Multi Stage and Multi Time Scale

Feedback Control of Linear Systems

with Applications to Fuel Cells Verica
Radisavljevi■-Gaji■
Visit to download the full and correct content document:
https://textbookfull.com/product/multi-stage-and-multi-time-scale-feedback-control-of-l
inear-systems-with-applications-to-fuel-cells-verica-radisavljevic-gajic/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Fixed Time Cooperative Control of Multi Agent Systems

Zongyu Zuo

https://textbookfull.com/product/fixed-time-cooperative-control-
of-multi-agent-systems-zongyu-zuo/

Classical Feedback Control with Nonlinear Multi-Loop

Systems: With MATLAB® and Simulink® 3rd Edition Boris
J. Lurie

https://textbookfull.com/product/classical-feedback-control-with-
nonlinear-multi-loop-systems-with-matlab-and-simulink-3rd-
edition-boris-j-lurie/

Networked control systems with intermittent feedback

1st Edition Hirche

https://textbookfull.com/product/networked-control-systems-with-
intermittent-feedback-1st-edition-hirche/

Multi-scale Extracellular Matrix Mechanics and

Mechanobiology Yanhang Zhang

https://textbookfull.com/product/multi-scale-extracellular-
matrix-mechanics-and-mechanobiology-yanhang-zhang/
Multi-View Geometry Based Visual Perception and Control
of Robotic Systems First Edition Chen

https://textbookfull.com/product/multi-view-geometry-based-
visual-perception-and-control-of-robotic-systems-first-edition-
chen/

Agents and Multi-Agent Systems: Technologies and

Applications 2018 Gordan Jezic

https://textbookfull.com/product/agents-and-multi-agent-systems-
technologies-and-applications-2018-gordan-jezic/

Multi time Wave Functions An Introduction Matthias

Lienert

https://textbookfull.com/product/multi-time-wave-functions-an-
introduction-matthias-lienert/

Design of Linear Multivariable Feedback Control Systems

The Wiener Hopf Approach using Transforms and Spectral
Factorization Joseph J. Bongiorno Jr.

https://textbookfull.com/product/design-of-linear-multivariable-
feedback-control-systems-the-wiener-hopf-approach-using-
transforms-and-spectral-factorization-joseph-j-bongiorno-jr/

Fundamentals of Chemical Reactor Engineering A Multi

Scale Approach 1st Edition Timur Do■u

https://textbookfull.com/product/fundamentals-of-chemical-
reactor-engineering-a-multi-scale-approach-1st-edition-timur-
dogu/
Mechanical Engineering Series

Verica Radisavljević-Gajić
Miloš Milanović
Patrick Rose

Multi-Stage and Multi-

Time Scale Feedback
Control of Linear Systems
with Applications to Fuel
Cells
Mechanical Engineering Series

Series Editor
Francis A. Kulacki
Department of Mechanical Engineering, University of Minnesota,
Minneapolis, MN, USA
The Mechanical Engineering Series presents advanced level treatment of topics on
the cutting edge of mechanical engineering. Designed for use by students,
researchers and practicing engineers, the series presents modern developments in
mechanical engineering and its innovative applications in applied mechanics,
bio-engineering, dynamic systems and control, energy, energy conversion and
energy systems, fluid mechanics and fluid machinery, heat and mass transfer,
manufacturing science and technology, mechanical design, mechanics of materials,
micro- and nano-science technology, thermal physics, tribology, and vibration and
acoustics. The series features graduate-level texts, professional books, and research
mono-graphs in key engineering science concentrations.

