Visit https://ebookluna.

com to download the full version and

explore more ebooks

Data-Driven and Model-Based Methods for Fault

Detection and Diagnosis 1st Edition - eBook PDF

_____ Click the link below to download _____

https://ebookluna.com/download/data-driven-and-model-
based-methods-for-fault-detection-and-diagnosis-ebook-
pdf/

Explore and download more ebooks at ebookluna.com

Here are some recommended products that might interest you.
You can download now and explore!

(eBook PDF) Bayesian Networks in Fault Diagnosis: Practice

and Application

https://ebookluna.com/product/ebook-pdf-bayesian-networks-in-fault-
diagnosis-practice-and-application/

ebookluna.com

Optimal State Estimation for Process Monitoring, Fault

Diagnosis and Control 1st Edition Ch. Venkateswarlu M.Tech
Phd - eBook PDF
https://ebookluna.com/download/optimal-state-estimation-for-process-
monitoring-fault-diagnosis-and-control-ebook-pdf/

ebookluna.com

Data-Driven Traffic Engineering: Understanding of Traffic

and Applications Based on Three-Phase Traffic Theory 1st
Edition - eBook PDF
https://ebookluna.com/download/data-driven-traffic-engineering-
understanding-of-traffic-and-applications-based-on-three-phase-
traffic-theory-ebook-pdf/
ebookluna.com

Data Driven: Harnessing Data and AI to Reinvent Customer

Engagement 1st Edition Tom Chavez - eBook PDF

https://ebookluna.com/download/data-driven-harnessing-data-and-ai-to-
reinvent-customer-engagement-ebook-pdf/

ebookluna.com
(eBook PDF) Model Driven Development for Embedded
Software: Application to Communications for Drone Swarm

https://ebookluna.com/product/ebook-pdf-model-driven-development-for-
embedded-software-application-to-communications-for-drone-swarm/

ebookluna.com

Quantitative systems pharmacology : models and model-based

systems with applications 1st Edition - eBook PDF

https://ebookluna.com/download/quantitative-systems-pharmacology-
models-and-model-based-systems-with-applications-ebook-pdf/

ebookluna.com

Highway Safety Analytics and Modeling: Techniques and

Methods for Analyzing Crash Data 1st edition - eBook PDF

https://ebookluna.com/download/highway-safety-analytics-and-modeling-
techniques-and-methods-for-analyzing-crash-data-ebook-pdf/

ebookluna.com

Data-driven Solutions to Transportation Problems 1st

Edition- eBook PDF

https://ebookluna.com/download/data-driven-solutions-to-
transportation-problems-ebook-pdf/

ebookluna.com

Urgent Care Dermatology: Symptom-Based Diagnosis 1st

Edition - eBook PDF

https://ebookluna.com/download/urgent-care-dermatology-symptom-based-
diagnosis-ebook-pdf/

ebookluna.com
DATA-DRIVEN
AND
MODEL-BASED
METHODS FOR
FAULT DETECTION
AND DIAGNOSIS
DATA-DRIVEN
AND
MODEL-BASED
METHODS FOR
FAULT DETECTION
AND DIAGNOSIS

MAJDI MANSOURI
MOHAMED-FAOUZI HARKAT
HAZEM N. NOUNOU
MOHAMED N. NOUNOU
Elsevier
Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
Copyright © 2020 Elsevier Inc. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, recording, or any information storage and
retrieval system, without permission in writing from the publisher. Details on how to seek
permission, further information about the Publisher’s permissions policies and our arrangements
with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency,
can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the
Publisher (other than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and experience
broaden our understanding, changes in research methods, professional practices, or medical
treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in
evaluating and using any information, methods, compounds, or experiments described herein. In
using such information or methods they should be mindful of their own safety and the safety of
others, including parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors,
assume any liability for any injury and/or damage to persons or property as a matter of products
liability, negligence or otherwise, or from any use or operation of any methods, products,
instructions, or ideas contained in the material herein.

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

ISBN: 978-0-12-819164-4

For information on all Elsevier publications

visit our website at https://www.elsevier.com/books-and-journals

Publisher: Susan Dennis

Acquisitions Editor: Anita Koch
Editorial Project Manager: Sara Valentino
Production Project Manager: Vignesh Tamil
Designer: Miles Hitchen
Typeset by VTeX
Contents

List of figures ix
List of tables xvii
About the authors xix
Acknowledgments xxi
List of acronyms xxiii
Nomenclature xxv

1. Introduction 1
References 6

2. PCA and PLS-based generalized likelihood ratio for fault detection 11

2.1. PCA and PLS-based generalized likelihood ratio for fault detection 12
2.1.1. Introduction 12
2.1.2. Principal component analysis (PCA) 13
2.1.3. Fault detection using PCA method 16
2.1.4. Statistical hypothesis testing 18
2.1.5. Fault detection using a PCA-based GLRT 21
2.1.6. PCA-based GLRT and applications 23
2.1.7. Conclusion 36
2.2. PLS-based generalized likelihood ratio for fault detection 37
2.2.1. Introduction 37
2.2.2. Partial Least Square (PLS) method 38
2.2.3. PLS-based GLRT for fault detection 40
2.2.4. PLS-based GLRT fault detection and applications 41
2.2.5. Conclusions 43
References 45

3. Kernel PCA- and Kernel PLS-based generalized likelihood ratio

tests for fault detection 49
3.1. Kernel PCA-based generalized likelihood ratio test for fault detection 49
3.1.1. Introduction 49
3.1.2. Kernel Principal Component Analysis (KPCA) description 50
3.1.3. Fault detection using KPCA method 53
3.1.4. Enhanced monitoring using kernel GLRT chart 54
3.1.5. Kernel GLRT fault detection chart with applications 57
3.1.6. Conclusion 62
3.2. Kernel PLS-based generalized likelihood ratio test for fault detection 63
3.2.1. Introduction 63
3.2.2. Kernel Partial Least Squares (KPLS) method 65

v
vi Contents

3.2.3. KPLS-based GLRT and application to fault detection in CSTR

process 69
3.2.4. Conclusion 73
References 75

4. Linear and nonlinear multiscale latent variable-based

generalized likelihood ratio for fault detection 79
4.1. Linear multiscale latent variable-based generalized likelihood ratio for
fault detection 79
4.1.1. Introduction 79
4.1.2. Multiscale PCA-based GLRT for fault detection 80
4.1.3. Multiscale PLS-based GLRT for fault detection 98
4.1.4. Conclusions 112
4.2. Multiscale nonlinear latent variable-based generalized likelihood ratio
test for fault detection 113
4.2.1. Introduction 113
4.2.2. Multiscale kernel PCA-based GLRT for fault detection 116
4.2.3. Multiscale kernel PLS-based GLRT for fault detection 122
4.2.4. Conclusion 132
References 132

5. Linear and nonlinear interval latent variable approaches for fault

detection 135
5.1. Interval latent variable approaches for fault detection 136
5.1.1. Introduction 136
5.1.2. Interval PCA-based GLRT for fault detection 137
5.1.3. Interval PLS-based GLRT for fault detection 170
5.1.4. Conclusion 185
5.2. Interval nonlinear latent variable approaches for fault detection 188
5.2.1. Introduction 188
5.2.2. Interval kernel PCA-based GLRT for fault detection 190
5.2.3. Interval kernel PLS-based GLRT for fault detection 204
5.2.4. Conclusion 214
References 215

6. Model-based approaches for fault detection 221

6.1. Introduction 221
6.2. State estimation 225
6.2.1. State estimation problem formulation 225
6.2.2. State estimation techniques 226
6.3. Fault detection-based state estimation approaches 230
6.3.1. Fault detection using multiscale EWMA chart 230
6.3.2. Application to wastewater treatment plant 233
 Contents vii

6.4. Fault detection-based state estimation approach 240

6.4.1. Fault detection using optimized weighted SS-DEWMA chart 240
6.4.2. Optimized WSS-DEWMA and application to fault detection 245
6.5. Conclusions 252
References 255

7. Conclusions and perspectives 259

7.1. Conclusions 259
7.2. Perspectives and research proposals 260
7.2.1. Project 1: water distribution networks: modeling, sensor
placement, leak and quality monitoring 260
7.2.2. Project 2: enhanced operation of wastewater treatment plants 266
7.2.3. Project 3: enhanced monitoring of photovoltaic systems 270
7.2.4. Project 4: enhanced data validation of an air quality monitoring
networks 273

Appendix 279
References 286
Index 289
List of figures

Figure 2-1 Geometric Interpretation of PCA. 17

Figure 2-2 Principle of PCA projection. 17
Figure 2-3 Outliers detection using PCA. 17
Figure 2-4 A schematic diagram of the PCA-based GLRT fault detection
method. 22
Figure 2-5 (A) Quarter-hourly ozone measurements and (B) ACF of ozone
measurements. 25
Figure 2-6 Variance captured by each principal component. 26
Figure 2-7 PC2 versus PC1. 26
Figure 2-8 PCA model prediction of ozone concentrations for the first
four network stations. 27
Figure 2-9 PCA model prediction of ozone concentrations for the last
three network stations. 28
Figure 2-10 Monitoring a simple sensor fault using PCA-based Q. 30
Figure 2-11 Monitoring a simple sensor fault using PCA-based GLRT. 30
Figure 2-12 Monitoring a multiple faults using PCA-based Q. 31
Figure 2-13 Monitoring a multiple faults using PCA-based GLRT. 31
Figure 2-14 The variance captured by each principal component. 32
Figure 2-15 Histograms showing the normality of the residuals. 33
Figure 2-16 Monitoring a fault in CA using PCA-based Q. 34
Figure 2-17 Monitoring a fault in CA using PCA-based GLRT. 34
Figure 2-18 Monitoring a fault in T using PCA-based Q. 35
Figure 2-19 Monitoring a fault in T using PCA-based GLRT. 35
Figure 2-20 Monitoring of multiple faults in CA and T using PCA-based Q. 36
Figure 2-21 Monitoring of multiple faults in CA and T using PCA-based
GLRT. 36
Figure 2-22 Correct detection rate vs false alarm rate for the PCA-based
GLRT fault detection method and the conventional PCA
method. 37
Figure 2-23 The time evolution of the generated data X . 42
Figure 2-24 Monitoring fault in temperature using PLS-based Q chart. 42
Figure 2-25 Monitoring fault in temperature using PLS-based GLRT chart. 43
Figure 2-26 Monitoring TEP IDV 2 fault using PLS-based Q chart. 43
Figure 2-27 Monitoring TEP IDV 2 fault using PLS-based GLRT chart. 44
Figure 3-1 Schematic illustration of KGLRT algorithm. 58
Figure 3-2 Evolutions of SPE, KGLRT, and EWMA-KGLRT in faulty case. 59
Figure 3-3 Evolutions of SPE, KGLRT, and EWMA-KGLRT in faulty case. 60

ix
x List of figures

Figure 3-4 The time evolution of the SPE (A), KGLRT (B) and
EWMA − KGLRT (C) statistics on a semi-logarithmic scale
in the presence of a single fault in T . 61
Figure 3-5 The time evolution of the SPE (A), KGLRT (B) and
EWMA − KGLRT (C) statistics on a semi-logarithmic scale
in the presence of simultaneous faults in CA and T . 62
Figure 3-6 KPLS diagram for nonlinear regression. 67
Figure 3-7 The time evolution of PLS-based Q statistic on a
semilogarithmic scale in the presence of a fault in CA . 69
Figure 3-8 The time evolution of KPLS-based Q statistic on a
semilogarithmic scale in the presence of a fault in CA . 70
Figure 3-9 The time evolution of PLS-based GLRT statistic on a
semilogarithmic scale in the presence of a fault in CA . 70
Figure 3-10 The time evolution of KPLS-based GLRT statistic on a
semilogarithmic scale in the presence of a fault in CA . 71
Figure 3-11 The time evolution of PLS-based Q statistic on a
semilogarithmic scale in the presence of a fault in T . 71
Figure 3-12 The time evolution of KPLS-based Q statistic on a
semilogarithmic scale in the presence of a fault in T . 71
Figure 3-13 The time evolution of PLS-based GLRT statistic on a
semilogarithmic scale in the presence of a fault in T . 72
Figure 3-14 The time evolution of KPLS-based GLRT statistic on a
semilogarithmic scale in the presence of a fault in T . 72
Figure 3-15 The time evolution of PLS-based Q statistic on a
semilogarithmic scale in the presence of simultaneous faults
in CA and T . 73
Figure 3-16 The time evolution of KPLS-based Q statistic on a
semilogarithmic scale in the presence of simultaneous faults
in CA and T . 73
Figure 3-17 The time evolution of PLS-based GLRT statistic on a
semilogarithmic scale in the presence of simultaneous faults
in CA and T . 74
Figure 3-18 The time evolution of KPLS-based GLRT statistic on a
semilogarithmic scale in the presence of simultaneous faults
in CA and T . 74
Figure 4-1 A schematic diagram of data representation at multiple
scales [1]. 81
Figure 4-2 Schematic illustration of MSPCA model. 83
Figure 4-3 Schematic illustration of proposed MSPCA-based MW-GLRT
algorithm. 86
 Visit https://ebookluna.com
now to explore a diverse
collection of ebooks available
in formats like PDF, EPUB, and
MOBI, compatible with all
devices. Don’t miss the chance
to enjoy exciting offers and
quickly download high-quality
materials in just a few simple
steps!
 List of figures xi

