Military Applications Of Data Analytics Data

Analytics Applications 1st Edition Kevin Huggins

Editor download

https://ebookbell.com/product/military-applications-of-data-
analytics-data-analytics-applications-1st-edition-kevin-huggins-
editor-11300924

Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Military Applications Of Data Analytics Huggins Kevin

https://ebookbell.com/product/military-applications-of-data-analytics-
huggins-kevin-9953730

Military Applications Of Modeling Selected Case Studies Francis P

Hoeber

https://ebookbell.com/product/military-applications-of-modeling-
selected-case-studies-francis-p-hoeber-44693610

The Democratic Offset Contestation Deliberation And Participation

Regarding Military Applications Of Ai Johannes Thumfart

https://ebookbell.com/product/the-democratic-offset-contestation-
deliberation-and-participation-regarding-military-applications-of-ai-
johannes-thumfart-49936434

Applications Of Mathematics And Informatics In Military Science

Nicholas J Daras

https://ebookbell.com/product/applications-of-mathematics-and-
informatics-in-military-science-nicholas-j-daras-4158188
Applications Of Operations Research And Management Science For
Military Decision Making 1st Ed 2019 William P Fox

https://ebookbell.com/product/applications-of-operations-research-and-
management-science-for-military-decision-making-1st-ed-2019-william-p-
fox-10797210

Military Displays Technology And Applications Daniel D Desjardins

Society Of Photooptical Instrumentation Engineers

https://ebookbell.com/product/military-displays-technology-and-
applications-daniel-d-desjardins-society-of-photooptical-
instrumentation-engineers-4723932

The Application Of The European Convention On Human Rights To Military

Operations Stuart Wallace

https://ebookbell.com/product/the-application-of-the-european-
convention-on-human-rights-to-military-operations-stuart-
wallace-10898690

Figurative Uses Of Animal Names In Latin And Their Application To

Military Devices A Study In Semantics 1912 Eugene Stock Mccartney

https://ebookbell.com/product/figurative-uses-of-animal-names-in-
latin-and-their-application-to-military-devices-a-study-in-
semantics-1912-eugene-stock-mccartney-55044232

Monitoring Metabolic Status Predicting Decrements In Physiological And

Cognitive Performance Committee On Metabolic Monitoring For Military
Field Applications

https://ebookbell.com/product/monitoring-metabolic-status-predicting-
decrements-in-physiological-and-cognitive-performance-committee-on-
metabolic-monitoring-for-military-field-applications-2134626
Military Applications
of Data Analytics
 Data Analytics Applications
Series Editor: Jay Liebowitz

PUBLISHED
Actionable Intelligence for Healthcare
Jay Liebowitz and Amanda Dawson

Data Analytics Applications in Latin America and Emerging Economies

by Eduardo Rodriguez

Big Data and Analytics Applications in Government:

Current Practices and Future Opportunities
by Gregory Richards

Big Data Analytics in Cybersecurity

by Onur Savas and Julia Deng

Data Analytics Applications in Education

by Jan Vanthienen and Kristoff De Witte

Intuition, Trust, and Analytics

by Jay Liebowitz, Joanna Paliszkiewicz, and Jerzy Gołuchowski

Research Analytics: Boosting University Productivity

and Competitiveness through Scientometrics
by Francisco J. Cantú-Ortiz

Big Data in the Arts and Humanities: Theory and Practice

by Giovanni Schiuma and Daniela Carlucci

Analytics and Knowledge Management

by Suliman Hawamdeh and Hsia-Ching Chang

Data-Driven Law: Data Analytics and the New Legal Services

by Edward J. Walters

Military Applications of Data Analytics

by Kevin Huggins

For more information about this series, please visit:

https://www.crcpress.com/Data-Analytics-Applications/book-series/CRCDATANAAPP
Military Applications
of Data Analytics

Edited by
Kevin Huggins

AN AUERBACH BOOK
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

© 2019 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed on acid-free paper

International Standard Book Number-13: 978-1-4987-9976-8 (Hardback)

This book contains information obtained from authentic and highly regarded sources. Reasonable efforts
have been made to publish reliable data and information, but the author and publisher cannot assume
responsibility for the validity of all materials or the consequences of their use. The authors and publishers
have attempted to trace the copyright holders of all material reproduced in this publication and apologize
to copyright holders if permission to publish in this form has not been obtained. If any copyright material
has not been acknowledged please write and let us know so we may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information storage
or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.copyright.
com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and
registration for a variety of users. For organizations that have been granted a photocopy license by the
CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com
Contents

Preface ...............................................................................................................vii
Acknowledgments ..............................................................................................ix
Editor ..................................................................................................................xi
Contributors .....................................................................................................xiii
1 Bayesian Networks for Descriptive Analytics in Military
Equipment Applications .........................................................................1
DAVID AEBISCHER

2 Network Modeling and Analysis of Data and Relationships:

Developing Cyber and Complexity Science..........................................29
CHRIS ARNEY, NATALIE VANATTA, AND MATTHEW SOBIESK

3 Context for Maritime Situation Awareness ..........................................47

ANNE-LAURE JOUSSELME AND KARNA BRYAN

4 Harnessing Single Board Computers for Military Data Analytics.......63

SUZANNE J. MATTHEWS

5 Data Analytics and Training Effectiveness Evaluation ........................79

MICHAEL SMITH, SUSAN DASS, CLARENCE DILLON, AND
RODNEY LONG

6 Data Analytics for Electric Power and Energy Applications ................97

AARON ST. LEGER

7 The Evolution of Environmental Data Analytics in Military

Operations ..........................................................................................113
RALPH O. STOFFLER

8 Autoregressive Bayesian Networks for Information Validation and

Amendment in Military Applications ................................................129
PABLO IBARGÜENGOYTIA, JAVIER HERRERA-VEGA,
URIEL A. GARCÍA, L. ENRIQUE SUCAR, AND EDUARDO F. MORALES

v
vi ◾ Contents

9 Developing Cyber-Personas from Syslog Files for Insider Threat

Detection: A Feasibility Study ............................................................151
KEVIN PURCELL, SRIDHAR REDDY RAVULA, ZIYUAN HUANG,
MARK NEWMAN, JOSHUA RYKOWSKI, AND KEVIN HUGGINS

10 Analytics for Military Training in Virtual Reality Environments .....167

MIGUEL PÉREZ-RAMÍREZ, BENJAMIN EDDIE ZAYAS-PÉREZ,
JOSÉ ALBERTO HERNÁNDEZ-AGUILAR, AND NORMA JOSEFINA
ONTIVEROS-HERNÁNDEZ

Index ...........................................................................................................187
Preface

Military organizations around the world are normally huge producers and consumers
of data. Accordingly, they stand to gain from the many benefits associated with data
analytics. However, for leaders in defense organizations—either government or
industry—accessible use cases are not always available. The reason that I embarked on
this book was to provide a diverse collection of detailed use cases that could serve as a
representation of the realm of possibilities in military data analytics. I trust you will find
it useful, and I look forward to hearing about your successful analytics projects.

vii
Acknowledgments

This, like most other works, was truly a team effort. First, I’d like to thank my
family for their patience and encouragement. Additionally, I’m grateful to Jay
Liebowitz for giving me this opportunity. It was definitely an adventure. I am also
indebted to my colleagues at Harrisburg University for their insights and recom-
mendations. Finally, I would like to thank my awesome team of reviewers. They
generously shared their expertise to raise the quality of this work significantly. The
review team consisted of the following experts:

◾ Nathaniel Ashby, PhD

◾ Feyzi Bagirov
◾ Rand Ford, PhD
◾ Philip Grim
◾ Marvine Hamner, PhD
◾ Pablo Ibargüengoytia, PhD
◾ Suzanne J. Matthews, PhD
◾ Arnie Miles
◾ Glenn Mitchell, MD
◾ Ali Motamedi, PhD
◾ Stephen Penn, PhD
◾ Kevin Purcell, PhD
◾ Mandana Rezaeiahari, PhD
◾ Roozbeh Sadeghian, PhD
◾ Aaron St. Leger, PhD
◾ Doaa Taha, PhD

ix
Editor

Kevin Huggins, PhD, is professor of Computer

Science and Analytics at Harrisburg University of
Science and Technology, Harrisburg, Pennsylvania.
He is also a retired military officer who spent the
early part of his career in military intelligence,
with extensive experience in Latin America. The
remainder of his career was dedicated to academia,
primarily as a faculty member in the Department of
Electrical Engineering and Computer Science at the
U.S. Military Academy. While there, Dr. Huggins
served as the director of Research in Network
Science as well as the director of the Information Technology Program.
Additionally, Dr. Huggins was a visiting scientist at the École de Techniques
Avancées in Paris, France, where he studied parallel algorithms for multiprocessor
system-on-chip (MPSoC) architectures. His current research interest lies at the
intersection of data science and information security, exploring novel ways of secur-
ing computing systems by leveraging the enormous amounts of available data.
He holds a PhD in computer science from Mines Paris Tech.

xi
Contributors

David Aebischer is chief of the Special Operations

Branch of the U.S. Army Communications
Electronics Command (CECOM) Training
Support Division (TSD) and is the inventor and
leader of the CECOM Equipment Diagnostic
Analysis Tool, Virtual LAR (CEDAT VLAR)
project, which uses advanced operations research
techniques to develop diagnostic applications for
Army combat equipment. Aebischer is a veteran
of the U.S. Air Force with over 36 years of gov-
ernment service—including over 4 years deployed
to Iraq and Afghanistan supporting soldiers and their equipment. Aebischer holds
a master’s of business administration and technology management degree from
Monmouth University and is pursuing a doctoral degree in applied management
and decision science. Aebischer is a class of 2016 Franz Edelman laureate and his
VLAR team was selected as a finalist for the 2016 Franz Edelman award for sig-
nificant contribution to operations research and innovation in applied analytics.
He has extensive technical training and field experience in electrical, electro-
mechanical, and electronics systems and has advanced training on Bayesian net-
works and Cognitive Task Analysis (CTA) processes. Aebischer’s current research
interest is in using Bayesian networks to optimize fuel consumption and equip-
ment readiness.

xiii
xiv ◾ Contributors

Chris Arney is a professor of mathematics at the

U.S. Military Academy. He helped to develop a
network science minor at the academy and often
teaches courses in modeling, interdisciplinary
problem solving, and differential and difference
equations. He served in the active duty Army for
30 years, was a dean and vice president at a college,
and directed and managed research programs in
network science, mathematics, and cooperative
systems. He is the founding director of the
Interdisciplinary Contest in Modeling.