More information about this series at http://www.springer.com/series/1161

Verica Radisavljević-Gajić • Miloš Milanović •
Patrick Rose

Multi-Stage and Multi-Time

Scale Feedback Control of
Linear Systems with
Applications to Fuel Cells
Verica Radisavljević-Gajić Miloš Milanović
Department of Mechanical Engineering Department of Mechanical Engineering
Villanova University Villanova University
Villanova, PA, USA Villanova, PA, USA

Patrick Rose
Department of Mechanical Engineering
Villanova University
Villanova, PA, USA

ISSN 0941-5122 ISSN 2192-063X (electronic)

Mechanical Engineering Series
ISBN 978-3-030-10388-0 ISBN 978-3-030-10389-7 (eBook)
https://doi.org/10.1007/978-3-030-10389-7

Library of Congress Control Number: 2018966525

© Springer Nature Switzerland AG 2019

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

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
“I dedicate this book to my parents
Ljiljana and Miroljub Radisavljević
for their unconditional love and support”
Dr. Verica Radisavljević-Gajić

“To my dear family and Saša,

for endless support and lifelong inspiration”
Miloš Milanović

“To my friends and family

for their endless support and confidence
in me”
Patrick Rose
Preface

This monograph is intended for researchers and practitioners in control systems

interested in complex, time-invariant, linear dynamic systems composed of several
subsystems and/or large-scale linear, time-invariant, dynamic systems operating in
several time scales. It can be used by engineering graduate students majoring in
control systems and automation, practicing control engineers, as well as all engi-
neering, applied mathematics, economics, and computer science faculty interested in
control of dynamic systems. Due to numerous applications of the derived feedback
controller design algorithms presented in this book to fuel cells (clean electric energy
sources), particularly to proton-exchange membrane (PEM) fuel cells, the book is
also of interest for the fuel cell community researchers, graduate students, and
practitioners. One of the most important features of the presented design methodol-
ogy is that different types of linear feedback controllers can be designed for different
subsystems and/or in different time scales of a complex linear dynamic system.
This research monograph presents completely the two-stage feedback controller
design algorithms in both continuous- and discrete-time domains including all
design formulas and algebraic equations for general linear time-invariant dynamic
systems. The results presented are specialized and simplified for the two-time scale
linear time-invariant dynamic systems (singularly perturbed systems). The
corresponding presentation is completely done also for the three-stage continuous-
time feedback controller designs for general linear time-invariant dynamic systems.
Extensions of three-stage feedback controllers to linear dynamic time-invariant
continuous- and discrete-time systems operating in the three-time scales have been
done only in the continuous-time. The discrete-time three-time scale systems remain
an open research area for the three-stage design of linear feedback controllers,
mostly due to unsettled issues about the formulations of the three-time scale linear
dynamic systems. The four-stage linear feedback controller design is presented only
for general class of linear, time-invariant, continuous-time dynamic systems. How to
extend those results or, more precisely, how to solve the obtained design algebraic
equations for the four-time scale, time-invariant, continuous-time, linear dynamic

vii
viii Preface

systems is discussed. Hence, the four-stage four-time scale feedback controller

design is left as an open future challenging research area.
The designs of linear-quadratic optimal controllers and eigenvalue assignment
(“pole placement”) feedback controllers are fully covered in this research mono-
graph. The book provides opportunities for future research of other types of multi-
stage linear feedback controllers for general and multi-time scale linear dynamic
systems, including observer-based and Kalman filter-based deterministic and sto-
chastic controllers. Due to the fact that different types of linear feedback controllers
can be designed (partial controllers) for different subsystems of a large-scale linear
dynamic system and/or system operating in different time scales, producing the dual
results for observers and the Kalman filter, it will be possible, for example, to design
partially optimal Kalman filter and new classes of reduced-order observers. Exten-
sions to other types of existing linear feedback controllers and filters are possible,
including the design of new hybrid linear dynamic feedback controllers and filters.