Figure 4-4 The effect of the window length on the performance of

the moving-window GLRT (the fault size equals twice the
residuals standard deviation). (A) Missed detection rate, (B)
false alarm rate and (C) average run length (ARL) vs. window
length. 88
Figure 4-5 Monitoring a fault of magnitude unity using PCA-based T 2
chart. 89
Figure 4-6 Monitoring a fault of magnitude unity using MSPCA-based
T 2 chart. 90
Figure 4-7 Monitoring a fault of magnitude unity using PCA-based Q
chart. 90
Figure 4-8 Monitoring a fault of magnitude unity using MSPCA-based Q
chart. 90
Figure 4-9 Monitoring a fault of magnitude unity using MSPCA-based
GLRT chart. 91
Figure 4-10 Monitoring a fault of magnitude unity using MSPCA-based
MW-GLRT chart (WL = 4). 91
Figure 4-11 Monitoring a fault of magnitude unity using MSPCA-based
MW-GLRT chart (WL = 8). 91
Figure 4-12 Monitoring a fault of magnitude 1σ using PCA-based T 2 chart. 92
Figure 4-13 Monitoring a fault of magnitude 1σ using MSPCA-based T 2
chart. 92
Figure 4-14 Monitoring a fault of magnitude 1σ using PCA-based Q chart. 92
Figure 4-15 Monitoring a fault of magnitude 1σ using MSPCA-based Q
chart. 93
Figure 4-16 Monitoring a fault of magnitude 1σ using MSPCA-based
GLRT chart. 93
Figure 4-17 Monitoring a fault of magnitude 1σ using MSPCA-based
MW-GLRT chart (WL = 4). 93
Figure 4-18 Monitoring a fault of magnitude 1σ using MSPCA-based
MW-GLRT chart (WL = 8). 94
Figure 4-19 Monitoring TEP fault 12 using PCA-based T 2 chart. 101
Figure 4-20 Monitoring TEP fault 12 using MSPCA-based T 2 chart. 102
Figure 4-21 Monitoring TEP fault 12 using PCA-based Q chart. 102
Figure 4-22 Monitoring TEP fault 12 using MSPCA-based Q chart. 102
Figure 4-23 Monitoring TEP fault 12 using MSPCAbased GLRT chart. 103
Figure 4-24 Monitoring TEP fault 12 using MSPCA-based MW-GLRT chart
(WL = 4). 103
Figure 4-25 Monitoring TEP fault 12 using MSPCA-based MW-GLRT chart
(WL = 8). 103
xii List of figures

Figure 4-26 Monitoring a fault of magnitude unity using the

MSPCA-based MW-GLRT (WL = 2) chart. 104
Figure 4-27 Monitoring a fault of magnitude unity using the
MSPCA-based MW-GLRT (WL = 4) chart. 104
Figure 4-28 Monitoring a fault of magnitude unity using the
MSPCA-based MW-GLRT (WL = 8) chart. 104
Figure 4-29 Monitoring a fault of magnitude unity using the
MSPCA-based EWMA-GLRT chart. 105
Figure 4-30 Monitoring a fault of magnitude 1σ using the MSPCA-based
MW-GLRT (WL = 2) chart. 105
Figure 4-31 Monitoring a fault of magnitude 1σ using the MSPCA-based
MW-GLRT (WL = 4) chart. 105
Figure 4-32 Monitoring a fault of magnitude 1σ using the MSPCA-based
MW-GLRT (WL = 8) chart. 106
Figure 4-33 Monitoring a fault of magnitude 1σ using the MSPCA-based
EWMA-GLRT chart. 106
Figure 4-34 Representation of MSPLS fault detection model. 107
Figure 4-35 Monitoring multiple faults in temperature using PLS-based Q
chart. 109
Figure 4-36 Monitoring multiple faults in temperature using MSPLS-based
Q chart. 110
Figure 4-37 Monitoring multiple faults in temperature using PLS-based
GLRT chart. 110
Figure 4-38 Monitoring multiple faults in temperature using MSPLS-based
GLRT chart. 111
Figure 4-39 Monitoring TEP IDV 2 fault using PLS-based Q chart. 113
Figure 4-40 Monitoring TEP IDV 2 fault using MSPLS-based Q chart. 114
Figure 4-41 Monitoring TEP IDV 2 fault using PLS-based GLRT chart. 114
Figure 4-42 Monitoring TEP IDV 2 fault using MSPLS-based GLRT chart. 115
Figure 4-43 Representation of MSKPCA fault detection model. 116
Figure 4-44 Evolutions of SPE, KGLRT, and MS-KGLRT in faulty case. 118
Figure 4-45 Evolutions of SPE, KGLRT, and MS-KGLRT in faulty case. 118
Figure 4-46 Evolutions of SPE, KGLRT, and MS-KGLRT in faulty case. 119
Figure 4-47 Monitoring faults (A) IDV1 , (B) IDV2 , (C) IDV3 and (D) IDV4
using SPE, KGLRT, and MS-KGLRT charts. 121
Figure 4-48 Monitoring faults (A) IDV5 , (B) IDV6 , (C) IDV7 and (D) IDV8
using SPE, KGLRT, and MS-KGLRT charts. 122
Figure 4-49 Monitoring faults (A) IDV9 , (B) IDV10 , (C) IDV11 and (D) IDV12
using SPE, KGLRT, and MS-KGLRT charts. 123
Figure 4-50 Monitoring faults (A) IDV13 , (B) IDV14 , (C) IDV15 and (D) IDV16
using SPE, KGLRT, and MS-KGLRT charts. 124
 List of figures xiii

Figure 4-51 Monitoring faults (A) IDV17 , (B) IDV18 , (C) IDV19 , (D) IDV20 and
(E) IDV21 using SPE, KGLRT, and MS-KGLRT charts. 125
Figure 4-52 Representation of MSKPLS fault detection model. 126
Figure 4-53 The time evolution of the generated data X . 128
Figure 4-54 Time evolution of detection using PLS-based GLRT method. 128
Figure 4-55 Time evolution of detection using KPLS-based GLRT method. 129
Figure 4-56 Time evolution of detection using MSPLS-based GLRT method. 129
Figure 4-57 Time evolution of detection using MSKPLS-based GLRT
method. 130
Figure 4-58 Time evolution of detection using KPLS-based GLRT method. 131
Figure 4-59 Time evolution of detection using MSKPLS-based GLRT
method. 131
Figure 5-1 Time evolution of simulated interval-valued data. 150
Figure 5-2 Measurements and estimations of interval-valued data using
CIPCA model. 151
Figure 5-3 Time evolution of univariate interval GLR. 151
Figure 5-4 Time evolution of univariate interval weighted GLR. 152
Figure 5-5 Time evolution of multivariate GLR. 152
Figure 5-6 Time evolution of multivariate interval weighted GLR. 152
Figure 5-7 Distillation column interval-valued measurements. 153
Figure 5-8 Evolution of the VIRE with respect to the number of principal
components . 154
Figure 5-9 Measurements and estimations. 154
Figure 5-10 Time evolution of univariate interval GLR. 155
Figure 5-11 Time evolution of univariate interval weighted GLR. 155
Figure 5-12 Time evolution of multivariate interval GLR. 155
Figure 5-13 Time evolution of multivariate interval weighted GLR. 156
Figure 5-14 Ozone concentrations for single-valued and interval-valued
data. 157
Figure 5-15 Evolution of VIRE with respect to the number of principal
components. 157
Figure 5-16 Measurements and estimations of O3 station 1. 158
Figure 5-17 Measurements and estimations of O3 station 3. 158
Figure 5-18 Time evolution of univariate interval GLR with a fault on x7 . 159
Figure 5-19 Time evolution of univariate interval weighted GLR with a
fault on x7 . 159
Figure 5-20 Time evolution of multivariate interval GLR with a fault on x7 . 159
Figure 5-21 Time evolution of multivariate interval weighted GLR with a
fault on x7 . 160
Figure 5-22 Time evolution of simulated data. 163
xiv List of figures

Figure 5-23 The time evolution of the MRPCA-based SPE statistic in the
presence of faults in x2 and x3 . 165
Figure 5-24 The time evolution of the MRPCA-based EWMA statistic in the
presence of faults in x2 and x3 . 165
Figure 5-25 The time evolution of the MRPCA-based GLRT statistic in the
presence of faults in x2 and x3 . 166
Figure 5-26 The time evolution of the MRPCA-based EWMA statistic in the
presence of faults in x2 and x3 . 166
Figure 5-27 Measurements and estimations of O3 station 1. 167
Figure 5-28 Measurements and estimations of O3 station 3. 167
Figure 5-29 The time evolution of the MRPCA-based SPE statistic in the
presence of faults in O3 . 168
Figure 5-30 The time evolution of the MRPCA-based EWMA statistic in the
presence of faults in x2 and x3 . 168
Figure 5-31 The time evolution of the MRPCA-based GLRT statistic in the
presence of faults in O3 . 169
Figure 5-32 The time evolution of the MRPCA-based EWMA statistic in the
presence of faults in O3 . 169
Figure 5-33 Time evolution of interval-valued simulated variables. 177
Figure 5-34 Scatter plots of predicted and observed training data y1 . 177
Figure 5-35 Scatter plots of predicted and observed training data y2 . 178
Figure 5-36 Evolution of Q1,x in both fault-free and faulty cases. 178
Figure 5-37 Evolution of Q1,y in both fault-free and faulty cases. 179
Figure 5-38 Evolution of Q2,x in both fault-free and faulty cases. 179
Figure 5-39 Evolution of Q2,y in both fault-free and faulty cases. 180
Figure 5-40 Evolution of Q3,x in both fault-free and faulty cases. 180
Figure 5-41 Evolution of Q3,y in both fault-free and faulty cases. 181
Figure 5-42 Evolution of Q4,x in both fault-free and faulty cases. 181
Figure 5-43 Evolution of Q4,y in both fault-free and faulty cases. 182
Figure 5-44 Evolution of I Gx in both fault-free and faulty cases. 182
Figure 5-45 Evolution of I Gy in both fault-free and faulty cases. 183
Figure 5-46 Distillation column interval-valued measurements. 184
Figure 5-47 Evolution of Q1,x and Q1,y with a fault on variable x2 . 184
Figure 5-48 Evolution of Q2,x and Q2,y with a fault on variable x2 . 185
Figure 5-49 Evolution of Q3,x and Q3,y with a fault on variable x2 . 185
Figure 5-50 Evolution of Q4,x and Q4,y with a fault on variable x2 . 186
Figure 5-51 Evolution of IGLRx and IGLRy with a fault on variable x2 . 186
Figure 5-52 Evolution of Q4,x and Q4,y with a fault on variable y2 . 187
Figure 5-53 Evolution of IGLRx and IGLRy with a fault on variable y2 . 188
Figure 5-54 3-D scatter plot of the generated interval-valued data. 199
 List of figures xv

