore more or less dependent upon his good-will.
 Created Perhaps at that time, and in view of the limited population,
the racial rivalry and religious and educational
complications, it was just as well that such a body should not be elective, as
some desired. Twenty years later, however, when conditions had somewhat
changed, a representative Legislature was established composed of a Council
and a House of Assembly. Members were to be elected upon a wide
franchise, with no distinction of race or color, excepting that a Kaffir had to
hold some small amount of property and to have given up the tribal system.
There were very few natives in this condition. Meanwhile the dissensions
between the Dutch part of the population and the missionaries continued,
and they extended at times to the English settlers also. There can be no doubt
of the intense irritation aroused by this controversy. The Dutchman looked
upon the native as created and existing for his special benefit, and through
the effect of contiguity and similarity of conditions often induced the
English farmer to agree with him. The missionary, on the other hand,
believed himself appointed to guard the interests of the weaker race, and was
too apt to forget the suffering caused by Kaffir raids from the outside, in his
general sympathy for the downtrodden representatives of the race in the
Colony itself.

From about 1820 to 1860 this struggle lasted. It

A Long
Struggle weakened the hands of the Governors, who usually shared
the Colonial view of the Kaffir wars, as against the
missionaries. It injured the reputation of the Colonial Office throughout
South Africa from the widespread belief that its officials were inspired, or
guided, by the friends of the missionaries and by the impracticable
sentiments of Exeter Hall, rather than by the wishes of the people of Cape
Colony. It seriously affected the continuity of policy which should have
marked the action of the British Government, in these regions of all others,
and which, unfortunately, so seldom characterized their treatment of either
Cape Governors or native questions. In 1846 commenced the seventh Kaffir
or Kosa war. Sandili was the heir of Gaika, the Kosa chief who had figured
in a previous conflict, and he had for some time prior to this date permitted
raids upon the settlers of the Colony's eastern territory, and had entirely
disregarded pledges and arrangements. Finally, Sir Peregrine Maitland sent a
military force to occupy the region controlled by Sandili and bring him to
terms. With incomprehensible but oft-repeated carelessness in South African
warfare, a long ammunition wagon train following the expedition was left
practically unguarded, and was, of course, surprised and seized by the
A Sweeping
Kaffirs. The result of the ensuing retreat of the British
Raid troops was a combination of the Kosa and the Tembu tribes,
a sweeping raid along the entire frontier, the murder of
settlers, the capture of cattle, and the burning of dwellings. The local forces
of the Colony were hastily got together, and operations carried on in a
scattered sort of way for some months until the arrival of several British
regiments from abroad. A temporary submission was then made by the
natives with a view to the planting of their maize. As soon as this was
garnered the war broke out again.

The Governor had meantime been recalled, and was

The province
of British succeeded for a few months by Sir Henry Pottinger.
Kaffraria Sandili, however, soon had enough of the struggle, and, in
1847, peace was made after an enormous cost to the British
authorities and amid the clamor of ruined Eastern farmers. At the end of the
year Lieutenant-General Sir Harry Smith came out as Governor and High
Commissioner, with unusual personal powers and under the awakening
perception of the Colonial Office that it was better to let the man on the spot
guide affairs than to attempt the real government of South Africa from six
thousand miles away. It was not a permanent awakening, but it was useful so
long as it lasted. Sir Harry Smith adopted the repudiated native policy of Sir
Benjamin D'Urban; proclaimed the territory between the Kei and the
Keiskama as a British possession for the absolute use of the western clans of
the Kosa tribe; appointed a Commissioner to exercise general authority over
the Chiefs and sent a strong body of troops to garrison various forts; and
named the region—which once for a brief season had been called after
Queen Adelaide—the Province of British Kaffraria. A few years later the
eighth Kaffir war took place. The tribes seem to have considered the peace
as nothing more than a truce, and as soon as the British authorities began to
suppress the worst of their savage customs—notably the murders and
tortures arising out of the hunt for witchcraft—discontent very speedily
developed into the war of 1850-51. The usual struggle followed, with
surprises, raids, murders and the ravaging of the frontier. The war was the
most costly of all the conflicts with these restless tribes, and was specially
marked by an event memorable in the annals of British bravery—the loss of
H. M. S. Birkenhead with 400 soldiers on board. It occurred near Algoa Bay,
where the ship had struck a reef in the middle of the night. The women,
children and sick people were sent away safely, in all the available boats,
while the troops remained drawn up in line as though on parade, with the
ship breaking up under them and a sea swarming with sharks around them.