Karna Bryan has worked for over 20 years on

data-driven decision-making research topics in
various domains including underwater warfare,
maritime domain awareness, and civil aviation.
She currently works as an operations research ana-
lyst for the Department of Transportation in
Cambridge, Massachusetts. Previously, she worked
for 18 years at the Center for Maritime Research
and Experimentation and was program manager
for the Maritime Security program. She holds a
master’s degree in statistics from Yale University.

Susan Dass is a senior instructional designer and

technical specialist at ICF. She is also an adjunct
professor at George Mason University. Dass
received a PhD in education and an MEd in
instructional design. She has more than 17 years
of experience in instructional design supporting
both private and government entities. Dass has
written many publications, including coauthoring
three book chapters that explore the design of
help-seeking and self-regulated learning skills in
3D learning environments as well as faculty adop-
tion of virtual worlds as a learning environment.
Her current projects include designing 3D serious
games for the medical industry and exploring data analytics for learner perfor-
mance and course improvements.
 Contributors ◾ xv

Clarence Dillon is a senior consultant with ICF

and has been working with data, strategic plan-
ning, and management for 24 years, including
time working with the Office of the Secretary of
Defense, the Office of the Secretary of the Army,
the Army staff, and the Air Force staff. He is a
computational social science PhD candidate in the
Department of Computational and Data Sciences
at George Mason University. He holds an MSocSci
from the University of Tampere, Finland, and a BA
from the University of Colorado at Boulder.

Uriel A. García received a BsSc degree in engi-

neering in computer systems from the Technological
Institute of Comitan, Chiapas, Mexico. Currently,
he is pursuing a master’s degree in optimization and
applied computing at the Autonomous University
of Morelos, Mexico. His research interests are
focused on machine learning, artificial intelligence
in renewable energy, and probabilistic graphical
models.

José Alberto Hernández-Aguilar obtained a PhD

from Autonomous University of Morelos (UAEM)
in 2008, a master’s degree (summa cum laude) in
business management (2003) from the Universidad
de las Américas Puebla (UDLAP), and a BSc in
computer engineering from the National University
of Mexico (UNAM). He has been a full-time
professor since 2010, at the Accounting,
Management and Computer Sciences School at
UAEM. His main interest areas are databases,
artificial intelligence, online assessment systems,
data mining, and data analytics. He is a member of the organizing staff of different
international congresses such as the Healthcare Infection Society (HIS) and the
Mexican International Conference on Artificial Intelligence (MICAI).
xvi ◾ Contributors

Javier Herrera-Vega received BSc and MSc degrees

in computer science from the Autonomous
University of Puebla, Mexico. Currently, he is
finishing a PhD degree in computer science from
the National Institute of Astrophysics, Optics and
Electronics in Puebla, Mexico. His research inter-
ests are focused on inverse problems, probabilistic
graphical models, and medical imaging.

Ziyuan Huang is an experienced SQL server

developer with experience in the financial services
and healthcare industries. He also has experience
in analytical skills, SQL Server R services, SQL
Server Reporting Services (SSRS), Extract,
Transform, Load (ETL), databases, and Microsoft
Access. Huang is currently a PhD candidate in
data sciences at Harrisburg University of Science
and Technology.

Pablo Ibargüengoytia received a degree in elec-

tronic engineering from the Universidad
Autónoma Metropolitana (UAM). He received a
master’s degree from the University of Minnesota.
Finally, he received a PhD in computer science from
the University of Salford in the United Kingdom. He
has been a full-time investigator at the National
Institute of Electricity and Clean Energies (INEEL)
since 1983. He has been a professor of intelligent
systems at the ITESM Campus Cuernavaca since
1998. He has also directed several BSc, MSc, and
PhD theses. He is a senior member of the Institute of
Electrical and Electronics Engineers (IEEE), a
member of the Mexican Society of Artificial Intelligence, and a member of the
National Researchers System (SNI) in Mexico.
 Contributors ◾ xvii

Anne-Laure Jousselme is with the NATO Centre

for Maritime Research and Experimentation
(CMRE) in La Spezia, Italy, where she conducts
research activities on maritime anomaly detection,
high‐level and hard and soft information fusion,
reasoning under uncertainty, information quality
assessment, and serious gaming approaches. She is
a member of the board of directors of the
International Society of Information Fusion (ISIF)
where she serves as vice president of membership.
She serves on the board of directors and steering
committee of the Belief Functions and Applications
Society (BFAS) as well as on program committees
of the FUSION and BELIEF conferences. She co-organized the Information Fusion
conference in Quebec City in 2007 (tutorial chair) and served on the organizing
committee of Fusion 2015 in Washington (international co chair). She is associate
editor of the Perspectives on Information Fusion magazine. She has been an adjunct
professor at Laval University, Quebec, Canada for 10 years.

Rodney Long is a science and technology manager

at the Army Research Laboratory in Florida and is
currently conducting research in adaptive training
technologies. Long has a wide range of simulation
and training experience spanning 29 years in the
Department of Defense and has a bachelor’s degree
in computer engineering from the University of
South Carolina, as well as a master’s degree in
industrial engineering from the University of
Central Florida.
xviii ◾ Contributors

Suzanne J. Matthews is an associate professor of

computer science in the Department of Electrical
Engineering and Computer Science at the U.S.
Military Academy. She received her BS and MS
degrees in computer science from Rensselaer
Polytechnic Institute and her PhD in computer sci-
ence from Texas A&M University. Her honors
include a dissertation fellowship from Texas A&M
University, a master teaching fellowship from
Rensselaer Polytechnic Institute, and memberships
in the honor societies of Phi Kappa Phi and Upsilon
Pi Epsilon. She is a member of the Association of
Computing Machinery, and an affiliate of the
IEEE Computer Society. Her research interests
include parallel computing, single board comput-
ers, data analytics, experimental computing, and computational biology.

Eduardo F. Morales received a PhD from the

Turing Institute at the University of Strathclyde, in
Scotland. He has been responsible of more than 25
research projects and has written more than 150
peer-review papers. He was an invited researcher at
the Electric Power Research Institute (1986), a
technical consultant (1989–1990) at the Turing
Institute, a researcher at the Instituto de
Investigaciones Electricas (1986–1988 and 1992–
1994), and at ITESM–Campus Cuernavaca
(1994–2005). He is currently a senior researcher at
the Instituto Nacional de Astrofísica, Óptica y Electrónica (INAOE) in Mexico
where he conducts research in machine learning and robotics.

Mark Newman is an experienced senior engineer

with a demonstrated history of working in a vari-
ous data-heavy industries. Skilled in Azure, .Net,
Microsoft SQL, and Microsoft Dynamics CRM,
Newman is a strong engineering professional with
an MS focused in computer science from Hood
College. He is currently a PhD candidate in data
sciences at Harrisburg University of Science and
Technology.
 Contributors ◾ xix

Norma Josefina Ontiveros-Hernández holds an

MSc in computer science (1995) from the Centro
Nacional de Investigación y Desarrollo Tecnológico
(CENIDET), Mexico (software engineering) and a
BSc in computer science from Durango Institute of
Technology (1984). She has lectured at the Mexico
National System of Technology since 1986, at
Durango Institute of Technology from 1986 to 1997,
and at Zacatepec Institute of Technology since 1997.
Her primary areas of interest are databases, artificial
intelligence, assessment systems, and analytics.

Miguel Pérez-Ramírez holds a PhD in computer

science (artificial intelligence from the University
of Essex, UK; MSc in computer science (software
engineering) from Centro Nacional de Investi-
gación y Desarrollo Tecnológico (CENIDET),
Mexico, and a BSc in computer science from
Benemérita Universidad Autónoma de Puebla
(BUAP), Mexico. He has been a research member
of the Mexican National Institute for Electricity
and Clean Energies, since 1992, in the informa-
tion technologies department. He has partici-
pated in different projects, which include
development of information systems and data warehouses. He has also worked
with expert systems and knowledge management. Since its creation in 2003, he
has led the Virtual Reality Group, which focuses on the development of training
systems for the energy sector, where the learning context is high-risk tasks, such as
maintenance to energized lines within the electrical domain. These training systems
have been installed and used across the entire country in the Comisión Federal
de Electricidad (CFE), the Mexican electricity utility company. He is a member
of the IEEE.
xx ◾ Contributors

Kevin Purcell is the chief scientist at WildFig, a

data science and analytics consultancy. Purcell has
an MS and PhD in the life sciences and conducts
research focused on applying computational and
statistical methods to understand disturbance in
complex systems. He is currently an adjunct faculty
member in the Analytics Department at Harrisburg
University, teaching graduate courses in data
science. He currently heads WildFig’s first research
and experiential learning laboratory, with a mission
to develop novel analytical approaches with com-
mercial applications and cultivate a new generation
of data scientists.

Sridhar Reddy Ravula is an entrepreneur, asset

manager, and business consultant. He is passionate
about solving business problems and providing
cost-effective solutions using emerging technolo-
gies. He holds an MBA, with specializations in
finance and operations, and an MS in data analyt-
ics. He is currently pursuing a PhD in data science
at Harrisburg University. His research interests
are applications of natural language processing and
AI in the domains of cybersecurity, finance, and
healthcare.

Joshua Rykowski is the director for Strategic

Initiatives for the Cyber Protection Brigade at the
U.S. Army Cyber Command.
 Contributors ◾ xxi

Michael Smith is a senior technical specialist

with ICF and has over 14 years of experience
in data analytics, strategic planning, and
management. Smith advises government clients
on how to adapt analytics practices to improve
organizational performance. He has a BA in
international economics from Longwood
University and a master’s degree in public policy
from Georgetown University.

Matthew Sobiesk has a BA in mathematics and

Near Eastern studies from Cornell University and
is currently an operations research PhD student at
the Massachusetts Institute of Technology. His
research interests include studying machine
learning through an optimization lens.