Figure 5-55 Time evolution of indices T 2 and SPE in KPCA model with a
fault on variable x1 . 200
Figure 5-56 Time evolution of indices T 2 and SPE in IKPCACR model
with a fault on variable x2 . 200
Figure 5-57 Time evolution of x3 . 201
Figure 5-58 Time evolution of indices T 2 and SPE in KPCA model with
IDV-1 fault. 202
Figure 5-59 Time evolution of indices T 2 and SPE in IKPCA model with
IDV-1 fault. 203
Figure 5-60 Time evolution of indices T 2 and SPE in IKPCA model with
IDV-1 fault. 203
Figure 5-61 Time evolution of indices IGLRTUL in IKPCA model with IDV-1
fault. 204
Figure 5-62 Time evolution of indices IGLRTCR in IKPCA model with IDV-1
fault. 204
Figure 5-63 Time evolution of univariate GLRT index based on IKPLSUL
model. 211
Figure 5-64 Time evolution of multivariate GLRT index based on IKPLSUL
model. 211
Figure 5-65 Time evolution of univariate GLRT index based IKPLSCR model. 213
Figure 5-66 Time evolution of multivariate GLRT index based on IKPLSCR
model. 213
Figure 6-1 Plots of samples of normal and faulty signals. 222
Figure 6-2 General flow-chart of the multiobjective optimization process. 232
Figure 6-3 Multiscale EWMA strategy. 233
Figure 6-4 PF-based MS-EWMA fault detection strategy. 234
Figure 6-5 State estimation of the variables (A) XDCO , (B) SO and (C)
XBH using UKF and PF. 236
Figure 6-6 State estimation of the variables (A) SNH , (B) SNO and (C)
XBA using UKF and PF. 237
Figure 6-7 Monitoring a bias fault in SO using (A) Shewhart, (B) EWMA,
and (C) MS-EWMA methods. 239
Figure 6-8 Monitoring a drift fault in SO using (A) Shewhart, (B) EWMA,
and (C) MS-EWMA methods. 241
Figure 6-9 Monitoring a fault in X using (A) EWMA, (B) SS-DEWMA, and
(C) OWSS-DEWMA charts. 247
Figure 6-10 Qualitative model of the CSEC (simplified). 248
Figure 6-11 Estimation of state variables using various state estimation
techniques. 249
Figure 6-12 Monitoring a multiple faults in cadaverine Cadav using (A)
EWMA, (B) SS-DEWMA and (C) OWSS-DEWMA charts. 251
xvi List of figures

Figure 6-13 Monitoring a fault in Cadav using (A) EWMA, (B) SS-DEWMA
and (C) OWSS-DEWMA charts. 253
Figure 6-14 Monitoring a multiple faults in Cadav and Lys using (A)
EWMA, (B) SS-DEWMA and (C) OWSS-DEWMA charts. 254
Figure 1 Tennessee Eastman process. 279
Figure 2 Basic distillation column controlled with LV-configuration. 283
Figure 3 Ozone concentrations for the first three stations. 285
Figure 4 Ozone concentrations for the first station, single-valued and
interval-valued representations. 285
List of tables

Table 2-1 Summary of MDRs (%), FARs (%) and ARL1 . 42

Table 2-2 Missed detection rates (MDR %), False alarm rates (FAR %),
and ARL1 values for TEP data. 44
Table 3-1 FAR %, MDR %, and ARL1 for the presented fault detection
charts (simulated example). 60
Table 3-2 FAR % and MDR % for the presented fault detection charts
(CSTR process) for a single fault. 63
Table 3-3 FAR % and MDR % for the presented fault detection charts
(CSTR process) for multiple faults. 63
Table 4-1 Summary of missed detection (%), false alarms (%), and
ARL1 for simulated data using PCA and MSPCA models. 95
Table 4-2 Summary of missed detection rates (%) for TEP data using
PCA and MSPCA. 96
Table 4-3 Summary of missed detection rates (%) for TEP data using
MSPCA-based GLRT and MW-GLRT. 97
Table 4-4 Summary of false alarm rates (%) for TEP data using PCA and
MSPCA. 98
Table 4-5 Summary of false alarm rates (%) for TEP data using
MSPCA-based GLRT and MW-GLRT. 99
Table 4-6 Summary of ARL1 for TEP data using PCA and MSPCA. 100
Table 4-7 Summary of ARL1 for TEP data using MSPCA-based GLRT and
MW-GLRT. 101
Table 4-8 Summary of missed detection (%), false alarms (%), and ARL1
for simulated data using MSPCA. 106
Table 4-9 Summary of MDR (%), FAR (%), and ARL1 . 110
Table 4-10 Missed detection rates (%) for TEP data. 111
Table 4-11 False alarm rates (%) for TEP data. 112
Table 4-12 ARL1 values for TEP data. 113
Table 4-13 FAR %, MDR %, and ARL1 for the presented fault detection
charts (simulated example). 118
Table 4-14 FAR %, MDR %, and ARL1 for the presented fault detection
charts (CSTR process). 119
Table 4-15 Summary of MDR, FAR, and ARL1 values for TEP data. 120
Table 4-16 Summary of missed detection rate (%), false alarm rate (%),
and ARL1 . 130
Table 4-17 Summary of missed detection rate (%), false alarm rate (%),
and ARL1 (CSTR process). 132

xvii
xviii List of tables

Table 5-1 VIRE of different interval-valued for CIPCA model. 150

Table 5-2 FAR, MDR, and ARL1 . 153
Table 5-3 FAR, MDR rates, and ARL1 for the distillation column. 156
Table 5-4 FAR, MDR, and ARL1 for air quality data. 160
Table 5-5 MSE using CPCA, CIPCA, and MRPCA models. 164
Table 5-6 Summary of missed detection (MDR %), false alarms (FAR%)
and ARL1 . 166
Table 5-7 Summary of missed detection (MDR %), false alarms (FAR %),
and ARL1 . 169
Table 5-8 FAR % and MDR % for the presented fault detection charts
(simulated example). 183
Table 5-9 FAR % and MDR % for the presented fault detection charts
(scenario 1 of distillation column). 187
Table 5-10 FAR % and MDR % for the presented fault detection charts
(scenario 2 of distillation column). 188
Table 5-11 Summary of MDR and FAR values for the simulation example. 201
Table 5-12 Missed detection rate (MDR %) and False Alarm Rate (FAR %)
values for TEP data sets. 205
Table 5-13 Selected monitored variables in the TE process. 210
Table 5-14 Selected output measured variables in the TE process. 211
Table 5-15 Missed detection rate (MDR %) and false alarm rate (FAR %)
values for TEP data sets using IKPCA-based GLRT approach. 214
Table 5-16 Missed detection ratio (MDR %)-based Q and GLRT for the 21
faults of the TEP. 215
Table 6-1 Comparison of the MSE for the UKF and PF techniques. 238
Table 6-2 Summary of MDR (%) and FAR (%) for bias fault. 240
Table 6-3 Summary of MDR (%) and FAR (%) for drift fault. 242
Table 6-4 Summary of MDR (%) and FAR (%) for different values of s. 242
Table 6-5 Summary of MDR (%) and FAR (%) for different values of a. 242
Table 6-6 MDR (%) and FAR (%) evaluation. 248
Table 6-7 CSEC parameters. 249
Table 6-8 RMSE of estimated states using EKF, UKF, and PF methods. 250
Table 6-9 MDR (%) and FAR (%) evaluation. 252
Table 6-10 MDR (%) and FAR (%) evaluation. 252
Table 6-11 MDR (%) and FAR (%) evaluation. 255
Table 1 Manipulated variables. 280
Table 2 Measured variables. 281
Table 3 Process faults of TEP. 282
Table 4 Distillation column process variables. 284
About the authors

Majdi Mansouri Dr. Majdi Mansouri received the engineering degree in

Electrical Engineering in 2006 from the Higher School of Communication
of Tunisia (SUPCOM), Tunisia. He received his master degree of Electrical
Engineering from the School of Electronics, Informatics and Radiocom-
munications in Bordeaux (ENSEIRB), France, in 2008. He received his
PhD degree of Electrical Engineering from the University of Technology
of Troyes (UTT), France, in 2011. In December 2019, he received the
degree of HDR (Accreditation To Supervise Research) of Applied Mathe-
matics and Statistics for Electrical Engineering from University of Orleans
in France. He joined the Electrical Engineering Program at Texas A&M
University at Qatar, in 2011, where he is currently an Associate Research
Scientist.
He has over ten years of research and practical experience in systems
engineering and signal processing. His work focuses on the utilization
of applied mathematics and statistics concepts to develop statistical data
and model driven techniques and algorithms for modeling, estimation,
fault detection, fault classification, monitoring and diagnosis, which aim
to improve process operations and enhance the data validation. Dr. Majdi
Mansouri is the author of more than 150 refereed journal and conference
publications and book chapters, and has worked on several projects as lead
principal investigator (LPI) and principal investigator (PI). Dr. Mansouri is
a member of IEEE.

Mohamed-Faouzi Harkat Dr. Mohamed-Faouzi Harkat received his

Eng. degree in Automatic control from Annaba University, Algeria in 1996,
his Ph.D. degree from Institut National Polytechnique de Lorraine (INPL),
France, in 2003. He is now Professor in the Department of Electronics at
Annaba University, Algeria. His research interests include fault diagnosis,
process modeling and monitoring, multivariate statistical approaches, and
neural networks. Dr. Harkat is the author of more than 100 refereed jour-
nal and conference publications and book chapters.

Hazem Numan Nounou Dr. Hazem N. Nounou (SM’08) is a professor

in the Electrical and Computer Engineering Program and the Assistant
Dean for Academic and Student Services at Texas A&M University at
xix
xx About the authors

Qatar. In 2015–2017, he was the holder of Itochu Professorship. He re-

ceived the B.S. degree (Magna Cum Laude) from Texas A&M University,
College Station, in 1995, and the M.S. and Ph.D. degrees from Ohio State
University, Columbus, in 1997 and 2000, respectively, all in electrical en-
gineering. In 2001, he was a Development Engineer for PDF Solutions, a
consulting firm for the semiconductor industry, in San Jose, CA. Then, in
2001, he joined the Department of Electrical Engineering at King Fahd
University of Petroleum and Minerals in Dhahran, Saudi Arabia, as an
Assistant Professor. In 2002, he moved to the Department of Electrical
Engineering, United Arab Emirates University, Al-Ain, UAE. In 2007, he
joined the Electrical and Computer Engineering Program at Texas A&M
University at Qatar, Doha, Qatar, where he is currently a professor. He
published more than 200 refereed journal and conference papers and book
chapters. He served as Associate Editor in technical committees of sev-
eral international journals and conferences. His research interests include
data-based control, monitoring and fault detection, intelligent and adaptive
control, control of time-delay systems, system biology, and system identifi-
cation and estimation. Dr. Nounou is a senior member of IEEE.