An
For two years a large force of soldiers, farmers and
Extraordinary auxiliaries of various kinds were employed in trying to end
Incident a war with enemies who had the fleetness of the antelope
and powers of disappearance equal to that of a bird. When
their food was exhausted, and not before, the Kosas gave in and asked for
peace. As usual in such cases, the Governor was recalled, and Sir George
Cathcart appointed his successor. The government of British Kaffraria was
reorganized and the region subdivided amongst the Tembus, a section of the
Kosa tribe under a chief named Kreli, the western clans of the Kosa and the
loyal Fingos. Several regiments of regular troops were maintained in the
Province and a body of local police formed from amongst the younger white
colonists. In 1857 there took place one of those extraordinary incidents
which can only occur in a region such as South Africa. The Kosas, prompted
by some wizard who professed to wield unknown and vast powers and to
hold communication with the unseen world, destroyed all their cattle and
stores of grain in the belief that their ancestors would, as a reward for their
faith, join them in driving the white man out of the country and in creating
for them a boundless stock of new cattle and a limitless supply of fresh
crops. Famine naturally followed, and some 30,000 natives perished of
hunger or disease despite all that Sir George Grey, who, in 1854, had
become Governor at Cape Town, could do for them in a hurried supply of
provisions and work. Some good came out of the evil. Large tracts of
depopulated land were taken possession of by European settlers, peace came
to the exhausted region, and in 1865 it was annexed to Cape Colony. It may
be added here that some small risings occurred in 1877, termed the ninth
Kaffir war, and that in 1880 the region held by the Pondos was formally
annexed to the Colony, and its borders thus became coterminous with those
of Natal.
A GENERAL VIEW OF ESTCOURT, TWENTY-FIVE MILES SOUTH
OF LADYSMITH.
GENERAL VIEW OF CITY OF LADYSMITH, NATAL (From Photo
by Henry Kisch).
 MAP SHOWING COUNTRY FROM DURBAN TO LADYSMITH

Meanwhile, the history of Cape Colony was by no

A Vexed
Question means confined to conflicts with border natives or to the
controversies with the Orange Free State, which have been
detailed in preceding pages. In 1850 occurred one ol the most striking
illustrations of what mistakes a fair-minded and well-meaning Home
Government may at times be involved in when dealing with far-away
regions. There seems to have been no perception in those days of the wrong
which might be inflicted upon a Colony by the exportation of convicts
undergoing various terms of penal servitude. Confinement in Australia or
South Africa seemed to British statesmen, and especially to Earl Grey, who
presided over the Colonial Office at this time, no more objectionable on
principle than it would be if they were kept at home in the British Isles. They
forgot that on being released these men—some punished for serious crimes,
some for slight offenses—were let loose upon a community widely scattered
and isolated and composed of many persons who, taken in this way, were
easy victims to robbery or attack. And they entirely overlooked the danger of
allowing hundreds, or in time thousands, of men without personal
responsibility or character, to roam at will amongst a large and restless
population of natives. They appear to have felt only that in the vast and
vacant spaces of the Colonies there was room and verge for a released
convict, or a ticket-of-leave man, to make for himself a new career
untrammelled by the past, or by the danger of drifting again into the deeps of
the great cities at home.

Penal
When it was understood at the Cape that the Imperial
Settlement in Government proposed to establish a penal settlement in the
the Colony Colony, similar to the one which had been formed at
Botany Bay, the indignation aroused was immediate and
intense, petitions and protests were sent in great number to London,
meetings were held throughout the Colony, and when the Neptune arrived in
Simon's Bay, Cape Town, with convicts on board, nearly all the people of the
Peninsula bound themselves together in a pledge to supply nothing to the
ship or to have any dealings with persons connected with it. Sir Harry Smith,
who was then Governor, had expressed his own strong opposition to the
plan; but he was compelled to obey his orders from home and could not
therefore send the vessel back. For five months it lay in the Harbor, supplied
from passing men-of-war and treated by the Colonists as though the plague
were within its wooden walls. And then, at last, came the order—in frank
and acknowledged response to the petitions of the Colonists—transferring its
convict cargo to Tasmania.

A New
Four years after the satisfactory settlement of this vexed
Constitution question came the grant of Parliamentary institutions to the
Colony. This action was part of a general Colonial plan by
which full responsible or ministerial government was established in Canada,
under Lord Elgin—there had long been elective legislatures in the British-
American Provinces—and a system formulated in the Australias similar to
that of the Cape. The details of the proposed changes were left by the
Colonial Office largely in the hands of the Governor and the appointive
Legislative Council, which had been created in 1834, and it was therefore
not expected that the result would be extreme in a democratic sense. The
new constitution was promulgated on March 11, 1853, and by its terms an
elective House of Assembly numbering forty-six members was created—
afterwards increased to seventy-six, and with a five years' limit in time as
against the earlier seven years period. The Upper Chamber or Legislative
Council was, to the surprise of many, also made elective. It consisted of
fifteen members, who were afterwards increased to twenty-two, with the
Chief Justice of the Colony as an additional member and ex-officio
President. The right to vote for both Houses was given to every male British
subject over twenty-one years of age who occupied a house or land worth
$125, or was in receipt of a salary or mixed remuneration valued at $250.
There was no distinction as to race, color, religion or mode of life, and this
pronounced measure of electoral liberty was a matter of constant friction in
the minds of the Dutch settlers—so far as they cared in these years to think
or trouble themselves about the affairs of an alien rule. The legislation,
however, was more important as the enunciation of a principle than because
of its working out in practice at this particular period. There were few
natives for many years in a position to take advantage of even this low
franchise, and, of course, all who continued to share in the tribal system
Right to Vote
were absolutely debarred. In 1892 the right to vote was
Limited limited by fresh legislation—resulting from the rising
political power of the Afrikander Bund and the Dutch
dislike to the natives—to such adult males as were able to sign their names
and write down their addresses and employment. The franchise qualification
was raised to a property one of $375, while the wage qualification was
allowed to remain as it had been.