Aaron St. Leger is an associate professor at the

U.S. Military Academy (USMA). He received
BSEE, MSEE, and PhD degrees from Drexel
University. His research and teaching interests
include alternative energy, electric power sys-
tems, modeling, and controls. He has published
more than 50 papers on these subjects. His recent
work focuses on integrating alternative energy
and demand response controllers to improve
electric power systems for military forward oper-
ating bases, and anomaly detection in smart
grids. He is the director of the Electrical Power
Systems, Alternative Energy, and Operational Energy Laboratories at USMA.
xxii ◾ Contributors

Ralph O. Stoffler is a member of the Senior

Executive Service, is the director of Weather, dep-
uty chief of staff for Operations, Headquarters,
U.S. Air Force, Washington, DC. In this capacity,
he is responsible for the development of weather
and space environmental doctrine, policies, plans,
programs, and standards in support of Army and
Air Force operations. He oversees and advocates
for Air Force weather resources and monitors the
execution of the $320 million per year weather
program and is the functional manager for 4300
personnel.

L. Enrique Sucar has a PhD in computing from

Imperial College, London, an MSc in electrical
engineering from Stanford University, and a BSc in
electronics and communications engineering from
ITESM, Mexico. Currently, he is a senior research
scientist at the National Institute for Astrophysics,
Optics and Electronics, Puebla, Mexico. Sucar is a
member of the National Research System and the
Mexican Science Academy. He is the associate editor
of the Pattern Recognition and Computational
Intelligence journal, and a member of the advisory
board of International Joint Conferences on Artificial Intelligence (IJCAI). He was
awarded the National Science Prize by the Mexican president in 2016.

Natalie Vanatta is a U.S. Army cyber officer and

currently serves as the deputy chief of research at
the Army Cyber Institute. Here, she focuses on
bringing private industry, academia, and
government agencies together to explore and
solve cyber challenges facing the U.S. Army in
the next 3 to 10 years in order to prevent strate-
gic surprise. She holds a PhD in applied
mathematics, as well as degrees in computer
engineering and systems engineering. Vanatta
has also served as distinguished visiting professor
at the National Security Agency.
 Contributors ◾ xxiii

Benjamin Eddie Zayas-Pérez is a researcher at the

Supervision Department of Electrical Research
Institute. He obtained a PhD in computer sciences
and artificial intelligence from Sussex University,
following predoctoral studies in human-centered
computer systems and computer sciences. His
research interests include human–computer inter-
action and virtual reality environments for learning
and training purposes. Currently, he is initiating
applied research in Big Data analytics. He is mem-
ber of the IEEE.
Chapter 1

Bayesian Networks for

Descriptive Analytics
in Military Equipment
Applications
David Aebischer

Contents
Introduction ..........................................................................................................1
DSEV ...................................................................................................................4
Frame the Problem ................................................................................................5
Ready the Experts..................................................................................................5
Pre-DSEV ...........................................................................................................12
Define .................................................................................................................13
Structure .............................................................................................................14
Elicit ...................................................................................................................17
Verify ..................................................................................................................21
Critical Decision Method (CDM) .......................................................................23
References ...........................................................................................................28

Introduction
The lives of U.S. soldiers in combat depend on complex weapon systems and
advanced technologies. Operational command, control, communications, computers,
intelligence, surveillance, and reconnaissance (C4ISR) systems provide soldiers the

1
2 ◾ Military Applications of Data Analytics

tools to conduct operations against the enemy and to maintain life-support. When
equipment fails, lives are in danger; therefore, fast and accurate diagnosis and repair of
equipment may mean the difference between life and death. But in combat conditions,
the resources available to support maintenance of these systems are minimal.
Following a critical system failure, technical support personnel may take days to
arrive via helicopter or ground convoy, leaving soldiers and civilian experts exposed
to battlefield risks. What is needed is a means to translate experiential knowledge and
scientific theory—the collective knowledge base—into a fingertip-accessible, artificial
intelligence application for soldiers. To meet this need, we suggest an operations research
(OR) approach to codifying expert knowledge about Army equipment and applying
that knowledge to troubleshooting equipment in combat situations. We infuse a
classic knowledge-management spiral with OR techniques: from socializing advanced
technical concepts and eliciting tacit knowledge, to encoding expert knowledge in
Bayesian Belief Networks, to creating an intuitive, instructive, and learning interface,
and finally, to a soldier internalizing a practical tool in daily work. We start development
from this concept with counterfactuals: What would have happened differently had a
soldier been able to repair a piece of critical equipment? Could the ability to have made
a critical diagnosis and repair prevented an injury or death? The development process,
then, takes on all the rigor of a scientific experiment and is developed assuming the
most extreme combat conditions. It is assumed that our working model is in the hands
of a soldier who is engaged with the enemy in extreme environmental conditions,
with limited knowledge of, and experience with, the equipment that has failed, with
limited tools, but with fellow soldiers depending on him (her) to take the necessary
steps to bring their life-saving equipment back into operation and get the unit out of
danger (Aebischer et al., 2017). The end product is a true expert system for soldiers.
Such systems will improve readiness and availability of equipment while generating
a sustainable cost-savings model through personnel and direct labor reductions. The
system ensures fully functional operation in a disconnected environment, making it
impervious to cyber-attack and security risks. But most importantly, these systems
are a means to mitigate combat risk. Reducing requirements for technical support
personnel reduces requirements for helicopter and ground-convoy movements, and
this translates directly to reductions in combat casualties (Bilmes, 2013).
In On War, Carl Von Clausewitz wrote, “War is the realm of uncertainty; three
quarters of the factors on which action is based are wrapped in a fog of greater or lesser
uncertainty” (Kiesling, 2001). Clausewitz’s “Fog of War” metaphor is well-made and
oft-used, becoming so popular as to become part of the modern military lexicon.
Analysis of his writing has, naturally, led to discussion about how information tech-
nology can be brought to bear to reduce battlefield entropy. From this familiar terri-
tory, let us deliberately examine data analytics in context with combat uncertainty. Let
us further refine that to the combat soldier and his equipment—in the most extreme
scenario of enemy threat, geography, and environment—and determine what data
analytics can put at a soldier’s fingertips. This task requires a fine balance of complex-
ity and usability. It must be mathematically precise, yet infused with tacit knowledge.
 Bayesian Networks for Descriptive Analytics in Military ◾ 3

It must be exhaustive, as the consequences of failure can be dire. Most of all, the task
must establish the primacy of soldiers at the point of need. We submit that this is a
doable and necessary task and an opportunity to apply a unique analytics solution
to this, and other, multivariate problems. This chapter will provide an introductory
guide for tapping into an expert knowledge base and codifying that knowledge into
a practical, useful, and user-friendly working model for diagnostics. In order, we will
provide a general description of all the tools, techniques, and resources needed to con-
struct the model, and then proceed through a practical example of model building.
The category of Descriptive Analytics, and diagnostics as a sub-category, serves
well to frame our initial efforts. Korb and Nicholson (2011) describe:
By building artifacts which model our best understanding of how
humans do things (which can be called descriptive artificial intelligence)
and also building artifacts which model our best understanding of what
is optimal in these activities (normative artificial intelligence), we can
further our understanding of the nature of intelligence and also produce
some very useful tools for science, government, and industry. (p. 21)
Causal analysis is at the heart of descriptive analytics. To perform this analysis,
we will need a couple of tools: a knowledge engineering tool to capture knowledge
about the domain of interest, and Bayesian networks to codify and represent that
knowledge. The knowledge engineering tool and the Bayesian network tool work
hand-in-hand, so they will be detailed as such. We will also need to designate a spe-
cific C4ISR system as a use case and as a practical example to demonstrate how each
tool works separately and in combination with the other to model complex systems.
For this, we will use the Army’s diesel engine-driven tactical generator, the primary
source of power for Army Command Post operations. Generators meet our needs in
terms of multi-domain complexity (electrical, electronics, and electro-mechanical)
and in terms of ubiquity in combat environments. We will look at these tools indi-
vidually in some detail, but we will focus more on how they overlap and how they
integrate around our diesel generator use case. All of this goes toward learning how
to work with limited or nonexistent data. Our richest source of knowledge is experts,
and our actual data-generating process is from facilitating effective and efficient
knowledge elicitation sessions with those experts. Our first tool will help us navigate
the challenging, but rewarding, world of working with experts and expertise.
It is easy to recall instances where we have witnessed experts at work. The great
artists—composers, painters, poets, musicians—seem to possess something that
transcends talent, and they describe their art as something that occurs when their
ideas have gone somewhere that was not intended. The process by which they cre-
ate and do is not describable, existing only in their heads. It is much the same with
equipment experts—possessing a blend of experiential and theoretical knowledge
and an innate ability to apply that knowledge in ways few of us can comprehend.
Coming to understand, and make explicit, even a fraction of this knowledge is the
goal of knowledge engineering and the process of knowledge elicitation (Figure 1.1).
4 ◾ Military Applications of Data Analytics

Human expert
Dialogue
Knowledge
engineer
Explicit
Knowledge base
of expert system

Figure 1.1 Knowledge engineering.

This process is painstaking, labor-intensive, and fraught with the danger of

introducing many different biases, but the benefits of a well-constructed and exe-
cuted knowledge elicitation make it worth the effort and outweigh the risks. It is in
this process that we tease out the heuristic artifacts buried in the brains of experts.
It is these artifacts that are carefully crafted into the artificial intelligence network
that sits in a soldier’s hands.
The knowledge engineering process has its roots firmly in Cognitive Task
Analysis (CTA). CTA involves a number of processes by which decision making,
problem solving, memory, and judgment are all analyzed to determine how they are
related in terms of understanding tasks that involve significant cognitive activity. For
our purposes, CTA supports the analyses of how experts solve problems.

DSEV
Define, Structure, Verify, Elicit (DSEV) is a field-proven method for developing
Bayesian models strictly from expert knowledge with a unique blend of soft and hard
skills. DSEV is infused with techniques that mitigate bias and ensure the elicitation
process aggregates expert judgments into a unified network. We will first outline the
methods we use to execute each phase of the model and then, later, demonstrate how
it is applicable, flexible, and scalable to a practical problem domain.
DSEV starts, stops, and cycles with experts. In both the theory and practice of
Bayesian networks, assumptions—expert knowledge about any specific problem
domain—are a necessary component. But it is inherently difficult for experts to explain
what they know, and equally difficult for non-experts to understand what experts
are saying. What is needed is a formalism for bridging the gap between experts and
non-experts and codifying complex technical concepts into graphical structure and
probability distributions. Our DSEV objective is an exhaustive probabilistic model of
how the generator can fail—with expert-based pathways through each failure mode.
But DSEV requires unique facilitation skills, so, prior to starting the DSEV process, we
want to do some preparatory tasks so that facilitators and experts—the whole team—
understand the tools, tasks, and processes associated with Figure 1.2 (Hepler et al., 2015).
 Bayesian Networks for Descriptive Analytics in Military ◾ 5

Frame the Ready the Build the Validate and Use the
problem experts network update model

Define Structure

DSEV

Verify Elicit

Figure 1.2 Define, Structure, Elicit, Verify (DSEV).