Mohamed Numan Nounou Dr. Mohamed Nounou is a professor of

Chemical Engineering at Texas A&M University-Qatar. He received the
B.S. degree (Magna Cum Laude) from Texas A&M University, College
Station, in 1995, and the M.S. and Ph.D. degrees from the Ohio State
University, Columbus, in 1997 and 2000, respectively, all in chemical en-
gineering. From 2000 to 2002 he was with PDF Solutions, a consulting
company for the semiconductor industry, in San Jose, CA. In 2002, he
joined the Department of Chemical and Petroleum Engineering at the
United Arab Emirates University. In 2006, he joined the Chemical Engi-
neering Program at Texas A&M University at Qatar, where he is currently
a professor. He has received research funding over $5M and published more
than 190 refereed journal and conference papers and book chapters. He also
served as Associate Editor in technical committees of several international
journals and conferences. His research interests include process modeling,
monitoring, estimation, system biology, and intelligent control. He is a se-
nior member of the American Institute of Chemical Engineers (AIChE)
and a senior member of the Institute of Electrical and Electronics Engi-
neers (IEEE).
 Visit https://ebookluna.com
now to explore a diverse
collection of ebooks available
in formats like PDF, EPUB, and
MOBI, compatible with all
devices. Don’t miss the chance
to enjoy exciting offers and
quickly download high-quality
materials in just a few simple
steps!
Acknowledgments

This work was made possible by NPRP grant NPRP9-330-2-140 from

the Qatar National Research Fund (a member of Qatar Foundation). The
statements made herein are solely the responsibility of the authors.

xxi
List of acronyms

GLRT Generalized likelihood ratio test

KGLRT Kernel generalized likelihood ratio test
EWMA Exponential weighted moving average
CUSUM Cumulative sum
SPE Squared prediction error
LVR Latent variable regression
MSE Mean square error
PCA Principal component analysis
PLS Partial least square
MSPLS Multiscale partial least square
MSPCA Multiscale principal component analysis
IPCA Interval principal component analysis
KPCA Kernel principal component analysis
KPLS Kernel partial least square
MSPLS Multiscale kernel partial least square
MSKPCA Multiscale kernel principal component analysis
IKPCA Interval kernel principal component analysis
CPCA Centers principal component analysis
CPLS Centers Partial least square
MRPCA Midpoints-radii principal component analysis
MRPLS Midpoints-radii partial least square
CIPCA Complete information principal component analysis
CIPLS Complete information partial least square
SNR Signal-to-noise ratio
CSTR Continuous stirred tank reactor
AQMN Air quality monitoring network
TEP Tennessee Eastman process
FAR False alarm rate
MDR Missed detection rate
ARL1 Average run length
WL Window length
MW Moving window

xxiii
Nomenclature

Latin letters
X ∈ RN ×m Input data matrix
x ∈ Rm Input vector
y ∈ Rp Output vector
t ∈ R Latent variable
P∈R Eigenvector matrix
p Eigenvector
I Identity matrix
E Residual matrix
e Residual vector
m Number of inputs
n Number of samples
 Number of retained principal components
T2 Hotelling statistic
Q Squared prediction error statistic
T Generalized likelihood ratio test statistic
w Window length

Greek letters
 Eigenvalue matrix
 Model error
λ Eigenvalue
σ Standard deviation
φ(·) Nonlinear mapping
ψ Wavelet basis function
φ Orthonormal scaling function

xxv
CHAPTER 1

Introduction
Process monitoring is essential for proper and safe operation of various
industrial processes (like chemical and environmental processes), and it has
recently become even more important than ever before.
Proper operation of complex chemical processes, such as those in the oil
and gas industries, requires careful monitoring of certain key process vari-
ables to enhance the productivity of these processes and, more importantly,
to avoid disasters in the cases of failure [1]. Many serious accidents have
occurred in the past few decades in various chemical and petrochemical
plants all over the world. These accidents include the Union Carbide ac-
cident [2,3], the Piper Alpha accident [4,5], and the Al-Ahmedi (Kuwait)
accident [6]. The Union Carbide accident occurred in Bhopal, India, in
1984, where a major toxic gas leak resulted in over 3000 fatalities and in-
jured 400,000 others in the surrounding neighborhoods [2,3]. The 1988
accident in Piper Alpha (an oil production plant operated by Occidental
Chemical in the North Sea) involved an explosion killing 167 men, leaving
only 61 survivors [4,5]. The accident in Mina Al-Ahmedi in 2000, on the
other hand, was due to a failure in a condensate line in a refinery plant caus-
ing the death of 5 people and injuring 50 others [6]. These accidents show
that tight monitoring of chemical and petrochemical processes is essential
for safe and profitable operation of these plants.
Also, monitoring the atmospheric air pollution levels is extremely im-
portant for the safety of humans and the marine life, especially in areas with
large fuel productions or consumptions and large climate fluctuations [7].
For example, the heat wave in France in the summer of 2003 was linked to
an exceptional ozone pollution that affected the entire European commu-
nity [8]. The consequences of this heat wave demonstrated the importance
of having reliable warning systems to detect unexpected pollution levels
and any unforeseeable events [8]. Proper monitoring of air pollutants pro-
vides useful information that can help people take the needed precautions
to avoid undesirable consequences. During the past few decades, a lot of
effort has been made to improve air quality.
Fault detection is often used for process monitoring. Possible faults can
be due to malfunctioning sensor/s (called sensor faults) or to abnormal
changes in the process. Sensor faults are usually quantified by sudden (or
Data-Driven and Model-Based Methods for Fault Detection and Diagnosis Copyright © 2020 Elsevier Inc.
https://doi.org/10.1016/B978-0-12-819164-4.00010-8 All rights reserved. 1
2 Data-Driven and Model-Based Methods for Fault Detection and Diagnosis

quick) changes in a small number of process variables. Process faults, on

the other hand, are abnormal changes caused by deviations in the process
itself. These faults are usually quantified by slow drifts across several vari-
ables. The need for monitoring techniques that can accurately and quickly
detect abnormal situations (sensor or process faults) has greatly attracted the
attention of researchers and engineers. Over the past few decades, several
monitoring techniques have been developed [6,9–11]. Generally, fault de-
tection techniques can be classified into two main categories: data-based or
model-free techniques and model-based techniques.
Model-based monitoring methods rely on comparing the process mea-
surements with knowledge obtained from a mathematical process model,
which is usually derived using some fundamental understanding of the
process under fault-free conditions. The residuals, which are the differ-
ences between the measurements and the model predictions, can be used
as an indicator about the existence or absence of faults [12,13]. When the
monitored process is under normal operating conditions (no faults exist),
the residuals are zero or close to zero in cases of modeling uncertainties and
measurement noise. However, when a fault occurs, the residuals deviate sig-
nificantly from zero indicating the presence of a new condition that is sig-
nificantly distinguishable from the normal faultless working mode [12,13].
The model-based monitoring approaches include the observer-based meth-
ods [14,15], parity space approaches [16–19], and interval approaches [20].
Of course, the effectiveness of these model-based monitoring methods de-
pends on the accuracy of the models used.
The effective performance of various practical systems requires proper
process operations, such as modeling and monitoring. In these operations, it
is often assumed that the state variables of the process model are measurable
and that the model parameters are available. In many cases, however, obtain-
ing such measurements or determining the model parameters can be costly,
difficult, or sometimes impossible. To deal with this problem, state and/or
parameter estimators are often utilized. Several estimation techniques, such
as the extended Kalman filter, unscented Kalman filter, and more recently
the sequential Monte Carlo method have been developed and utilized in
many applications. The classical Kalman filter (KF) was developed in the
1960s [21] and has been widely applied in various engineering and science
areas, including communications, control, machine learning, neuroscience,
and many others. In the case where the model describing the system is
assumed to be linear and Gaussian, the KF provides an optimal solution
[22–25]. KF has also been formulated in the context of Takagi–Sugeno
 Introduction 3

fuzzy systems, which can be described by a convex set of multiple linear

models [26–28]. It is known that KF is computationally efficient; however,
it is limited by the nonuniversal linear and Gaussian modeling assumptions.
To relax such assumptions, the extended Kalman filter [22,23,29–31] and
the unscented Kalman filter [22,23,32–34] have been developed. In ex-
tended Kalman filtering the model describing the system is linearized at
every time sample (which means that the model is assumed to be differ-
entiable). Therefore, for highly nonlinear models, EKF does not usually
provide a satisfactory performance. The UKF, on the other hand, instead
of linearizing the model to approximate the mean and covariance matrix
of the state vector, uses the unscented transformation to approximate these
moments. In the unscented transformation a set of samples (called sigma
points) are selected and propagated through the nonlinear model to im-
prove the approximation of these moments and thus the accuracy of state
estimation.
Other state estimation techniques use a Bayesian framework to esti-
mate the state and/or parameter vector [35]. The Bayesian framework relies
on computing the probability distribution of the unobserved state given a
sequence of the observed data in addition to the state evolution model.
Consider an observed data set y, which is generated from a model defined
by a set of unknown parameters z [36]. The beliefs about the data are
completely expressed via the parametric probabilistic observation model
P(y|z). The learning of uncertainty or randomness of a process is solved by
constructing a distribution P(z|y), called the posterior distribution, which
quantifies our belief about the system after obtaining the measurements.
According to the Bayes rule, the posterior distribution can be expressed as

P(z|y) ∝ P(y|z)P(z),

where P(y|z) is the conditional distribution of the data given the model
parameter vector z, which is called the likelihood function, and P(z) is
the prior distribution, which quantifies our belief about z before obtain-
ing the measurement. Thus the Bayes rule specifies how our prior belief,
quantified by the priori distribution, is updated according to the mea-
sured data y. Unfortunately, for most nonlinear systems and non-Gaussian
noise observations, a closed-form analytic expression of the posterior dis-
tribution of the state is untractable [37]. To overcome this drawback, a
nonparametric Monte Carlo sampling-based method called the sequen-
tial Monte Carlo method (SMC) (also known as particle filtering (PF))
4 Data-Driven and Model-Based Methods for Fault Detection and Diagnosis

[38–40] has recently gained popularity. SMC methods approximate the

posterior probability distribution by a set of weighted samples, called par-
ticles. Since real-world problems usually involve high-dimensional random
variables with complex uncertainty, the nonparametric and sample-based
estimation of uncertainty (provided by the PF) has thus become quite pop-
ular to capture and represent the complex distribution P(z|y) in nonlinear
and non-Gaussian models [41].
The model-based methods, mainly the Shewhart chart [42], the expo-
nentially weighted moving average (EWMA) chart [43,44], the cumulative
sum (CUSUM) chart [45], and the generalized likelihood ratio test (GLRT)
chart [46,47], have been also used to improve the model-based FD capabil-
ities.
Whereas the Shewhart chart considers solely the present data sam-
ple to evaluate performance, the CUSUM and EWMA chart considers
a weighted sum of past observations. The CUSUM chart provides same
weight for all past observations, whereas the EWMA chart gives more
importance to the more recent observations [48–50]. Both CUSUM and
EWMA charts perform almost equally in detecting small mean shift, but
the EWMA chart is somewhat easier to set up and operate. Moreover,
since EWMA statistic is the weighted average of all previous and present
observations, it is less sensitive to the normality assumption [48,51].
However, the classical EWMA chart is not suitable to deal with an ex-
tensive range of fault sizes, since it has to be tuned according to different
sizes. The improved chart called max-double EWMA (M-DEWMA) has
showed higher detection performances than the classical EWMA chart in
detecting minor and moderate shifts in the mean and/or variance [52].
The Max-DEWMA control chart considers the highest of the absolute
values for two EWMA statistics, one controlling the mean and the other for
the variance. It has been presented that the Max-DEWMA chart performed
higher than the Max-EWMA chart in detecting shifts in the mean and/or
variance.
The authors in [53,54], developed an enhanced single chart named sum
of squares-DEWMA (SS-DEWMA) chart aims at detecting shifts of all
sizes in the mean and/or variance. It has been shown that the SS-DEWMA
chart performed higher than the Max-DEWMA chart in detecting shifts
in the mean and/or variance, and both of them outperformed the classical
EWMA [53,54].
Unfortunately, sometimes it is very difficult to derive accurate models
of the monitored systems, especially for complex processes, such as in the
 Introduction 5

cases of many chemical and environmental processes. For example, mod-

eling the ozone level is very challenging because of the complexity of the
ozone formation mechanisms in the troposphere and the uncertainty about
the meteorological conditions. Also, modeling chemical and petrochemical
processes is a challenging task because of the complexity and sometimes of
the lack of understanding about these processes. In these cases, data-based
monitoring techniques are more commonly used.
Data-based monitoring methods, on the other hand, rely on the avail-
ability of historical data obtained from the monitored fault-free process [10].
These data are first used to build an empirical model, which is then used
to detect faults in future data. Data-based monitoring methods include
the latent variable regression methods, such as partial least square (PLS)
regression, principal component analysis (PCA), independent component
analysis (ICA), canonical variate analysis (CVA), [55,10], neural networks
[56], fuzzy systems [57], pattern recognition methods [58], and support
vector machine (SVM) based methods [59,60]. SVM-based fault detection
methods can be applied to nonlinear systems and offer advantages over
conventional nonlinear optimization-based techniques. Data-based moni-
toring methods, especially those that utilize PCA or its extensions, have
been widely used in many applications in a very wide range of industries,
for example, air quality monitoring [61], chemical industry [62,63], wa-
ter treatment [64,65], pharmacology [66], biology and biotechnology [67],
agriculture [68], health [69], semiconductors [70], and many others.
In the scientific synthesis, several objectives are sought. First, the ad-
vantages of the hypothesis testing fault detection methods are exploited in
the cases where process models are not available by developing linear and
nonlinear latent variable regression (LVR) based hypothesis testing fault
detection methods. LVR techniques (such as PCA and PLS) are made to
achieve further improvements and widen the applicability of the developed
methods in practice. Also, kernel LVR (including kernel PCA and kernel
PLS) methods are used to deal process nonlinearities. Secondly, to account
for uncertainty in the measured data, two approaches are followed. In one
approach, multiscale representations of data are utilized to reduce the false
alarm rate through enhancing the noise-feature separation in the data and
decorrelating autocorrelated errors in the measurements. In another ap-
proach, to deal with uncertainty, interval LVR (ILVR), which quantifies
uncertainty in the data by defining intervals in which the data may fall, is
adopted. Integrating hypothesis testing and ILVR helps develop fault de-
tection techniques that are more sensitive to the presence of faults.
6 Data-Driven and Model-Based Methods for Fault Detection and Diagnosis