The First
The first Parliament of the Colony met in June, 1854,
Parliament of and from that time onward all laws had to be sanctioned by
the Colony both Houses and approved by the Governor. As elsewhere
in the Empire the right of disallowance was reserved to the
Queen for a given period after such laws reached London, but in practice the
power was, and is, seldom used. Like so many of the apparently dormant
prerogatives of the Crown it is, however, available for an emergency.
Following this creation of Parliamentary institutions came the usual struggle
for Parliamentary control over the appointments to office, over the
expenditure of money, and over the personnel of the Governor's Council. As
in other Colonies, it was found impossible to construct in a day, or a year, an
exact imitation of Great Britain's Cabinet and governmental system, with all
its complex Parliamentary code, its elaborate constitutional checks and
counter-checks, its numerous traditions and precedents. And there was, of
course, the same difficulty as Canada had already faced and overcome—the
presence of a large electoral population with no hereditary or natural
adaptability to the British constitutional system, and without, in some cases,
the basis of cordial loyalty which is so essential to its successful operation.
At first, therefore, the officials of the Executive Council (or what afterwards
became the Ministry) were appointed by the Colonial Secretary. They
framed the financial legislation of the Government and introduced it to the
House of Assembly, and they held the right of discussion, though not of
voting, in both Houses. This system was maintained for eighteen years, and,
in view of England's heavy financial responsibilities in South Africa, the
racial condition of Cape Colony itself and the continuous troubles
everywhere with natives and Boers, it was, perhaps, as well that the threads
of government should be largely held in London. And this may be said
despite all the vacillations of the Colonial Office. Had there been firmness
and continuity in the general Home policy concerning South Africa, there
could be no question at all upon this point.

Meanwhile, Sir George Grey had been distinguishing

Wise
Administration himself by a singularly wise administration between the
years 1854 and 1859. He conciliated the Hottentots of the
Colony by granting certain claims which had been long and fruitlessly
pressed upon the authorities. He settled tor a time the native troubles in
Kaffraria, and founded a great hospital for natives, in which, by 1890, more
than 130,000 cases had been treated, and the resulting cures heralded in
many corners of "Darkest Africa" as a proof of the Englishman's power and
unexpected beneficence. He despatched troops to India at a critical period of
the Mutiny and upon his own responsibility, settled the German Legion from
the Crimea in the Colony, and brought out a number of German families for
its members to marry into. Finally, during his first Governorship, he urged
the union of the Legislatures of the Cape, Natal and Orange Free State in a
common federal system, and at a time when the Free State might easily have
been persuaded to accept the policy. But the Colonial Office would have
none of it. Unfortunately, and to the lasting injury of South Africa, the Home
Government distrusted him, and in 1858 he was recalled.

Sir George
The Derby Administration, however, met with defeat
Grey while Sir George Grey was on the sea, and when he
Reappointed reached London it was to find that he had been reappointed
to his position. It long afterwards became known that this
was done by the personal command of the Queen, who had appreciated the
policy he pursued and had sympathized with his proposed federal scheme.[1]
But despite this fact the new Government, as a whole, was so strongly
opposed to the much-feared increase of responsibilities, under a federation in
South Africa, that Sir George Grey was obliged to forego the hope of even
attempting to carry his schema further. During his second administration,
which only lasted until 1861, he entertained Prince Alfred (the Duke of
Edinburgh), and traversed with him a great part of Cape Colony, Kaffraria
and Natal; improved to an immense extent the splendid natural Harbor at
Cape Town; visited the Orange Free State and established at Bloemfontein,
as a token of friendship, the Grey Institute, in which so much has since been
done for the higher education of the youth of that State. In
Annexation of
Basutoland 1861 he accepted the Governorship of New Zealand, and
was succeeded by Sir P. E. Wodehouse, whose
administration was chiefly distinguished for the annexation of Basutoland. In
1870 Sir Henry Barkly took charge of affairs and assumed possession for
Great Britain of the Diamond Fields. With the coming of Sir Bartle Frere, in
1877, arose new developments along the lines of Sir George Grey's
disappointed hopes and hampered policy. This time, however, a check was to
be given from within the Colony instead of by the Colonial Office. The
wheel of fate refused to reverse itself.

[1] Life and Times of Sir George Grey. By W. L. Rees. London, 1892.
Vol. XI., p. 298.
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

textbookfull.com