Frame the Problem

This part of the DSEV includes establishing and scoping objectives, identifying use
cases, and conducting preliminary research. While our overarching objective is an
exhaustive probabilistic model of generator failure modes, we need to properly scope
that objective for a discrete DSEV session. In practice, this means breaking the generator
system into logical subsystems and working on modeling each of those subsystems in
turn. Defining use cases then becomes a straightforward task related to the subsystem:
operation of the system in normal conditions, operation under abnormal/extreme
environmental conditions, and operation under other special conditions specific to
the combat scenario. The last step in this pre-phase block of the process is to ensure
everybody on the team is speaking the same technical language. It is incumbent upon
the facilitators to be conversant with the subject matter (theory, operation, technical
terms, acronyms) and to know what questions will be most effective. Background
research, training classes, and seminars are all ways for facilitators to ready themselves to
talk to experts. CTA techniques, such as the critical decision method (CDM), provide
a framework for how to structure the conversation. We will delve into details of CDM
during the practical modeling portion of the chapter. With objectives and use cases
agreed to and a well-versed facilitator team in place, it’s time to organize the group.

Ready the Experts

Deciding how many, what types, and what combination of experts to use in an
elicitation is the first challenge. Statistically, four to five experts is the right amount
to mitigate some biases and to avoid bogging the discussion down with too many
points of view and too much interaction between members. This is directly related
to the number of possible social interactions in the group dynamic. Above a group
of five, this number grows exponentially (Table 1.1).
6 ◾ Military Applications of Data Analytics

Table 1.1 Group Dynamics versus Group Size

250

Number of social interactions

200

150

100

50

0
0 2 4 6 8
Group size

Research on practical group events bears that out, with group member ratings
converging to between four and five members for optimum efficiency (Hackman and
Vidmar, 1970). It is usually best to choose “complementary experts”: some with specific
hands-on experience with the equipment, and some with more academic or theoretical
perspectives. It also helps to have an expert with significant general knowledge in the
field (domain experience) but not specific experience on the system. This combination
provides a good setting for “productive friction” between experts, stimulating the
CDM “deepening” process that reveals new insights into the domain. The expert team
is combined with a facilitation team consisting, normally, of two knowledge engineers,
with one focusing on doing the direct elicitation—extracting and interpreting the
expert knowledge—and one focusing on building a model based on the elicitation and
interpretation. Regular dialogue between knowledge engineers to clarify statements is
expected, and it is often helpful to reverse roles at some interval during the elicitation.
With a well-organized and well-balanced team in place, the next step is to ensure the
experts know how their knowledge will be used and how will it be represented.
Bayesian networks (BNs) are probabilistic models of how we believe our “world”
works (Pearl, 1988). BNs are based on Bayes theorem, an equation that calculates
the probability of an event (A) given we know an event (B) has occurred, or P (A|B)
(Figure 1.3).
Bayes Theorem
• Bayes theorem is most commonly used to estimate the
state of a hidden, causal variable H based on the
measured state of an observable variable D:
Likelihood
Prior
p(D|H)p(H)
p(H|D) =
p(D)

Posterior Evidence

Figure 1.3 Bayes theorem.

 Bayesian Networks for Descriptive Analytics in Military ◾ 7

Cause Effect

Figure 1.4 Causal relationships in a Directed Acyclic Graph (DAG).

BNs have a qualitative component and a quantitative component. The

qualitative, or structural, component is a Directed Acyclic Graph (DAG). DAGs
are constructed using nodes, representing variables in the domain of interest, and
arcs, representing the causal (temporal) relationship between variables (Figure 1.4).
Relationships between variables can be modeled using any of three basic
structures, individually or in combination: (1) indirect connection (A causes B,
which causes C); (2) common cause (A can cause B and C); or (3) common effect
(A or B can cause C). Each structure has unique characteristics that guide the
process of choosing how to represent specific relationships (Figure 1.5).

1 Note: These symbols are used

Cause A2 Intermediate B Effect C3 throughout to indicate where the
structure allows or blocks
information flow in both directions.
Cause A1 Cause A Cause B
2
3

Effect B Effect C2 Effect C

Figure 1.5 Basic structures.

In an indirect connection structure, A and C are marginally dependent (each

variable is in its natural distribution with no evidence entered). For example,
knowing that the ambient temperature is cold changes our belief about an engine
starting failure through an open path between A (Ambient Temperature) and C
(Engine Cranks but Will not Start) through B (Battery) (Figure 1.6).
Critical to the understanding of Bayesian networks is the concept of
omnidirectional inference: the path is open in both directions. In the small network

“The arrows in the diagram represent real

Cause A2 Intermediate B Effect C3 causal connections and not the flow of
information during reasoning ... Reasoning
processes can operate on Bayesian
networks by propagating information in
Ambient Battery[1] Engine Cranks any direction.” (Conrady, Jouffe)
Temperature[1] but w/n Start[3]

Figure 1.6 Indirect connection structure.

8 ◾ Military Applications of Data Analytics

above, the arcs do indicate a cause and effect, or temporal, ordering. But information
flows in both directions—both with and against the arc direction. So knowing
the Engine Cranks but w/n Start and setting evidence accordingly will influence
the probability distribution of Battery and Ambient Temperature, just as setting
evidence on Ambient Temperature will influence the distribution of Battery and
Engine Cranks but w/n Start (Figure 1.6).
However, if we know B (Battery state of charge is LOW, as in Figure 1.7), and set
evidence accordingly, this closes the information path in both directions. We say, then,
that A and C are independent given (conditioned on) B, or that setting any evidence on
B makes A and C independent. This is intuitive because if we know the Battery state
of charge is low or high, then knowing Ambient Temperature is no longer useful with
respect to the “Engine Cranks but w/n Start” variable, and vice versa. This structure
type is particularly useful for modeling electrical wiring or fiber optic runs, as it will
translate an expert’s thought such as “a reading at test point B will make a test point A
reading unnecessary with respect to C” into a compact structural model.

LOW Conditioning on, or setting evidence on,

Ambient Battery[1] Engine Cranks an intermediate variable in an indirect
Temperature[1] but w/n Start[3] connection structure closes the
information path in both directions

Figure 1.7 Indirect connection—Evidence.

In the common cause structure (A can cause B or C), B and C are marginally
dependent. Information can flow from B to C through A, and from C to B through A.
Consider a case where the “Cylinder Pressure” variable (A) can be the cause of both the
“Engine Cranks but w/n Start” (B) and “Compression Test” (C) variables. Knowing
the state of the “Engine Cranks but w/n Start” variable will certainly change our
reasoning (and probability distribution) on the “Compression Test” variable, and vice
versa (Figure 1.8).

Cause (A) Cylinder Pressure With no evidence entered,

information can flow in both
directions from effect to effect in
a common cause structure. The
effects are marginally dependent.
Engine Cranks Compression Test (B ⫫ C)
Effect 1 (B) Effect 2 (C)
but w/n Start

Figure 1.8 Common cause (confounder) structure.

 Bayesian Networks for Descriptive Analytics in Military ◾ 9

However, if we already know the state of the “Cylinder Pressure” variable (we
know the cause) and enter evidence accordingly, this blocks the path between
effects through the cause in both directions. This is intuitive because knowing
the state of “Cylinder Pressure” makes knowing the state of “Engine Cranks but
w/n Start” no longer useful to our belief about “Compression Test,” and vice versa.
B and C are, then, independent given evidence for A (Figure 1.9).

LOW Given evidence for the cause in

the common cause structure
makes the effects independent
given the cause (B⫫C|A)
Cylinder Pressure

Engine Cranks Compression Test

but w/n Start

Figure 1.9 Common cause (confounder)—Evidence.

The common cause structure translates expert’s statements such as “low cylin-
der pressure can cause engine starting problems and a marginal or low compression
test” directly into a compact model that reasons like the expert; that is, what tests
are most economical and most sensitive?
In the common effect structure, A or B can cause C. Comment effect structures
are also known as colliders. Information cannot flow from A to B in a collider
because C blocks the path of information in both directions. A and B are, then,
marginally independent. Consider a case where both the “Battery” and the “Fuel
System” could cause “Engine Cranks but w/n Start.” This is intuitive, as knowing
the state of “Battery” tells us nothing about “Fuel System” (Figure 1.10).

Battery Fuel System

Cause 2 (B)
Cause 1 (A)

Engine Cranks
Effect (C)
but w/n Start

With no evidence entered, no information flows between Cause A

and Cause B in either direction because C blocks the path. Cause A
and B are then marginally independent in the common
effect/collider structure (A ⫫ B)

Figure 1.10 Common effect (collider) structure.

10 ◾ Military Applications of Data Analytics

But, if we know the state of “Engine Cranks but w/n Start” and we enter evidence
accordingly, then this opens the path from A to B, making them dependent given
evidence for C. This is also intuitive. If we know the effect and we know the state of
one of the possible causes, then we can reason about the state of the other cause. In
network terms, if we know that “Battery” state is “low” and we have evidence that
“Engine Cranks but w/n Start” is “Yes,” then we can reason that “Fuel System” is less
likely to be the cause. This is known as “explaining away” and is a powerful tool for
articulating expert knowledge. The collider structure translates expert’s statements
such as “there are multiple causes for the engine to crank but not start” into a model
that reasons like an expert; that is, if A is true and we know C, then it is unnecessary
to test B (Figure 1.11).

LOW
With evidence entered for the effect and one
Battery Fuel System
cause, information flows between Cause A and
Cause B in either direction because C opens the
path. Cause A and B are, then, dependent
conditioned on C in the common effect/collider
structure (A ⫫ B)

YES
Engine Cranks
but w/n Start

Figure 1.11 Common effect (collider)—Evidence.