To deal with scenarios where the process model is available and has
a pre-defined structure (obtained using material and energy balances), an
improved model-based fault detection technique that aims at enhancing the
monitoring of industrial systems is developed. The objectives are twofold.
Firstly, a state estimation technique that can accurately estimate the state
variables in such systems will be developed; secondly, new fault detection
charts based on exponentially weighted moving average (EWMA) will be
proposed.
Finally, the developed fault detection methods are applied to enhance
monitoring various chemical, biological, and environmental processes.
This scientific synthesis is organized in seven chapters. Chapter 2
presents detailed descriptions of the developed linear LVR-based gener-
alized likelihood ratio test (GLRT) methods for fault detection. Chapter 3
extends the linear LVR-based GLRT approaches to deal with nonlinearities
on the systems by developing nonlinear LVR (including kernel PCA and
kernel PLS) based GLRT fault detection techniques. In Chapter 4, multi-
scale LVR-based fault detection approaches are presented, which improve
the monitoring abilities of the classical LV-based techniques. Chapter 5
presents interval LVR-based approaches for monitoring uncertain systems,
followed by their applications to chemical and environmental processes.
Then, in Chapter 6 the model-driven fault detection techniques based on
particle filter (PF) and exponentially weighted moving average (EWMA)
and their applications to monitoring biological processes are presented. Fi-
nally, a summary of the main conclusions out of this report and the future
works are presented in Chapter 7.

References
[1] F. Khan, S. Abbasi, Major accidents in process industries and an analysis of causes
and consequences, Journal of Loss Prevention in the Process Industries 12 (5) (1999)
361–378.
[2] V. Dhara, The Union Carbide disaster in Bhopal: a review of health effects, Archives
of Environmental Health 27 (5) (2002) 391–404.
[3] B. Bowonder, The Bhopal accident, Technological Forecasting and Social Change
32 (2) (1987) 169–182.
[4] L. Cullen, The Public Inquiry into the Piper Alpha Disaster, HMSO, 1990.
[5] M. Pate-Cornell, Learning from the Piper Alpha accident: a postmortem analysis of
technical and organizational factors, Risk Analysis 13 (2) (1993) 215–232.
[6] V. Venkatasubramanian, R. Rengaswamy, S. Kavuri, K. Yin, A review of process fault
detection and diagnosis part I: quantitative model-based methods, Computers and
Chemical Engineering 27 (2003) 293–311.
 Introduction 7

[7] M. Al-Maslamani, Assessment of Atmospheric Emissions Due to Anthropogenic Activ-

ities in the State of Qatar, PhD thesis, Institute for the Environment, Brunel University,
2011.
[8] M. Poumadere, C. Mays, S.L. Mer, R. Blong, The 2003 heatwave in France: dangerous
climate change here and now, Risk Analysis 25 (6) (2005) 1483–1494.
[9] Y. Qingsong, Model-Based and Data Driven Fault Diagnosis Methods With Applica-
tions to Process Monitoring, PhD thesis, Case Western reserve University, May 2004.
[10] V. Venkatasubramanian, R. Rengaswamy, S. Kavuri, K. Yin, A review of process fault
detection and diagnosis part III: process history based methods, Computers and Chem-
ical Engineering 27 (2003) 327–346.
[11] K. Chaitanya, Data-Based Modeling: Application in Process Identification, Monitoring
and Fault Detection, PhD thesis, 2011.
[12] M. Kinnaert, Fault diagnosis based on analytical models for linear and nonlinear sys-
tems – a tutorial, in: Proceedings of the 15th International Workshop on Principles of
Diagnosis, 2003, pp. 37–50.
[13] M. Nyberg, C.M. Nyberg, Model Based Fault Diagnosis: Methods, Theory, and Au-
tomotive Engine Applications, PhD thesis, 1999.
[14] R. Clark, D. Fosth, V. Walton, Detecting instrument malfunctions in control systems,
IEEE Transactions on Aerospace and Electronic Systems AES-11 (4) (1975) 465–473.
[15] R.J. Patton, P. Frank, R. Clarke (Eds.), Fault Diagnosis in Dynamic Systems: Theory
and Application, Prentice-Hall, Inc., 1989.
[16] M. Staroswiecki, Redondance analytique, in: Automatique et statistiques pour le diag-
nostic, Hermes Science Europe, 2001.
[17] X. Ding, P. Frank, Fault detection via factorization approach, System Control Letters
14 (5) (1990) 431–436.
[18] R.J. Patton, J. Chen, A review of parity space approaches to fault diagnosis, in: Pro-
ceedings of SAFEPROCESS’91, 1991, pp. 239–255.
[19] E. Chow, A. Willsky, Analytical redundancy and the design of robust failure detection
systems, IEEE Transactions on Automatic Control 29 (7) (Jul 1984) 603–614.
[20] K. Benothman, D. Maquin, J. Ragot, M. Benrejeb, Diagnosis of uncertain linear
systems: an interval approach, International Journal of Sciences and Techniques of Au-
tomatic Control Computer Engineering 1 (2) (2007) 136–154.
[21] R.E. Kalman, A new approach to linear filtering and prediction problem, Transactions
of ASME, Ser. D, Journal of Basic Engineering 82 (1960) 34–45.
[22] D. Simon, Optimal State Estimation: Kalman, H∞ , and Nonlinear Approaches, John
Wiley and Sons, 2006.
[23] M. Grewal, A. Andrews, Kalman Filtering: Theory and Practice Using MATLAB, John
Wiley and Sons, 2008.
[24] V. Aidala, Parameter estimation via the Kalman filter, IEEE Transactions on Automatic
Control 22 (3) (1977) 471–472.
[25] L. Matthies, T. Kanade, R. Szeliski, Kalman filter-based algorithms for estimating depth
from image sequences, International Journal of Computer Vision 3 (3) (1989) 209–238.
[26] G. Chen, Q. Xie, L. Shieh, Fuzzy Kalman filtering, Journal of Information Science
109 (1998) 197–209.
[27] D. Simon, Kalman filtering of fuzzy discrete time dynamic systems, Applied Soft Com-
puting 3 (2003) 191–207.
[28] H. Nounou, M. Nounou, Multiscale fuzzy Kalman filtering, Engineering Applications
of Artificial Intelligence 19 (2006) 439–450.
 Visit https://ebookluna.com
now to explore a diverse
collection of ebooks available
in formats like PDF, EPUB, and
MOBI, compatible with all
devices. Don’t miss the chance
to enjoy exciting offers and
quickly download high-quality
materials in just a few simple
steps!
Other documents randomly have
different content
 Mr. Drayton extricated his fingers, and examined them with a
rather rueful smile.

"There's no need to apologise," he said. "Your gratitude's quite

pressing enough for me." He rose to his feet and, turning towards
the table, pulled forward a bundle of papers. "There are one or two
things here that want signing, and then I think I had better take you
round to the bank, before they shut, and introduce you to the
manager. It's only just across the road in Holborn."

I sat down at the table, and scribbled my name in the spaces he

pointed out. I suppose I ought really to have read the documents
through, but so complete was my confidence in his good faith that I
made no attempt at this elementary precaution.

"You have got a delightfully trustful nature, Dryden," he said with

a sigh. "I wish I inspired the same sort of feeling in all my clients."

He rang the bell for his clerk, and, having informed the latter that
he would be out for the next quarter of an hour, he picked up his hat
and led the way downstairs.

"Are you staying in town to-night?" he enquired as we emerged

into Bedford Row. "If so, and you have nothing better to do, come
and dine with me at my club. It's very dull, but you're not likely to
notice that after a fortnight at Greensea Island."

Had I chosen I think I might have been able to disillusion him on

the subject of Greensea's dullness, but for the present at all events I
still thought it wiser to keep my own counsel. So with a perfectly
truthful remark that I was never bored in good company, I
laughingly accepted his invitation, and without further discussion we
turned down a narrow passage into the roar and bustle of Holborn.

Our interview with the bank was not a very formidable affair. We
were shown into a private room, where a brisk, bald-headed little
man with gold-rimmed eye-glasses was seated at a table several
sizes too large for him.

Mr. Drayton introduced me as the heir to the Jannaway fortunes,

and the manager—for such the bald-headed gentleman proved to be
—congratulated me cordially on what he termed my "romantic
inheritance." He announced that a sum of eleven thousand and
forty-five pounds seven shillings and six-pence was lying in the office
awaiting my attention—a statement which I tried to receive with
becoming nonchalance. At his request I wrote out a specimen of my
signature, receiving in return a useful-looking cheque-book. He then
informed me that if at any time I needed expert financial advice he
would be delighted to place himself at my disposal; after which he
again shook hands with us both, and escorted us in state to the door
of the bank.

"I like being a capitalist," I observed, as we stood for a moment

on the step. "It's a much more restful life than the sea, and
everybody one meets is so extraordinarily obliging."

Mr. Drayton chuckled appreciatively. "Wait till I've sent you in my

bill," he retorted. "You'll have some excuse for feeling cynical then."
He dived into his pocket and produced a card-case. "Here's the
address of my club," he added, "in case you forget it. I must be off
now, but I'll expect you at seven-thirty. Don't dress up and make
yourself beautiful—come along just as you are."

With a friendly wave he disappeared amongst the traffic, while

almost at the same moment a prowling taxi pulled up in the gutter. I
moved forward and accosted the driver.

"Have you ever heard of a place called Angel Court, somewhere

off Fleet Street?" I asked him.

He eyed me critically.

"Are you wantin' Inspector Campbell's office?" he enquired.

 "Yes," I said. "Do you know him?"

He leaned across and opened the door. "Know 'im," he repeated

rather scornfully. "Why, 'e pinched a bloke outer this very cab last
March twelve-months. There ain't a taxi driver in London as don't
know 'Foxy' Campbell."

Considerably impressed with this unexpected tribute to the

Inspector's reputation, I climbed inside the vehicle. We sped away
rapidly through a number of side turnings, coming out at last within
a few yards of the bottom of Fleet Street. A moment later the taxi
pulled up, and as I stepped out the driver jerked his thumb in the
direction of a narrow archway.

"That's Hangel Court," he said. "You'll find the party you're

looking for the second door on the right."

I thanked him, and, passing through the opening, entered a kind

of paved yard, on three sides of which were quaint narrow little
houses with old-fashioned window-panes. The second doorway bore
a carefully polished brass plate, with the inscription "James
Campbell. Private Enquiry Agent."

I was just hunting around for a bell when suddenly the door
swung back, and, looking up, I found myself face to face with the
Inspector himself. He was evidently on the point of going out, for he
was wearing a bowler hat and carrying a walking-stick.

Directly he saw me his big square-jawed face lit up in a smile of

welcome.

"Why, it's Mr. Dryden," he exclaimed, holding out his hand. "Well,
this is a bit of luck. If you had been a moment later you'd just have
missed me."

"It's my lucky day," I said "But, all the same, I mustn't keep you if
you've got an appointment."
 "It's nothing very important," he replied. "Come along inside. I
can spare a minute or two anyhow."

He led the way into a comfortably furnished office on the right,

and, pulling forward a chair, seated himself exactly opposite me. His
large and very wide-awake blue eyes scanned me with friendly
interest.

"I'd better start by apologising," he began. "You mustn't think I've

forgotten your invitation to come down and have a bit of shooting,
but the fact is I've been so busy I've had no time to write. People
have been misbehaving themselves the last three weeks in a way
you wouldn't believe; I suppose it's the hot weather."

"I shouldn't wonder," I said. "I've felt a little vicious myself once
or twice." I tossed away the stump of the cigar I was holding and
helped myself to another from the case which he offered me. "As far
as the invitation goes," I added, "it's open permanently. I am not the
sort of a person who wants a lot of notice. Just send along a wire as
soon as the crime wave subsides."

"The very minute," he said with a twinkle. Then, putting his

thumbs in his waistcoat, he leaned back comfortably in his chair.
"And how are you getting on down there?" he asked. "You know, I
take a special interest in your affairs, Mr. Dryden. There's something
about Greensea Island that I didn't quite get to the bottom of, and
no one likes to be hit in his professional pride."

He spoke in such a frank and good-humoured fashion that once

again I was sorely tempted to make a clean breast of my difficulties.
I stuck to my resolve, however, for the thought of Christine as the
central figure in some public scandal was too utterly repugnant to be
considered. At the same time I felt that in the case of Inspector
Campbell a certain measure of honesty would undoubtedly be the
best policy.
 "As a matter of fact," I said, looking him straight in the face, "I've
come here to ask you for your help."

He nodded his head.

"I thought you might," he said. "That was one reason why I gave
you my address."

"The trouble is this," I went on bluntly. "I want you to do

something for me, but for certain reasons which I can't explain it's
quite impossible for me to answer any questions. I know it sounds
pretty rotten, so I shan't be the least surprised or offended if you tell
me to go to the devil."

He smiled genially. "That's the last thing I should do, Mr. Dryden.
There's no sense in encouraging a rival firm." He paused. "You know
your own business best, of course, but if there's any particular way
in which I can be of assistance you can count on me from this
minute."

"Well, there is," I confessed. "You remember a chap called Dr.

Manning—the fellow who looked after my uncle when he was
dying?"

Once again the Inspector nodded. "Yes," he answered. "I

remember him very distinctly. He lived on a barge about three
hundred yards above the island."

"That's the gentleman," I said. "He's living there still; in fact, to all
intents and purposes he's my next-door neighbour. What I want you
to do, if you can possibly spare the time, is to get me a little
accurate information about his past life."

If my companion felt any surprise or curiosity he certainly didn't

show it.
 "That oughtn't to be very difficult," was his reply. "Doctors are
fairly easy to trace as a rule." He reached across to the table and
picked up a half sheet of notepaper. "Has he ever mentioned a
particular town, or given you any idea of what hospital he was at?"

"He once told me he had been a ship's surgeon," I said. "It was
probably a lie, so I shouldn't attach much importance to it."

The Inspector made a brief note. "One never knows," he

observed hopefully. "People sometimes tell the truth by accident.
They generally regret it afterwards."

A telephone bell tinkled sharply in the corner of the room, and

with a word of excuse he rose from his seat and walked across to
the instrument.

"Yes," he said. "I'm Campbell—speaking." There was a pause.

"What's that?" Another and longer pause followed. "Oh, very well.
Don't let him go. I'll be round almost immediately."

He hung up the receiver and turned back to me.

"I'm sorry," he said, "but I'm afraid I shall have to run away after
all. I have got to meet a man who wants to leave for Scotland this
afternoon." He folded up the paper he was holding, and put it away
carefully in his pocket "I'll get on to this little job at once. You shall
hear from me directly there's any news."

"Thanks," I said gratefully; "and, meanwhile, if the wicked should

happen to cease from troubling——"

"You'll not only hear from me, but see me."

He gave me a friendly grip, and, collecting his hat and stick, led
the way out into the yard. We walked together under the archway
and emerged on to the crowded pavement of Fleet Street.
 "There's one thing I should like to add, Mr. Dryden," he said, just
before we parted. "If you ever find yourself in real difficulty or actual
danger, don't hesitate to let me know. Whatever I'm doing I shan't
be too busy to lend you a hand."

I tried to thank him, but, without waiting to listen, he stepped off

into the roadway, and a moment later he was swallowed up amongst
the traffic.

Brief as our interview had been, however, it had left me in a very

satisfied frame of mind. From what I had seen of the Inspector I felt
pretty sure that if there was anything worth knowing about
Manning's past career he would soon succeed in ferreting it out.
Apart from that, the knowledge that in a tight corner I should have
him as well as Bobby at my back was the kind of encouraging
thought that would have put heart into a rabbit.

My good spirits were in no way damped by one or two drops of

rain, which at that moment, much to my surprise, began to dot the
pavement. Looking up, I noticed for the first time that a change was
taking place in the weather. A great bank of cloud was creeping
slowly up from the east, and spreading in a black, threatening line
all across the sky.

Manning's prophecy at once came into my mind, but, not being

particularly anxious for a ducking, I wasted no time in further
reflections. Hailing a passing taxi, I instructed the man to take me
back to the Great Eastern Hotel, which I reached in comfort and
dryness just as the rain began to descend in a steady drizzle.

There was nothing to do under the circumstances but to make the

best of things indoors. I am never at a loss how to amuse myself if I
can get a game of billiards, and I soon discovered that the hotel
possessed a couple of excellent tables, presided over by a red-
headed marker, who spoke Scotch and looked as if he drank it.
 We played several hundred up with varying success; and at six-
thirty, feeling quite ready to face a good dinner, I set out for Mr.
Drayton's club. As an additional encouragement to my appetite I
walked the whole way to St. James's Street, for by this time,
although the pavements were still wet, the rain itself had practically
ceased falling.

The club proved to be a big, sombre establishment, inhabited

chiefly by middle-aged gentlemen of a distinctly morose aspect.
What they had to be morose about I can't imagine, for a better meal
than the one put before us I never wish to eat. Mr. Drayton, who
was waiting for me in the hall and conducted me straight into the
dining-room, confessed frankly that he had joined the place purely
for the sake of its excellent cooking, and the respect which I already
had for his judgment rose steadily with each succeeding course.

After dinner we made our way to the smoking-room upstairs,

where, with the assistance of a couple of leather arm-chairs and a
plentiful supply of whisky and soda, the time passed away so
pleasantly that it was close on midnight before I got back to my own
bed at the hotel.

The whisky must have been of the same high quality as the rest
of the food, for in spite of the amount that I had put away, I woke
up next morning feeling remarkably fit and cheerful. One of the first
things I did was to pull aside the blind and have a look at the
weather. The result was not encouraging, for my window faced on to
a blank wall of dirty white brick, and in the intervening space a thin
grey drizzle was coming down with monotonous persistence.

Once more Manning's predictions about a sudden change in the

weather came back to my memory. If it were like this in London, the
odds were that Greensea Island would be wrapped in a regular sea
fog, which I knew from experience might last for several days.
However, I consoled myself with the comforting reflection that it
would save me a deal of hard work with the watering can, and that,
after all, I was no longer compelled to tramp up and down a
dripping bridge with a miniature trout stream trickling down the back
of my neck.

I had made up my mind to go back by the mid-day train, which

gets into Torrington somewhere about two o'clock. There were
several things I still wanted to buy, and, now that I was actually in
town, it seemed a pity to waste such a favourable opportunity. I had
no wish to get soaked through, however, so on my way down to the
dining-room I gave the hall porter a sovereign and told him to go
out and buy me an umbrella. He returned with a sporting-looking
affair, the handle of which terminated in a bull-dog's head. Though a
trifle spectacular, it was a sound article for the money, and, armed
with this and my cheque book, I set off after breakfast, pleasantly
determined to make a good opening in Uncle Richard's life savings.

As intentions go, it must be admitted that I was fairly successful.

In addition to half a dozen things I bought for myself, I discovered in
a Bond Street jeweller's a perfectly charming little emerald pendant,
which struck me at once as having been obviously designed for
Christine. Under the circumstances twenty-five pounds seemed to be
a ridiculously cheap price, but I felt that it was unnecessary to point
this out to the shopman, who appeared quite satisfied with the
transaction.

Returning to the hotel with my purchases, I made an early but

excellent lunch off cold duck and salad, and at a quarter to one I
was steaming out through the damp squalor of East London,
comfortably established in the corner seat of a first-class smoker.

It was a cheerless journey, for all the way down a driving rain
beat incessantly upon the windows, and most of the landscape was
blotted out by drifting clouds of mist.

At Torrington things were not much better. It was not actually

raining, but the little cobble-paved market-place reeked with wet,
and a depressing air of forlorn moisture brooded over the entire
place.

Packing myself and my belongings into a solitary growler which

was standing in the gutter, I instructed the man to take me as far as
the Gunner's Arms. We plodded off, and, after half an hour's drive
between dripping hedgerows, came out at last over the brow of the
long slope that overlooks the estuary. I lowered the sash, and,
putting my head out of the window, inspected the prospect. Down
below, the roof of the inn was just visible, but the harbour itself and
everything beyond was hidden under a pall of fog, which stretched
like a grey shroud over the whole length of the estuary. There was
something singularly sinister and desolate about the whole scene,
and for a moment I felt a pang of regret for the comfortable billiard-
room which I had so recently quitted.

We came to a halt on the quay in front of the inn, deserted for

once by its usual sprinkling of boatmen and barge hands. Our arrival
had been observed, however, for scarcely had we pulled up when I
saw the small figure of Jimmy emerge from the stable door and
come hurrying down the yard towards us. He touched his damp curls
with a cheerful grin of welcome.

"I reckoned it was you, guv'nor," he remarked "You ain't bin away
long, an' that's a fact."

"Quite long enough, Jimmy," I returned. "At all events, you've had
time to mess up the weather."

He took the bag which I handed out, and at the same time cast a
rather disgusted glance into the surrounding gloom.

"It's a treat, ain't it!" he observed. "Come on like this yes'day

afternoon." Then, drawing a little nearer, he added in a lowered
voice: "There's a letter for you at the 'ouse, guv'nor. The young laidy
give it me laite las' night, an' I took it across at once—saime as you
told me."
 I tried to cover my unseemly elation by lighting a cigarette.
"Where did you meet her?" I enquired.

"She come down to the quay. I see 'er standin' there in the dark,
an' I says to me self, As like as not she's wantin' me. So I just slips
over alongside of 'er, an' without sayin' nothin' she shoves the letter
in my 'and and orf she goes."

I looked at him gratefully. "You're a stout fellow, James," I said "I

knew I could depend on you directly I saw you."

He waited until I had paid the cabman, and then, shouldering my

bag, accompanied me to the boat.

"I seen 'er again since then," he continued confidentially. "I was
comin' along Duke's Laine about 'alf an hour ago, an' she was waitin'
to go aboard the doctor's barge."

I stopped short with a very distinct shock.

"Are you sure about that?" I demanded

He nodded a little scornfully. "Dead certain, guv'nor. There ain't

no one else you'd mistake for 'er—not round these parts."

"But how do you know she was waiting to go on the barge?" I

persisted.

"Why, I 'eard 'er call out, and I seed the doctor comin' orf to fetch
'er. I reckon she's there now—that's to say, if 'e ain't put 'er ashore
again."

I said nothing further, but walked on to the boat in anything but a

happy state of mind. The thought of Christine boxed up alone in the
fog with those two scoundrels filled me with a horrible sense of
uneasiness. Knowing her mistrust for Manning, and having seen
something of the relations between them, I felt that only the
strongest reasons would have induced her to take such a step.

It was possible, of course, that she had explained her action in

the letter which she had given to Jimmy. If that were the case, the
quicker I got back to the house the sooner my anxiety would be
relieved, so without wasting any time I unshipped the oars and
started off on my journey down the creek.

The tide was slack when we reached the mouth; and trusting to
luck, for one could see only a few feet ahead, I cut straight across
the estuary, and came out triumphantly almost exactly alongside the
landing-stage.

Before stepping ashore I handed Jimmy a well-earned ten bob in

return for his services. Then, with my bag in my hand, I strode
forward through the mist and entered the narrow path which ran up
through the shrubbery to the garden.

As I crossed the lawn and approached the house I noticed with

some surprise that the front door was standing open. I was just
thinking that it was rather unlike Bascomb to be so casual when I
suddenly heard the sound of footsteps coming round the corner of
the drive. I pulled up where I was, and the next moment the square,
burly figure of my retainer loomed into view.

I was about to make some remark, but, whatever it was, the

words remained unspoken. One glance at Bascomb's face was quite
sufficient to drive everything else clean out of my head.

"Good God!" I exclaimed. "What's the matter?"

He stood there, glaring at me with sullen, blood-shot eyes—a

silent and menacing figure in the mist. Then, taking a step forward,
he laid his hand on my arm.
 "You come along o' me," he said hoarsely. "I got something to
show you."