The key takeaway from this section is that ANY piece of expert knowledge
(variables and relationships between variables) can be encoded using these three struc-
tures, or combinations thereof, by the knowledge engineering team. We now have all
the tools we need to develop the qualitative, or structural, component of BNs: the
DAG. The BN is incomplete, however, without a means to specify the relationships.
The quantitative component of the CBN is constructed using conditional prob-
ability tables (CPT) and conditional probability statements for each parent-child
relationship. These statements specify probabilities of each state of a child node
given the state of its parents (Figure 1.12).
 Bayesian Networks for Descriptive Analytics in Military ◾ 11

Figure 1.12 Conditional Probability Table (CPT): Quantitative component of a

Bayesian network.

We will discuss, in detail, some techniques for working with CPTs in the Elicit sec-
tion of DSEV. For now, it is easiest to consider this analogy when thinking about the
quantitative portion: experts use the DAG to model what they think the world of their
domain looks like; experts use the CPT to model how they move through that world.
The qualitative and quantitative components combine to make a BN. Note that all of the
structures described earlier are in use in the small network shown below (Figure 1.13).

Ambient Temperature

Battery Fuel System Mechanical

Cranking Speed Cylinder Pressure

Specific Gravity Test Engine Cranks Compression Test

but w/n Start

Figure 1.13 DAG and CPT: Qualitative and quantitative components of a

Bayesian network.
12 ◾ Military Applications of Data Analytics

The experts need to understand the basics and building blocks of BNs, need
to know what their knowledge looks like in a BN, and need to see how the BN
changes in response to different evidence. This is usually accomplished by building
a small sample BN—6–8 nodes, as above—for a familiar domain and demonstrat-
ing how it changes, what structures are used to model different types of relation-
ships, and what biases are in play. The BN in Figure 1.13 draws from our use case.
While it is very basic, it demonstrates how easily relationships can be articulated
and how it can frame further development. The translation of expert knowledge is
such that the completed BN looks like, behaves like, and learns like the experts that
contributed to its construction. With our experts now knowledgeable of the process
and the tools, we move to a few other pre-DSEV tasks: framing the problem, iden-
tifying a target node, and generating the list of nodes (variables) that will provide
the input for our DSEV process.

Pre-DSEV
This initialization stage starts with creating one to two very descriptive and specific
sentences that fit under the objective and the use cases and that will guide the facili-
tation and model building. For our use case, we could use, “The generator type is
a 60 kilowatt (kW) B model. The generator’s diesel engine will crank (turn over)
but not start in high ambient temperature.” Using that as a guide, the symptom
“diesel engine cranks but will not start” serves well as a target node. A note on tar-
get nodes: the goal is to find the right magnification. Too small, or narrow, and a
target (i.e., Generator fuel pump inoperative) may miss many possible causes of the
engine start problem; too large, or too wide, and a target (i.e., the Generator won’t
run) will cause the network to be too large and more difficult to elicit. A good target
finds the middle ground with sufficient specificity to identify all problems but with
a manageable network size. The experts are then asked to develop a list of nodes
that influence, or are influenced by, the target node. A number of good, CTA-based
techniques are useful for this process. Nominal group technique (NGT) is widely
used because its silent voting process eliminates many of the biases associated with
expert elicitations—most importantly, those that involve personalities influencing
the development of the node list (e.g., more charismatic experts may influence other
experts). In our use case, we would expect to see nodes representing causes of the
engine problem and nodes representing effects of the engine problem (Figure 1.14).
We are now ready to start the DSEV process proper.

Node List: Ambient Temperature, Engine Cranking Speed, Compression, Battery,

Fuel System, Specific Gravity, Mechanical, Cylinder Pressure

Figure 1.14 Node list.

 Bayesian Networks for Descriptive Analytics in Military ◾ 13

Define
The node list is voted on, prioritized, and then discussed at length, removing, by
consensus, duplicate or similar nodes. The imperative of the Define phase is to
ensure the state space in each node meets all criteria. The state space is a simple
description of what values the variable can take on in nature. The state space must
be exhaustive (all possible states enumerated) and mutually exclusive (the variable
can only be in one state). Most importantly, the state space must be intuitive to the
KE team and to the experts. Our CBN tools provide some help in that regard by
enabling us to add contextual information on values to establish thresholds, assign
upper and lower bounds, and make the states more specific. These can be as simple
as binary state space (true–false; yes–no), or multi-state where the value field pro-
vides the necessary threshold levels (Figure 1.15).

Figure 1.15 State space.

The “clairvoyance test” is unique to DSEV and provides the mechanism by

which we can test our work on developing state spaces. The premise of this test
is that if a person with perfect information provided a number or name value of
a variable, would this value make sense given only the variable name and state
description? In our “Ambient Temperature” example earlier, a clairvoyant value of
“28” would be easily translated into the “Cold/7.5–45 degree” state. The primary
goal of the KE team and experts is to remove vagueness and ambiguity from the
state space. With this thoroughly vetted list of variables in hand, we are ready to
start the Structure phase. In this next section, we will discuss how to categorize
node types and how to integrate the nodes on the list in a network.
14 ◾ Military Applications of Data Analytics

Structure
The first step in the developing the DSEV Structure phase is to work with a small
“chunk” of nodes and connect them. This requires knowledge of the three basic
BN node structures (discussed previously) and of the different types/categories of
nodes in a BN. BNs consist of five types of nodes: hypothesis nodes, context nodes,
indicator nodes, report nodes, and intermediate nodes. The state space of all types
of nodes must be, as discussed in the Define phase, mutually exclusive (the node
can only be in one state) and exhaustive (the state space must cover all that are
possible). We will briefly describe the attributes of each.
Hypothesis nodes describe possible root causes for the symptom. Hypothesis
nodes precede indicator nodes and report nodes in temporal order (they are parents
of indicator and report nodes). This means directed arcs go from the hypothesis node
to the indicator or report node. Hypothesis nodes succeed context nodes in temporal
order (they are children of context nodes). This means directed arcs come into the
hypothesis node from the context node. “Battery,” “Fuel System,” and “Mechanical”
are examples of hypotheses nodes for our problem statement (Figure 1.16).

Ambient Temperature

Battery Fuel System Mechanical

Cranking Speed Cylinder Pressure

Specific Gravity Test Engine Cranks Compression Test

but w/n Start

Figure 1.16 Hypothesis nodes.

Context, or pre-dispositional, nodes precede the hypothesis node in temporal

order (they are parents of hypothesis nodes). This means directed arcs go from
context nodes to hypotheses nodes. “Ambient Temperature” is an example of context
for the “Battery” hypotheses, as it could predispose the failure condition (e.g., cold
temperatures are correlated with low battery cranking performance) (Figure 1.17).
 Bayesian Networks for Descriptive Analytics in Military ◾ 15

Ambient Temperature

Battery Fuel System Mechanical

Cranking Speed Cylinder Pressure

Specific Gravity Test Engine Cranks Compression Test

but w/n Start

Figure 1.17 Context nodes.

Indicator nodes are children of hypothesis nodes. Directed arcs come into indi-
cator nodes from hypothesis nodes. Indicator nodes are, in general, symptoms of the
hypothesis node, but they may not always be easily observable. “Cranking Speed”
and “Cylinder Pressure” are examples of indicators of the hypothesis, as they are
symptoms but not readily observable (Figure 1.18).

Ambient Temperature

Battery Fuel System Mechanical

Cranking Speed Cylinder Pressure

Specific Gravity Test Engine Cranks Compression Test

but w/n Start

Figure 1.18 Indicator nodes.

Report nodes are children of indicator nodes. Directed arcs come into report
nodes from indicator nodes. Report nodes, in general, are observables of the indicator
node. “Specific Gravity Test” and “Compression Test” are examples of report nodes
for the hypotheses as they are both observables for the indicator node (Figure 1.19).
16 ◾ Military Applications of Data Analytics

Ambient Temperature

Battery Fuel System Mechanical

Cranking Speed Cylinder Pressure

Specific Gravity Test Engine Cranks but Compression Test

w/n Start

Figure 1.19 Report nodes.

Intermediate nodes—a part of the indirect connection structure described

previously—can be used with all types of nodes as the name suggests; for
example, context nodes can influence the hypotheses through an intermediate
node, hypotheses nodes can have intermediate effects between them and indicator
nodes, and so on. All of the same rules apply to intermediate nodes. “Cylinder
Pressure” would be an example of an intermediate node between “Mechanical”
and the hypothesis, as it describes one of the mechanisms whereby mechanical
issues affect engine start functions (Figure 1.20).

Mechanical

Cylinder Pressure

Engine Cranks
but w/n Start

Figure 1.20 Intermediate nodes.

 Bayesian Networks for Descriptive Analytics in Military ◾ 17

Another example from our “diesel engine cranks but will not start” problem
statement ties all these node types together: Cold ambient temperatures (pre-
disposition) slow down the flow of electrons through a battery’s electrolyte
(hypothesis) causing slow engine cranking speed (indicator, not directly observable/
quantifiable) that can be detected using a specific gravity test (report) (Conrady and
Jouffe, 2015) (Figure 1.21).

Ambient Battery[2] Cranking Specific

Temperature[2] Speed[1] Gravity Test[1]

Figure 1.21 All node types in action.

The key knowledge engineering task of the Structure phase involves deciding
what types of nodes we have in the list from the Define phase, where those nodes fit
in the model, and what structures to use to represent the knowledge. With all this
in hand, we can move to the description of the Elicit phase, where the structural
relationships described previously can be specified.

Elicit
This phase involves drawing out from the experts the probabilities for the structural
model you have built in the previous phase. This estimation process is burdened with a
number of biases that the team must be (a) aware of, and (b) have a mitigation plan for.
Before outlining the soft and hard skill requirements for the Elicit phase, Table 1.2 is
provided as a quick reference on common biases and mitigation techniques.
18 ◾ Military Applications of Data Analytics

Table 1.2
Anchoring: Initial impressions, whether quantitative or qualitative in nature,
can cause bias as the human brain tends to give more weight to the first
information received.

Mitigate by making the team aware, by avoiding using specific numbers

or specific contexts, and by providing a multi-perspective background.