Turning on his heel, he led the way round the corner of the
house. Without saying anything further I followed him along the
path until we reached the back garden, where with equal abruptness
he came to a sudden standstill. Then, raising his hand, he pointed
towards a dim, motionless heap, which was just visible in the centre
of the lawn.

"There you are," he muttered. "You go an' look at that."

Two strides brought me to the spot, and a wicked oath broke

from my lips as I stared down at the sight in front of me. It was the
huge body of Satan, stretched out stiff and dead on the wet grass.

CHAPTER TWELVE
For several seconds I remained still, without speaking. The wave
of grief and anger that swept through me left me sick and shaken; I
could only stand there with clenched hands waiting until I could
control my voice.

"Who did it, Bascomb?" I said at last.

He came up to me, and, bad as I felt myself, I was almost

shocked by the sight of his face. It was like a horrible grey mask,
twisted and distorted with passion.

"Who d'yer think?" he demanded hoarsely. "There's only one devil

in the world who'd go fur to do a thing like that." He sank down on
his knees beside the body, and, sliding his hand under the big head,
lifted it tenderly from the grass.

"You mean Manning?" The words came out mechanically. I knew

the answer before I uttered them.

"Aye!" he muttered savagely; "that's who I mean. It was Mannin'

right enough—the blarsted, poisonin' 'ound." Very slowly he laid
down his burden and got up again on to his feet. "I'll be even with
'im for this," he added in a choking voice. "You mark my words. I'll
be even with 'im for this or my name ain't John Bascomb."

"When did it happen?" I asked.

He stared at me, as if only half understanding the question.

"When did it 'appen?" he repeated. "Why, las' night 'about eleven

o'clock. 'E were alive an' well at the 'alf hour—that I can swear to. I
was in the kitchen, an' I seen 'im pass the window—seen 'im with
me own eyes. I can't say exac'ly 'ow long arter it was when I went
to the back door; mebbe a matter o' five an' twenty minutes.
Any'ow, there 'e was, stretched out on the path, too bad even to
make a sound. Crawled back 'ome to die, 'e 'ad, pore beggar, an' me
inside not knowin' nothin' about it."

His voice trembled, and, raising the back of his hand, he brushed
it roughly across his eyes.

I wetted my lips, which were dry as leather.

"If Manning did this—" I began.

Bascomb turned on me with glowing eyes.

"You'll leave 'im to me," he said. "This is my job, this is; an' no
one ain't comin' in between us—not till I've finished with 'im."
 I saw that the man was half off his head with rage, and I felt that
for the present the best thing was to leave him to himself. In the
mood we were both in any further discussion would probably only
end in an explosion.

"We must bury the body at all events," I said shortly. "If you want
any help in digging the grave you will find me in the house."

I walked towards the back door, my mind still in a turmoil of

emotion, and, crossing the kitchen, made my way through into the
hall. The place was dark and cheerless, and in a mechanical fashion
I struck a match and set a light to the fire. Then, pulling up an easy
chair, I flung myself down, and stared at the crackling sticks.

I suppose it ought to have been obvious from the first that an

attempt might be made to poison Satan, but somehow or other the
possibility had never entered my mind. The dastardly act had come
on me as a complete surprise, and in the first rush of anger and
indignation I found it difficult to collect my thoughts. That Bascomb
was right in his suspicions, however, I felt convinced, though I had
no more evidence than he had in support of my belief. There was a
cold-blooded ruthlessness about the whole thing which pointed
clearly to Manning, and such a passion of hatred for him swept
through me that for a moment I felt almost stifled.

What new threat, I asked myself, lurked behind this apparently

wanton piece of spite? It was utterly unlike Manning to show his
hand so clearly, unless driven to it by the most urgent reasons. He
must have realised that, however easily deceived I might be,
Bascomb would at once pitch on him as the author of the crime, I
might or might not accept the latter's opinion, but it was at least
probable that my suspicions would be aroused and that I should
take prompt steps to try and find out the truth.

That he would run such a risk through sheer malice was a thing
that I declined to believe. It was much more likely that he was fixing
up some new and devilish plan, in which the removal of Satan was
the first and perhaps the most essential step. Now that this was
accomplished he was not likely to waste time, for every hour that he
delayed would only add to the danger of discovery. I felt that I must
act, and act quickly, but what the deuce I ought to do was another
and more baffling question.

In my perplexity I suddenly remembered Bobby. A letter to

Harwich would reach him first post in the morning, and unless he
was up to his neck in work he would probably be with me by mid-
day. I didn't want to bother him unnecessarily, but things seemed to
be approaching a crisis, and if that were the case his advice and
help would b& simply invaluable.

I jumped up from my chair with the intention of writing him a

note at once. As I turned towards the desk, my eyes fell upon a
letter which was lying on the further side of the centre table. Even at
that distance I recognised Christine's hand, and at the same instant
the recollection of my talk with Jimmy rushed back into my mind.

In a second I had crossed the room and picked up the envelope.

Carrying it to the window, I tore open the flap, and with feverish
haste pulled out its contents. There was a single sheet of notepaper,
closely covered on both sides with small but very legible writing.

"I know that you have gone to London, but I am not

sure how long you intend to stay, or when you will get
this letter. Anyhow, it will be waiting for you as soon as
you reach the house.

"I had to write to you or see you. I feel certain that

something has happened—something that fills me with
terrible fear for your safety. If I knew what it was I
would tell you; I believe it must be connected with Dr.
 Manning's visit to the island, but that is only what I
think, and I may be wrong.

"I am absolutely sure, however, that at any moment

you may be in the greatest possible danger. Please
don't think that I am silly or fanciful in writing to you
like this. You must know by now the kind of people
you have to deal with, and surely you will understand
that I shouldn't have sent you this second warning
without some very real and serious reason for it.

"Whatever happens, don't under any circumstances

allow yourself to be alone on the island. I believe that
your man Bascomb is honest, but it would be better if
you were to have your naval friend with you as well.
Above all, be careful of Dr. Manning. Pay no attention
to any suggestion he makes, and mistrust everything
he says or does.

"If there is any further way in which I can help you I

will do so. Should you want to send me a message,
give it to the boy at the inn. My uncle has gone away
for two days, so I shall have no difficulty in getting
down to the quay in the evening.

"You are not to worry about me. I am in no danger

myself, and I implore you to think only of your own
safety.

"CHRISTINE DE RODA."

I read it hurriedly to the end, and then with a strange pleasure I

turned back and went through it a second time. It was just such a
letter as I should have expected her to write. Mysterious and reticent
as it was, there shone through every line an honesty and personal
courage which confirmed all I had felt about her ever since our
adventure at Leixoes.

Of one thing there could be no doubt. Whatever had prompted

her to send me this note, she had evidently been ignorant of the
brutal plan to poison Satan. Had she known she would certainly
have given me some warning, even if de Roda himself had been a
party to the scheme.

On that point, however, I had already and perhaps quite

unreasonably made up my mind. Nothing but the clearest evidence
would shake my belief that it was Manning's work, and that in all
probability he had arranged and carried out the whole thing for
some purpose of his own. More than ever I felt convinced that he
was playing a double part; that de Roda, though no doubt originally
the moving spirit, was now merely a tool in the hands of a much
cleverer man than himself.

Once more I glanced at Christine's letter. Her urgent pleading that

I would take the matter seriously was not really needed; I should
have done that even if the dead body of Satan had not been lying on
the grass outside. I knew well that it was no vague or imaginary
peril which had led a girl of her sense to send me this additional
warning—a step, which, for anything she might say to the contrary,
must have been fraught with considerable personal danger.

The last consideration, indeed, was much the most disturbing

factor as far as I was concerned. I was in no mood to worry about
my own safety when it was more than possible that, owing to her
efforts to help me, Christine might be exposing herself to all sorts of
horrible risks. People who don't hesitate to attempt murder are apt
to take a particularly unpleasant view of anything which they regard
as treachery. Of course de Roda was her uncle, but he looked the
kind of fanatic who would attach precious little importance to family
affection if it began to interfere with the success of his schemes.
Besides, according to the letter, de Roda was away. In his absence
Manning was presumably in charge of operations, and the mere
thought of Christine finding herself in the power of that gentleman
was sufficient to send a chill down the small of my back.

Suppose he had discovered that she was in communication with

me! It was unlikely, but unlikely things do happen in this world, and,
in spite of all our precautions, the secret might have leaked out. In
view of what I had seen outside "The Laurels" my feelings can be
easily understood, for he was just the sort of devil to make use of a
chance like this without compunction and without mercy.

I crumpled the letter in my hand, and thrust it into my pocket.

Whatever had induced Christine to visit the barge, the idea of her
shut up in that lonely cabin with no one else but Manning on board
was utterly unbearable. A frantic longing to do something rose up
inside me, and, walking to the window, I stared out into the drifting
mist.

I had stood there for perhaps thirty seconds when a really

inspired notion suddenly dawned on me. In a fog like this nothing
would be visible on the water more than a few yards away. If I were
careful about it there was no earthly reason why I should not get
into my boat and row up as far as the barge in perfect security. No
matter how sharp a look-out was kept, I should be practically
alongside before anyone could give the alarm.

Almost as soon as the idea came into my head I had made up my

mind. So strongly, indeed, did it attract me that I did not even stop
to consider what I should do when I reached my destination. I could
think about that while I was in the boat. The main thing was to get
started without wasting any more valuable time.

In spite of my eagerness to be off, I first of all sat down at my

desk and scribbled a note to Bobby. It had to be posted at Pen Mill
before six o'clock or else it would not be delivered in Harwich until
mid-day, and in that case he would very probably be out of the
office. I made no attempt to explain things, however; I merely told
him that I was in need of his help, and that if he could manage to
get down to Greensea the next day I should be uncommonly glad to
see him.

I was just fastening up the envelope when I saw Bascomb go

past the window. I called out to him from where I was sitting, and
he stepped in through the open doorway. He was carrying a heavy
garden spade.

"Where are you going to dig the grave?" I asked him.

He jerked his head toward the back of the house. "'Longside o'
the shed," he answered. "I can see it there from the kitchen
window."

He spoke as though the prospect of being continually reminded of

the crime afforded him a kind of sullen satisfaction.

"I am going ashore to post a letter," I said. "I mean to get to the
bottom of this infernal business, and I've asked Commander Dean to
come here to-morrow. I think he may be able to help us."

Bascomb looked at me more strangely than ever.

"It's likely enough," he remarked. "That's to say, if there's any 'elp

wanted."

He shouldered the spade as he spoke, and, stepping out on to the

path, walked off silently round the corner of the house. His manner
was so peculiar that for a moment I felt an uncomfortable doubt as
to whether he were in his right senses. I was too anxious about
Christine to worry over anything else, however, and, following him to
the door, I hastily secured the bolts and made sure that the windows
were properly fastened. Then, snatching up a cap from my travelling
bag, I left the house by the back door.
 The mist seemed thicker than ever when I came out into the
garden. I climbed the wet rail which led into the shrubbery, and,
having groped my way through the bushes, I struck off down the
path in the direction of the water.

A few minutes' walk brought me to the boathouse, where, by

gently coaxing the rusty lock, I managed to open the door without
making any unnecessary noise.

The next job was to find something with which to muffle the oars.
A glance round showed me exactly what I needed—a large piece of
cotton waste that was hanging from a nail just inside the threshold.
I tore this in two, and with the aid of some string soon accomplished
my purpose. Then, pushing off quietly, I drifted out into the mist.

It was just low water, and the young tide was already running up
the estuary. I knew its strength from bitter experience, so, keeping
the nose of my boat well into the current, I set off hopefully on what
I imagined to be the right course.

I was not far out in my reckoning, for, after about ten minutes'
steady rowing, I found myself approaching a mud flat, which I
recognised at once as being slightly to the east of the creek. I
allowed the current to carry me down as far as the entrance, and a
few moments later the rough wall of the jetty loomed into view on
my left-hand side.

There was no sign of anyone about, a circumstance that was

hardly surprising in view of the close proximity of the Gunner's Arms.
I ferried along till I reached the nearest post, where I shipped my
oars and made the boat fast. Then, climbing ashore, I started off
carefully down the slippery causeway.

I got to the pillar-box, which was just alongside the inn, without
meeting a soul. The knowledge that my letter was safely posted
gave me a distinct feeling of relief, and, retracing my steps as quickly
as possible, I was soon back in the dinghy and pulling off again
down the creek.

So far all was well, but as I drifted round the point into the open
water, I began to realize what a difficult job I had in front of me. In
the first place, it was quite conceivable that I was on a fool's errand.
Notwithstanding my fears, Christine might be in no need of my
assistance, and if that were the case my sudden appearance on the
scene would be about the most embarrassing ordeal I could possible
inflict on her. This was an unfortunate fact that had to be faced,
however; the only question was how to make certain that she was
safe without giving away the whole affair.

My strongest card was the one which had been thoughtfully

provided by Manning himself. His last words had been a pressing
invitation that I would come and see him on the barge as soon as I
got back from London, and he could hardly blame me if I happened
to select a particularly awkward moment for calling. Much as he
might object to my visit, he would find it uncommonly difficult to
invent any reasonable excuse for not asking me on board.

What I chiefly dreaded was the possibility that Christine might

betray herself before she could recover from her surprise. In that
event any further pretence on my part would be useless; Manning
would at once guess the truth, or something very near it, and what
the consequences might be God alone knew. Was I asking too much
in making such a demand upon a girl's courage and coolness? It was
a trial from which very few people would emerge successfully, but
then I had had ample proof in Leixoes harbour that Christine's nerve
was of no ordinary quality. Her self-possession on that occasion had
been truly delightful, and though she would now be up against a
much more severe test I had a kind of inward conviction that I could
still depend upon her. Anyhow, there was no way of avoiding the risk
unless I abandoned my purpose, an alternative on which I declined
to waste even a moment's consideration.
 All this time the tide was carrying me steadily up the estuary, my
own efforts being confined to keeping the boat as near the shore as
possible without losing the drift of the current. I made no attempt at
rowing, for, anxious as I was to reach the barge, I was afraid that if
Craill happened to be on deck the muffled creak of my rowlocks
would probably attract his attention. I wanted my arrival to be
absolutely unexpected, so that I could seize the chance of getting on
board before anyone was prepared to dispute my purpose.

Now that the critical moment was approaching I felt as cool as a

cucumber. Nothing steadies me like the prospect of immediate
action, and, although my hatred of Manning was as bitter as ever, it
no longer clouded or distorted my mind. It had, indeed, precisely the
opposite effect, bracing all my faculties to their clearest and sharpest
pitch.

I drifted along, listening intently, and keeping a constant look-out

over my shoulder. I knew by the curve of the shore that I must be
getting pretty near my goal, for just below where the barge lay the
estuary made a slight turn to the north.

Suddenly, a few feet ahead of me, I caught sight of a shadowy

anchor chain, rising from the water and stretching away into the
gloom. I was so close that in another second I should have been on
top of it, but a timely dig with my left-hand scull just saved me from
this calamity. Almost simultaneously the black outline of the barge
towered up above my head, and, gliding round the big blunt bow, I
came quietly alongside an empty dinghy which was bobbing about
on its painter at the foot of the accommodation ladder.

For a couple of seconds I remained perfectly still, gripping tight

hold of the dinghy's gunwale, and staring up at the railing above me.
I half expected to see Craill's ugly face come peering over the top,
for if he were on deck he must have heard the splash which I had
made in trying to avoid the chain. No such apparition materialised,
however, and with a slight pull at the painter I drew into the side
and rapidly hitched up my boat. The next moment I had mounted
the ladder and was safely on board.

As far as I could see, I appeared to have the whole place to

myself. It was very unlike an ordinary barge, for both fore and aft a
high "coach-house" roof had been built up above the deck, so as to
provide the cabins below with further light and head room. Between
the two, and right in front of where I was standing, a broad
companion-way led down to the interior.

I advanced on tiptoe, keeping a watchful eye on the fo'c'sle, from

which quarter I still feared that at any moment Craill might make an
inconvenient appearance. I had taken about three steps when a
sudden and unexpected noise brought me to a dead stop. It was a
queer sound, as though a small but heavily laden table had been
violently overturned. I stood quite still, listening intently; then, faint
but clear enough to send the blood racing through my heart, came
the half-stifled cry of a woman's voice.

One stride forward brought me to the edge of the companion-

way, and, clearing the short flight of steps with a single leap, I found
myself facing a closed door which apparently led through into the
cabins. It swung open as I turned the handle, revealing a narrow
passage, with another door at the end. On the farther side of this a
scuffle of some sort seemed to be taking place, and a voice which I
recognised as Manning's rose plainly above the confused jumble of
sounds.

If there were a championship for sprinting up corridors, I think I

should have some claim to be the holder. Anyhow, I covered the
distance in considerably less time than it takes to write the words,
and, wrenching round the brass knob with a savage jerk, applied the
full force of my shoulder to the panel. Instead of being locked, as I
expected, the door flew back suddenly on its hinges. How I saved
myself from falling I don't know; I can only imagine that the sight
which met my eyes momentarily stiffened every muscle and sinew in
my body.

Amid a debris of broken cups and plates Christine and Manning

were struggling together in the centre of the cabin. He had gripped
her in his arms, and, in spite of the efforts she was making to free
herself, he was pressing fierce kisses upon her face and neck. Beside
them on the carpet sprawled an overturned tea-table, with one leg
sticking up grotesquely in the air.

I had only the briefest glimpse of what was happening, for at the
noise of my entrance the tableau broke up abruptly. Manning raised
his head with a swift, astonished glance; and then, releasing his
hold, took a step backwards. Freed in this unexpected fashion,
Christine for an instant seemed to be on the point of collapsing. By a
great effort of will she was just able to reach the arm of the sofa,
where she stood panting and exhausted, her eyes fixed on me in a
kind of half-incredulous fear.

"You must forgive me if I am intruding, Manning," I said. "You

asked me to look you up, and I have taken you at your word."

Even at that moment the man's extraordinary coolness never

deserted him. Whatever thoughts and emotions were passing
through his mind, he seemed, after that first glance of surprise, to
accept the situation with perfect composure.

"I am afraid you have chosen rather an unfortunate time," he

replied. "If you would care to come back in half an hour I should be
charmed. At present I have another visitor."

"So I see," was my answer. "And, if you want to know, that is

precisely the reason why I intend to stop."

Manning's curious blue eyes narrowed dangerously.

"I don't quite understand," he observed in his silkiest voice.

 "It's very simple," I assured him. "This lady, whoever she is,
seems to have mistaken you for a gentleman." I turned to Christine.
"I hope I am not being officious," I added, "but if there is any way in
which I can be of use, please consider me entirely at your service."

She rose to the occasion with all the quickness and courage that I
had expected.

"If it's not troubling you too much," she said quietly, "I should
certainly be obliged if you would row me to the shore."

Manning stepped forward, addressing himself directly to her. His

manner was politeness itself, but there was no misunderstanding the
veiled threat behind his words.

"For various reasons," he said, "I think you had better allow me
that privilege."

I saw that Christine was hesitating, so I gave her no chance to

reply.

"I expect you know your way up on to the deck," I remarked. "If
you will excuse me I will be with you almost immediately."

For a second she still wavered. Then with the slightest possible
bow she crossed the cabin, and passed out silently into the passage.
I closed the door behind her.

Only once before in my life, when Bobby and I sighted our first
German submarine, have I felt the same peculiar sensation with
which I turned round and faced Manning. He stood where he was,
an amused and half-mocking smile playing round the corners of his
mouth.

"You mustn't let me detain you, Mr. Dryden," he said. "I am sure
you are full of the noblest sentiments, but it would be a pity to keep
the lady waiting."
 If his intention was to make me lose my temper it went woefully
astray.

"I shan't be long," I replied. "I am only going to give you a

damned good hiding."

I was looking straight at him as I spoke, and I saw the lightning

glance with which he measured his distance from a small oak
sideboard that was clamped against the wall.

"That's the worst of you primitive people," he drawled slowly,

"you always—" Then suddenly his eyes travelled past me to the door,
and a harsh, exulting cry broke from his lips.

"Go on, Craill," he shouted; "let him have it."

With most men the trick would probably have succeeded, but I
was too old a hand to be caught in that time-honoured fashion. As
he sprang for the sideboard I hurled myself after him, and at the
very second that he wrenched open the drawer my fist crashed
home full in his face. He lurched wildly backwards, and, stumbling
over the corner of the mat, fetched up against the wall with a thud
that shook the cabin.

Of the next two minutes I have a glorious but slightly confused

recollection. All the rage which I had been bottling up inside me
seemed to break loose at that first blow, and with a red mist in front
of my eyes I leaped in to finish the business.

Whatever else Manning might be, he was game to the last inch.
Hopelessly cornered, with blood streaming down his face, he yet
fought back at me like a trapped and maddened wolf. He fought,
too, with all the skill of a trained boxer, but science and courage
were little use against the mad fury which had suddenly taken
possession of me.
 Keeping him pinned against the wall, I smashed home punch
after punch without even troubling to guard myself. No one could
stand up long under such a hail of punishment, and although he
managed to land one or two blows there was not sufficient force in
them to have any effect on me. At last, with a terrific right-hand jolt
just below the heart, I sent him staggering sideways. He made a
desperate attempt to recover, but, seizing the chance, I let him have
my left bang on the point of the jaw, and down he went on to the
carpet—a sprawling bundle of arms and legs.

For a moment I stood there breathing heavily, my whole being

aflame with savage satisfaction. Stretched out motionless on his
back, Manning presented a pretty ghastly spectacle, but there was
not a spark of pity for him in my heart as I stared down into his
battered and bleeding face. Christine's cry for help, and the thought
of poor Satan creeping back alone to die in the darkness, were still
vividly in my mind, and, turning away, I strode across towards the
open drawer in the oak sideboard.

As I expected, the first thing I saw on looking inside was a

revolver. It was a Smith and Wesson of the heaviest service pattern,
and on taking it out I found that every chamber was loaded. I
wondered grimly what my fate would have been if Manning had
succeeded in reaching the drawer a second earlier. Possibly he had
only meant to protect himself, but in any case the weapon would
certainly be safer in my possession than in his, so, slipping it into my
pocket, I strolled back to where I had left him.

Just as I came up he gave a deep groan and opened his eyes.

"Oh," I said, "you're recovering, are you?"

He gazed at me vaguely, as if only half understanding my words;

then in a sudden rush the full memory of what had happened
seemed to return to him, and with a painful effort he raised himself
up on one elbow.
 "Next time you want to insult a girl," I suggested, "you had better
see that the cabin door is properly locked."

Lifting his sleeve, he tried to wipe away some of the blood which
was trickling down his face.

"I always thought you were a fool, Dryden," he said in a faint

voice. "Now I am sure of it." With great difficulty he struggled up a
little farther, and sank back against the wall.

"I know a damned cad when I see one, anyhow," I replied. "Not
that it needed much intelligence in the present case." I walked to
the door, and, pausing on the threshold, took out the revolver. "You
will forgive my borrowing this," I added. "I don't want to be shot in
the back as I leave the barge."

His blue eyes fixed themselves on mine with a cat-like

malevolence.

"You might have a worse ending," he said softly. "I am not a very
safe person to quarrel with."

There was a venom in his tone which spoke volumes as to his

sincerity, but unless it was intended to frighten me the warning was
a little superfluous.

"I shan't grumble at the bill," I replied. "I am always ready to pay
well, especially when I have really enjoyed myself."

Then, turning the handle, I stepped out into the passage, and
closed the door behind me.
 CHAPTER THIRTEEN
I found Christine standing on the deck at the head of the
companion-way.

She looked terribly pale, and as I approached she started towards

me with a little sob of relief.

"Oh, thank God!" she whispered. "Thank God you're safe!"

She took my hands, and a low cry escaped her at the sight of
their scarred and bleeding knuckles.

"It's nothing, darling," I said reassuringly. "That's only a little

blood from Manning's nose. He's got plenty left to go on with."

With something between a laugh and another sob she let her
head sink forward against my shoulder. "Don't think me a coward. I
—I'm not one really. It was just having to wait here and——"

"A coward!" I interrupted. "Why, I think you're the bravest girl

that ever stepped this earth."

I put my arms round her, and for one dear moment she lay there
passive and still, like a tired, contented child.

Then, suddenly releasing herself, she gazed round in a kind of

startled panic.

"But we mustn't stay here," she exclaimed hurriedly. "We must go

at once."

"I suppose you're right," I admitted with some reluctance.

"What's happened to that ruffian Craill?"
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!

ebookluna.com