Availability: The human mind tends to respond with information supported

by what information is readily available. This means that particularly “sticky”
recall (traumatic, emotional, or vivid) of information tends to bias estimates.

Mitigate by making the team aware and attempting to take emotion out of
the estimation process by reviewing pertinent statistics.

Confirmation: The human mind tends to seek out confirming evidence. Given
a preferred outcome, an individual will tend to seek out evidence to confirm
that outcome, and interpret evidence to confirm that outcome, instead of
giving equal weight to evidence that conflicts with the desired outcome.

Mitigate by making the team aware and actively seeking out disconfirming
evidence. Try to draw out scenarios that would result in a different
conclusion.

With this knowledge in hand, we move our focus to the key DSEV techniques
for eliciting probabilities for each node. One of the key challenges in this process
is in establishing the base rate, or prior probability, for each node. The base rate
should be an estimate of the state in nature with no other forces acting upon it. For
example, when eliciting the prior probability of a “cranks but will not run” fault,
consider some relevant parametric data, such as your own automobile. Does the
engine fail to start 1 out 100 starts? 1 out of 1000? 1 out of 10,000? Consider these
ratios knowing the tendency is to set the base rate too high. Consider both data and
context. For example, the base rate for smokers vs. non-smokers in a chest clinic is
significantly different that the base rate for smokers in a random selection of people
on the street. The prior probability should be entered into the nodes conditional
probability table (Hepler et al., 2015) (Figure 1.22).
 Bayesian Networks for Descriptive Analytics in Military ◾ 19

Figure 1.22 Conditional Probability Table (CPT).

For root nodes (nodes with no parents) these probabilities can be entered
directly into the CPT. For any node with parents, the node must be temporarily
disconnected (arcs coming into the node disconnected) to input prior probabilities.
This should be done for all nodes before reconnecting all the arcs. The resulting
reconnected network will show the marginal probabilities for each node, or the
probabilities with no evidence entered for any node, but the nodes will indicate that
the underlying CPT still needs to be filled in. As detailed earlier, the CPT is a series
of statements that show the probability distribution for each state of a child node
given each state of its parents (Conrady and Jouffe, 2015) (Figure 1.23).
20 ◾ Military Applications of Data Analytics

Figure 1.23 Conditional probability statements.

The knowledge engineering team and the experts work through each of the
statements. This is the heart of the process as these estimations are what makes
the expert knowledge show through in the BN and what makes the network
think and act like an expert. There are many direct and indirect methods for
drawing out these estimations. Experts like to work with words. Direct meth-
ods use questions designed to elicit a number response; indirect methods use
questions designed to elicit relationships that can be converted to numbers
(Table 1.3).
In some cases the team may find it easier to use a formula, such as a normal dis-
tribution, to translate expert statements and fill in the table. The key techniques here
are to avoid bias and to use tools to convert language into probabilities. The elicited
probabilities can be “played back” to the experts in real time to ensure they agree with
how the network responds to evidence. This is the essence of the final phase (Hepler
et al., 2015).
 Bayesian Networks for Descriptive Analytics in Military ◾ 21

Table 1.3 Direct and Indirect Methods for Estimating Probabilities

Verify
The sub-headings under this phase—consistency, coherence, what-if analysis, and
sensitivity analysis—all have to do with ensuring that the model you are working
on is operating as intended. Working with your expert team, use these techniques
to exercise the model in different ways to uncover any possible discrepancies, or
to just fine-tune the model. Consistency tests involve going into the CPT of each
child node, rearranging the order of the parents in all possible combinations, and
observing patterns after each reordering. Note below that rearranging the order
of parents makes it more obvious to the experts that the relationship between
“Cylinder Pressure” and “Engine Cranks but w/n Start” is independent of the
other parents. This will show, at a wide angle, how each parent node affects the
child node distribution, and this provides one way for the experts to review their
work and acknowledge the veracity of the parent-child relationships (Figure 1.24).
Similarly, coherence tests just provide the experts and the team with a different
perspective on the model. By examining the marginal distributions (no evidence
entered), we can analyze the relationships between variables and see if there is good
fit between the expert’s knowledge and the model. In most cases, this is a way to
22 ◾ Military Applications of Data Analytics

Consistency: Does the CPT make sense when

varying context or causal variables?

Figure 1.24 Consistency.

Coherence: Do the marginal distributions

make sense?

Figure 1.25 Coherence.

verify that the prior probability distributions and the conditional distributions in
the CPT are representative of the expert’s thought processes (Figure 1.25).
The “what-if” tests, as the name implies, involves “playing” with the model by
entering different sets of evidence and seeing how the model acts under different
scenarios. Always prepare experts for this test by asking what they expect to see. The
perspective on elicited results and computed results (2) provides yet another way for
experts to think about their view of the domain (Figure 1.26).
The key technical takeaway for this phase is mutual information. In building
the model, we are interested in how much entropy is reduced in a variable by
instantiating evidence in other nodes in the model. Mutual information sensitivity
nests well with the “first best question” philosophy that guides the practical tool for
soldiers and diagnostics in general. We want to eliminate as many possibilities as

What-if: Does the model work as expected

when entering evidence?

Figure 1.26 What if?

 Bayesian Networks for Descriptive Analytics in Military ◾ 23

Mutual information: How much does one

random variable tell us about another?

Figure 1.27 Mutual information.

possible in the shortest amount of time, and we want to be cognizant of opportunity

cost with respect to alternate diagnostic paths. BN software packages provide easy-
to-use tools for this type analysis (Hepler et al., 2015) (Figure 1.27).
We will now fit a DSEV elicitation around our generator use case. Diesel engine-
driven generators are the life-blood of any combat operation. Generators provide
clean, reliable, and quiet power to critical weapon systems and life-support systems.
The generator itself is a complex piece of machinery, with electrical, electro-mechanical,
and electronic components all integrated for diesel engine operation, electrical power
production, and control and safety circuits for both. The Army has limited data on
maintenance of these generators in the field. What exists is mainly transactional sup-
ply records that are only weakly associated with failure modes. What’s more, this
weak connection results in suboptimal supply policies and provides no actionable
data for soldiers at the point of need. What the Army does have is a very rich source
of experiential knowledge from the field support personnel that are tasked to support
these systems in the field. In some cases, this knowledge is made explicit in the form
of quick reference guides and information papers; in some cases, this knowledge is
tacit. In all cases, this knowledge is most important to tap into. Using the combina-
tion of DSEV and BNs, expert knowledge will be transformed into a data-generating
process that starts with accurate diagnosis and flows through the enterprise as accu-
rate data with which to make supply and repair decisions.

Critical Decision Method (CDM)

Our group of experts includes personnel with specific experience with supporting gen-
erators in deployed areas, along with personnel with general experience in deployed
areas and personnel with engineering knowledge on the generator sets. This section will
show how CTA methods are woven into the DSEV to produce BNs that “think” like
this group of experts. Getting to this point, however, requires a means for putting these
experts “in the moment” and then taking advantage of that opportunity with a struc-
tured process for creating an explicit, time-ordered account of the experts’ reasoning
about particularly critical problems. One of the most effective methods for doing this is
the critical decision method. CDM involves a number of elicitation “sweeps,” each with
a different purpose but overlapping in key aspects (Hoffman et al., 1998) (Figure 1.28).
The art of CDM is in selecting the right event. In our case, we want to put our
power generator experts back into their deployed experiences: ones where they were
the decision makers, where their decisions mattered, and where the situations were
unusual or significant enough to create vivid memories for the experts. In the power
24 ◾ Military Applications of Data Analytics

(1) Incident selection (2) Constructing a timeline (3) Deepening (4) What-if ’s

Figure 1.28 Critical decision method “sweeps.”

generation world, these cases will usually involve calls for assistance where critical
systems are down due to a lack of power and the usual supply and maintenance
resources are not available. For example, in Iraq in 2003, water purification assets
were placed strategically along convoy and supply lines to ensure coalition troops
and civilians had a dependable source of potable water (Figure 1.29).
These assets were often located on river side farms owned and run by local
Iraqi families. The only power sources sufficient to run the water purification
systems were the diesel generators mounted on the system trailers. Failure of these
generators and the loss of purified water production put soldiers and civilians at
immediate risk. Our specific case involves a generator deployed to a farm along on
the Tigris River in Iraq (Figure 1.30).

Figure 1.29 Tactical water purification equipment.

Figure 1.30 Tactical water production in Iraq.

 Bayesian Networks for Descriptive Analytics in Military ◾ 25

During operation, an electrical short had caused catastrophic failure of the

generators Input/Output (IO) module—the “brain” of the generator control
system. This caused the generator to shut down during operation with no specific
shutdown indicators. A field support person, one of our experts, had only a matter
of hours with which to travel to the affected site via convoy, diagnose the problem,
and effect short- and long-term repair plans. This scenario meets all the criteria for
incident selection.
The second sweep in CDM involves an iterative process of developing a timeline
of events that covered the incident. In our case, we want to know, from the time the
expert was notified of the problem, what were his or her reactions, preparations, and
thought processes, and how did those translate into an ordered approach to address-
ing the problem—remembering that the combat environment introduces uncertainty
into the basic process of getting to the site. From this process, we gather the following:

1. The expert convoyed to the site and was delayed several times along the way
due to enemy activity.
2. Without any communication with the site during the convoy, the expert
started running possible cause and effect relationships and started developing
a set of questions.
3. Upon arrival at the site, the expert asked key questions of the crew to create a
sub-timeline of the failure itself, e.g., did it occur during full load operation,
no-load operation, starting, etc.

From a DSEV perspective, the knowledge engineering team uses this information
to confirm a target node and to compile a list of these potential “first-best”
questions that will become context and indicator variables. As the timeline gets
richer in detail, we gather that the expert had a good idea of what he or she wanted
to look at first when gaining access to the generator, and this was an “audible click”
test, not available in any technical manual or doctrine, that the expert himself
had developed from his own experience and from discussions with engineers.
The test was designed as a quick way to determine if power was available at the
fuel injection pump actuator. A positive test result (audible click) would verify all
control circuits were functioning and would place the problem in the mechanical
or fuel supply subsystems. A negative test result would mean that a failure in
the control circuit was nearly certain. The test effectively isolates the problem to
mechanical or electrical and reduces a significant amount of entropy with very
little time or effort required (Figure 1.31).
Per our DSEV process guidelines and based on our expert team make-up, the
majority of these experts will have similar experiences troubleshooting critical failures
in deployed areas. The discussion of the audible click test will stimulate discussion
on if, and how, the test should be a part of the troubleshooting network. It will also
stimulate discussion on what the other experts do in the same situation, possibly
revealing other such tests. These are exactly the types of discussions DSEV can
Other documents randomly have
different content
began his sermons in the fields and looked down from a green slope
on several thousand colliers grimy from the coalpits near Bristol to
see, as he preached, tears "making white channels down their
blackened cheeks." Later again, Hannah More drew sympathy to the
poverty and crime of the agricultural classes.
The Influence of Walnut on Cabinet-making.—If oak was the
wood which the country joiner loved best, he was not without some
sympathetic leaning towards the effects which could be produced in
the softer walnut. Such styles accordingly began slowly to have a
marked influence upon the farmhouse furniture in early-Georgian
days. It was not easy to produce curved lines in the refractory oak,
tough and brittle, but the village craftsman essayed his best to
please his patrons whose taste had been caught by the newer
fashions observed in the squire's parlour when paying rare visits.
In the two examples illustrated of farmhouse cupboard and bureau
bookcase (p. 163) it will be seen that here is the country maker
definitely trying his skill in his native wood to emulate the finer
walnut examples of town cabinet-makers. This is even more
noticeable in regard to some of the tables actually found in
farmhouses belonging to as early as the first quarter of the
eighteenth century. The two specimens illustrated (p. 165) exemplify
this tendency to imitate the designs of trained workers. The country
touch always betrays itself in the cabriole leg, whether in chair or in
table. The upper table has less naïveté than most examples found.
There is a balance in its construction rarely found in provincial work.
The legs, always the stumbling-block to the less experienced
artificer, are here of exceptionally fine proportions, terminating in
club feet. The lower table shows a less capable treatment of the
cabriole leg. The hoof foot and the carved knee have obviously been
copied from a fine Queen Anne model. In the underframing of both
tables there is an experiment in ornament and form rarely attempted
except in the highest flights of the country maker, and as such these
two fine examples must be regarded.
 OAK TABLE.
Showing clumsy corners and indicating the naïveté of the country
cabinet-maker.
 OAK TABLE.
Showing transition from cabriole leg to straight leg of 1760.

The Early Georgian Types.—Treating of the early-Hanoverian

period from the death of Queen Anne in 1714, and including the
reigns of George I. from 1714 to 1727 and George II. from 1727 to
1760, furniture of all types begins to assume a complexity of
construction. At the final outburst the fine masterpieces of creation
of the great schools of design during the last half of the eighteenth
century, embodied the life-work of Chippendale, the brothers Adam,
Hepplewhite, Sheraton, and many others. This period from 1750 to
1800 was the golden age of design in England. It has had a far-
reaching effect, and still casts its glory upon the present-day schools
of designers, whose adaptations and lines of progress are based
upon the finest flower of the eighteenth-century styles.
The massive walnut chairs with deep underframing and broad hoop
backs departed from the solid splats of the Anne style and
endeavoured to become less squat by the employment of banded
ribbon-work, coarse, heavy, and ponderous in style. Settees, arm-
chairs and single chairs in this style came as the final efforts of the
walnut school. The graceful ribbon designs interlacing each other in
knots, and the flowing carving in mahogany of Chippendale, put a
period to all dullness and heavy design. With the new style and the
new wood a splendid field was opened to cabinet-makers, and the
quick appreciation of these opportunities signalised their work as of
permanent artistic value.
Among more important pieces, though still falling under the category
of farmhouse styles, may be mentioned the Queen Anne glass or
china cupboard, and the Georgian corner cupboard, illustrated p.
171.
The former has heavy bars, which mark the early type prior to
tracery, and it has spun-glass doors. Porcelain factories at Bow,
Worcester, and Derby brought such cupboards into more general use
after the middle of the century. Staffordshire earthenware tea and
coffee services were found in great numbers in farmhouses and
cottages. After the days of delft and stoneware came the prized
china services of the housewife. Pewter was largely used, but the
number of ale-jugs of Toby form, or cider-mugs with rural subjects
to suit the tastes of the users, indicate that more modern ideas and
taste, once exclusive to the world of fashion, had penetrated the
country districts.
The Georgian corner cupboard shows the broken architraves and
cushion top. The hinges should be noticed as being original.
Chippendale and his Contemporaries.—At first using the
cabriole leg with ball-and-claw foot, not quite as he found it, but
reduced to slightly more slender proportions to be in symmetry with
his less massive backs to chairs, Chippendale came to the straight
line. He employed it in the legs of tables and in the seats of chairs,
in the bracket supports, and in the top rail of his chairs. Chippendale
in his day, made the first straight top rail to the chair. It is interesting
to note the phases of changing design in country-made furniture
prior to his time, and the sudden mastery of form which became the
common inheritance of all after his and other contemporary design-
books were promulgated broadcast.

QUEEN ANNE TEA TABLE. C. 1710.

With scalloped edge for cups. Height, 2 ft. 4 ins.; depth, 1 ft. 9 ins.;
length, 2 ft. 8 ins.
 OAK REVOLVING BOOK-STAND. C. 1720.
Rare form. Diameter of top, 2 ft.; height, 2 ft. 8 ins.
(In the collection of Miss Holland.)
 COUNTRY CHIPPENDALE TABLE.
Leg with exaggerated knee, claw, and ball foot. Accuracy in straight
joinery. Failure in curved work.
Top, 2 ft. 7 ins. × 1 ft. 3 ins.; height, 2 ft. 4 ins.
 SQUARE MAHOGANY FLAP TABLE. C. 1730.
Height, 2 ft. 4 ins.; length, 3 ft. 10-1/2 ins.; width, 2 ft. 1 in. Round cross
stretcher. Rare form.
 TRIPOD TABLE. C. 1760.
Chippendale style, probably unique. Elaborate rococo work.
(In the collection of Harold Bendixon, Esq.)

In the table the cabriole leg showed early signs of passing away. The
two examples illustrated (p. 173) clearly indicate this. The upper
one, of the time of Queen Anne, shows the cabriole leg in fine
proportion under due subjection, and is a delicate example of fine
cabinet-work. The lower one sees the leg losing its cabriole curve,
but still rounded and still possessing the club foot.
Even more interesting are the two tables illustrated (p. 177). The
country maker was slow to adopt the cabriole leg when it was
fashionable, but when it became unfashionable he was equally loth
to depart from his accustomed style. These clearly point to the
transition between the cabriole leg and the straight leg of
Chippendale, and are about 1760 in date.
The forms of design of tables of eighteenth-century date are
extremely varied in character, denoting the rapidly changing habits
of the people. The Queen Anne tea-table, with scalloped edges for
cups, marks the note of preciosity creeping into country life. A
revolving bookstand in table form, of about 1720 in date, is another
rare piece. The adjacent table (p. 181) is country Chippendale. The
exaggerated knee and the feeble ball-and-claw foot mark the failure
of the provincial hand at curved work, accurate though he might be
in straight joinery. The "Cupid's bow" underframing is interesting in
combination with the rest of the design.
The tripod table offered difficulties of construction and is not often
found. The example illustrated is probably unique in form. In date it
is about 1760, and is remarkable for the attempt at elaborate rococo
work. Sometimes, though not often, mahogany was used in
farmhouse examples. The table illustrated (p. 183) is an instance of
the use of this wood instead of oak. It is about 1730 in date, and
exhibits an unusual form in the round cross stretcher, a touch of
originality by the maker. It is, as will be seen, a square-topped table
with flaps.
Elaboration of a high order was happily not often attempted by the
country workman, or the results with his limited experience would
have been disastrous. Instead of a fine series of really good, solid,
and well-constructed furniture made for practical use we should
have had a wilderness of failures at attempting the impossible. A
copy of a fine Chippendale side-table illustrated (p. 187) is a case in
point. There is the usual want of balance in the poise of the leg, but
the carving is of exceptional character. The table beneath, with its
long and tapering legs, has all the characteristics of the Adam style.
The beaded decoration on the legs, the classic fluting and the carved
rosette claim distant relationship with the classic inventions of
Robert Adam. The wood is pinewood, and as an example it is of
singular interest.
The rapid survey of eighteenth-century influences bearing on the
class of furniture of which this volume treats will perhaps induce the
collector to scrutinise more carefully all pieces coming under his
notice, with a view to arriving at their salient features in connection
with the native design of more or less untutored craftsmen.

ELABORATE TABLE.
Country attempt to imitate fine Chippendale side table. Note the want of
balance in leg.
PINEWOOD COUNTRY-MADE ADAM TABLE.
Note the unusually long leg.
CHAPTER VII
THE EVOLUTION
OF THE CHAIR

OAK ARM-CHAIR. DATE C. 1675.

With elaborate scroll back.

OAK ARM-CHAIR. DATE 1650.

With scratched lozenge.
 CHESTNUT ARM-CHAIR. DATE 1690.

OAK ARM-CHAIR. DATE 1690.

(By the courtesy of Messrs. Phillips, Hitchin.)

CHAPTER VII
THE EVOLUTION OF THE CHAIR

Early days—The typical Jacobean oak chair—The evolution of

the stretcher—The chair-back and its development—
Transition between Jacobean and William and Mary forms
—Farmhouse styles contemporary with the cane-back
chair—The Queen Anne splat—Country Chippendale,
Hepplewhite, and Sheraton—The grandfather chair—
Ladder-back types—The spindle-back chair—Corner
chairs.
In order to deal exhaustively with the evolution of the chair from its
earliest forms to the latest developments in sumptuous upholstery, it
would be necessary to make an extended survey of furniture, dating
back to early classic days. To enumerate the manifold varieties
belonging to various countries and to trace the gradual progress in
form, which kept pace with the advance in civilisation, would be of
sufficient interest to occupy a whole volume. Man, as a sitting or
lounging animal, has grown to require more elaborate forms of chair,
or settee, or sofa, and the modern tendency has been towards
comfort and luxury.
In regard to English furniture the intense contrast between the days
of Elizabeth and those of Victoria is at once noticeable. According to
Lord Macaulay in his comparison between the manners of his day
and those of the past, the furniture of a middle-class dwelling-house
of the nineteenth century was equal to that of a rich merchant in the
time of Elizabeth. In general this may be true, though not as regards
the spacious structure and the massive grandeur of the Tudor house.
In many details the differences are most noteworthy. The wide gulf
dividing the modern world from the days of the Armada may be
realised by reflecting on such an astounding fact that Queen
Elizabeth possessed at one time the only pair of silk stockings in her
realm, which were presented to her by Mistress Montague, "which
pleased her so well that she would never wear any cloth hose
afterwards."
The sturdy character of the yeomen of the days of the Tudors is
exhibited in their furniture. The illustrations of this chapter in regard
to the chair and its structural development indicate the slowly
acquired tastes, running some decades behind the fashionable
furniture, strong with foreign influences, which had come into more
or less general use. "England no longer sent her fleeces to be woven
in Flanders and to be dyed in Florence. The spinning of yarn, the
weaving, fulling, and dyeing of cloth, was spreading rapidly from the
towns to the country-side. The worsted trade, of which Norwich was
the centre, extended over the whole of the Eastern Counties.
Farmers' wives everywhere began to spin their wool from their own
sheep's backs into a coarse homespun."
The rough and wattled farmhouses were being replaced by dwellings
of brick and stone. The disuse of salt fish and the greater
consumption of meat marked the improvement which was taking
place among the countryfolk. The wooden trenchers in the
farmhouses were supplanted by pewter, and there were yeomen
who could boast of their silver. Carpets in richer dwelling-houses
superseded the wretched flooring of rushes. Even pillows, now in
common usage, were articles of luxury in the sixteenth century. The
farmer and the trader deemed them as only fit "for women in child-
bed." The chimney-corner came into usage in Elizabethan days with
the general use of chimneys. The mediæval fortress had given place
to the grandeur of the Elizabethan hall in the houses of the wealthy
merchants. The rise of the middle classes brought with it in its wake
the corresponding advance of the yeomen and their dependents.
Visions of the New World "threw a haze of prodigality and profusion
over the imagination of the meanest seaman."
Early Days.—Of farmhouse types that can authoritatively be
attributed to Tudor days there are few, but the succeeding age of
the Stuarts is rich with examples of undoubted authenticity. Many of
them are dated, and they all bear a strong family resemblance to
each other, owing to the narrow range of motifs in the carved
panels. There is a fixed insularity in these early examples, and the
same traditional patterns in scrollwork or in conventional lozenge
design retained their hold for many generations. The oak arm-chair
of a farmhouse kitchen made in the days of Charles I. was still
followed in close detail in the days of George III., as dated examples
testify, and it would puzzle an expert, without the date to guide him,
to say whether the piece was eighteenth or seventeenth century
work. It may be added that as a general rule there is a marked
leaning towards generosity in imparting age to old furniture. It is
now very generally recognised that, like wine, it gains prestige with
length of years. It therefore grows in antiquity according to the
fancy of the owner or the imagination of the collector.
Among the early forms of chairs falling under the category of
farmhouse furniture may be noticed examples of rough and massive
build, eminently fit to serve the purpose for which they were
designed. Ornament is reduced to a minimum, and they stand as
rude monuments to the cabinet-maker's craft in fashioning them and
following tradition to suit his client's tastes.
In regard to the sixteenth century there cannot be said to be any
type falling under the heading of cottage or farmhouse chairs. We
have already illustrated (p. 35) an early form of Elizabethan days,
but such examples are rare. Practically cottagers had only stools in
common use. It was not until about 1650 that a simplified form of
the well-known variety of the chairs of the Jacobean oak period
came into general use.
The Typical Jacobean Oak Chair.—The seventeenth century
offers a wide field of selection, and many examples exist which
undoubtedly were in use in farmhouses at that period. The arm-chair
illustrated p. 191, with the initials "W.I A.", is evidently made for the
farmhouse. It is noticeable for its complete absence of ornamental
carving except a thinly scratched lozenge. In date this is from 1650
to 1700, and if made for a wealthier person at that date it would be
richly carved. The adjacent chair shows the next advance in type. It
is a superior farmhouse chair of the period. It has a carved top with
scroll cresting. The holes in the seat, it should be observed,
originally held ropes, upon which a cushion was supported. The
wooden seat is an addition made in the eighteenth century.
 YORKSHIRE CHAIR. DATE 1660.
Late example, with ball turning in stretcher.
 CROMWELLIAN CHAIRS. DATE 1660.
With indication of transition to Charles II. period.
(By the courtesy of Messrs. Phillips, Hitchin.)

The two other chairs illustrated on the same page are later
examples, in date about 1690. One of these is fashioned of chestnut.
The form of these backs is related to the contemporary high-back
cane chairs of the time of Charles II. and James II. But these
fashions influenced the proportions only of farmhouse chairs. In
arriving at the date of such specimens as these the bevelled panel is
an important factor in determining the late period.
Cushions had no place in the effects of the farmhouse in early days,
although ropes were sometimes used to support cushions, as we
have shown. But as a general rule the wooden seats show tangible
signs of rough usage of centuries, and the stretcher has its worn
surface marked by generations of owners who found it protective
against the cold flagged or rush-strewn floor and the draughts in
days prior to carpets and rugs.
The Evolution of the Stretcher.—In making a study of the
evolution of the chair the stretcher is an important factor. For
obvious reasons, as explained above, no early chairs were made
without the stretcher across the front, a good sound serviceable
piece of British oak to stand rough wear and tear. Gradually, keeping
time with the march of comfort, the front stretcher begins to leave
its old position near the floor, and in later examples it is half-way up
the front legs. It still had a use, and a very important one: it added
considerable strength and solidity to the chair, and is nearly always
found in chairs intended for use. In the series illustrated herein there
are only few examples without the front stretcher. Later it took
another form, as the illustrated specimens in this chapter show: it
united the two side stretchers, and crossed the chair underneath in
the centre at right angles to the side stretchers. Its purpose in
adding stability to this class of furniture was evidently never lost
sight of.
At first strictly utilitarian, the stretcher was a solid foot-rest; later,
when partly utilitarian in adding to the strength, it became suitable
for ornamentation, Although in the class of furniture here under
review such ornament never took an elaborate form, there are
examples slightly differing in character from chairs intended for the
use of the wealthier classes, and these are evidently a local effort to
keep in touch with prevailing taste.
Finely turned stretchers, such as are found in gate tables, are a
feature of a certain class of local chairs, such as those illustrated on
p. 197. This kind of chair without arms is rather more decorated and
conforms more to the styles of furniture made for higher spheres
than the farmhouse. The upper chair with its light open back and
ornate decoration is a Yorkshire type, and the ball turning in the
stretcher shows the transition period to Charles II. The other two
are Cromwellian chairs, but showing indications of the next period.
In date they are all three about 1660.

OAK SETTLE.
With back panel under seat made from older Oak Chest. Date 1675.
 OAK ARM CHAIR. DATE 1675.
With Bevelled Panels.

OAK ARM CHAIR. DATE 1777.

With initials A.S. C.B.

The Chair-back and its Development.—Another point in

connection with the ordered progress of the chair-maker is the
gradual development of the back of the chair. At first it was straight
upright, and no attempt was made to impart an angle to rest the
back of the sitter. Types such as the arm-chair with square panel (p.
191) and the upright settle with the five panels illustrated on p. 201
indicate this feature of discomfort. The next stage is a slight
inclination in the back, still possessing a flat panel. This angle, while
not conforming to modern notions of ease, was an attempt to offer
greater comfort than before. This style, in a hundred forms, with the
minimum of inclination in the back, continued for a very considerable
period. It is found in the nearly straight-backed chairs of Derbyshire
and Yorkshire origin, with the turned stretchers, and it actually in
later days became almost upright in the series of chairs following the
later Stuart types with cane back and cane seat, noticeable for their
tall narrow backs with a resemblance to the prie-dieu chair of
continental usage.
The settle illustrated is a plainer variety of the settle made for use by
fashionable folk with delicately panelled back. Very often, in cottage
furniture, chests and other pieces are broken up to make into
smaller furniture or to be incorporated into furniture of a later
design. Often it is found that the underframing of an old gate table
made in the seventeenth or eighteenth century is from an earlier
chest. In the present instance it will be seen that the back panels of
the settle have been made from an older chest, which bears the
inscribed initials, still visible, "I.E." In date this settle is about 1675,
and is contemporary with the square-backed chair illustrated on the
same page. Here the panel in back projects, that is, it is slightly
bevelled forward. The bevelling of the panel is always a sign that a
chair is later in date than the year 1670.
Illustrated on the same page is a remarkable chair having the initials
"A.S.C.B." and the date 1777 carved on it. It is a striking instance of
the adherence to old time-honoured form by the local cabinet-maker,
with touches that, even although the date were not present, would
tell their own story. This dull wood proclaims a message in accents
no less sure than the sturdy yeoman's to Lady Clara Vere de Vere,
and as a chair in date anno Domini 1777 may afford to "smile at the
claims of long descent" of more pretentious and fashionable
furniture. It is like a rich vein of dialect running in some old country
song ripe with phrase of Saxon days. It seems incredible that this
survival of early-Jacobean days should have been put together by a
village craftsman true to convention and exact in seat and arms and
stretcher. But it was not done unthinkingly. Here is a chair,
astounding to note, made when Sheraton was creating his new
styles to supplant Chippendale, and when Hepplewhite stood
between the two masters as a via media. And the back of this village
chair has two distinct features translated from Hepplewhite's school
—the wheatear crest and the panel with its broken corner!

OAK CHAIRS. DATE ABOUT 1680.

Showing the inclination of the craftsmen to assimilate designs then
being fashioned in walnut.
(By the courtesy of Messrs. Phillips, Hitchin.)

Transition between Jacobean and William and Mary Forms.

—The rapid growth of the finer specimens of furniture made in
walnut brought a new note into the farmhouse variety. The elegance
and grace of the newer styles were at once evident. In the same
manner as the grandiose splendour of Elizabethan woodcarving was
succeeded by a less massive style in oak, degenerating into a rude
simplicity in farmhouse examples, so in turn Jacobean lost favour.
Walnut lent itself to more intricate turning, and lightness and greater
delicacy claimed the popular favour of fashionable folk. The cane
seat and the cane back at once indicate this new taste. The use of
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookbell.com