Villanova, PA, USA Verica Radisavljević-Gajić

Villanova, PA, USA Miloš Milanović
Villanova, PA, USA Patrick Rose
Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 General Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Applications to Other Classes of Dynamic Systems . . . . . . . . . . . 6
1.3 Improved System Robustness, Reliability, and Security . . . . . . . . 7
1.4 Book Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 Continuous-Time Two-Stage Feedback Controller Design . . . . . . . . 11
2.1 Two-Stage Design of Linear Feedback Controllers . . . . . . . . . . . 11
2.2 Two-Stage Feedback Design for Systems with Slow and Fast
Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Two-Stage Control of a Hydrogen Gas Reformer Slow-Fast
Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.1 Hydrogen Gas Reformer Operation and Modeling . . . . . . 21
2.3.2 Eigenvalue Assignment for the Hydrogen Gas
Reformer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.3 Optimal Slow and Eigenvalue Assigned Fast
Subsystems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Two-Stage Feedback Control of a Two-Time-Scale PEM
Fuel Cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4.1 PEM Fuel Cell Dynamics Mathematical Model . . . . . . . . 30
2.4.2 Two-Time-Scale Structure of the PEM Fuel Cell . . . . . . . 33
2.4.3 PEM Fuel Cell Slow-Fast Two-Stage Controller
Design Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.5 PEM Fuel Cell Observer Design . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3 Discrete-Time Two-Stage Feedback Controller Design . . . . . . . . . . 47
3.1 Discrete-Time Two-Stage Feedback Controller Design . . . . . . . . 47
3.2 Slow-Fast Design for Systems Defined in the Slow Time Scale . . 51
3.2.1 Example: A Power System . . . . . . . . . . . . . . . . . . . . . . . 56

ix
x Contents

3.3 Slow-Fast Design for Systems Defined in the Fast Time Scale . . . 59
3.3.1 Example: A Steam Power System . . . . . . . . . . . . . . . . . . 63
3.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4 Three-Stage Continuous-Time Feedback Controller Design . . . . . . 71
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2 Three-Stage Design of Continuous-Time Feedback
Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.3 Three-Stage Three-Time Scale Linear Control Systems . . . . . . . . 81
4.4 Application to a Proton Exchange Membrane Fuel Cell . . . . . . . . 90
4.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5 Three-Stage Discrete-Time Feedback Controller Design . . . . . . . . . 97
5.1 Three-Stage Discrete-Time Linear Feedback Controllers . . . . . . . 98
5.2 Three-Stage Three-Time Scale Discrete Linear Control
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.3 Future Research Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6 Four-Stage Continuous-Time Feedback Controller Design . . . . . . . 109
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.2 Four-Stage Design of Continuous-Time Feedback Controllers . . . 111
6.3 Four-Stage Four-Time Scale Linear Control Systems . . . . . . . . . . 123
6.4 Future Research Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
7 Modeling and System Analysis of PEM Fuel Cells . . . . . . . . . . . . . . 129
7.1 Third-Order Linear Model of a PEM Fuel Cell . . . . . . . . . . . . . . 132
7.1.1 Controllability of the Linear PEM Fuel Cell Model . . . . . 133
7.1.2 System Analysis and Constraints of the PEMFC Model . . 135
7.2 Third-Order Bilinear PEM Fuel Cell Model . . . . . . . . . . . . . . . . 138
7.2.1 Steady-State PEM Fuel Cell Equilibrium Points . . . . . . . . 140
7.2.2 Fuel Cell System Stability Analysis . . . . . . . . . . . . . . . . . 142
7.2.3 PEM Fuel Cell Controllability and Observability
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.2.4 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
7.3 Greenlight Innovation G60 Station with TP50 PEMFC . . . . . . . . 150
7.3.1 TP50 PEMFC Modeling . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.3.2 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
7.4 A Fifth-Order Nonlinear PEMFC Model . . . . . . . . . . . . . . . . . . . 161
7.5 Eight-Order Mathematical Model of a PEMFC Used
in Electric Cars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
7.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8 Control of a Hydrogen Gas Processing System . . . . . . . . . . . . . . . . 167
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
8.2 Full- and Reduced-Order Observer and Optimal Controllers . . . . 172
8.2.1 Full-Order Observer Design . . . . . . . . . . . . . . . . . . . . . . 173
8.2.2 Reduced-Order Observer Design . . . . . . . . . . . . . . . . . . . 174
Contents xi

8.2.3
Optimal Linear-Quadratic Integral Feedback
Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
8.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
8.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Appendix 8.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
9 Extensions to Multi-stages and Multi-time Scales . . . . . . . . . . . . . . 187
9.1 Extensions to Multi-stage Multi-time Scale Linear Systems . . . . . 188
9.2 Multi-stage Feedback Design for Multi-time Scale Systems . . . . . 192
9.3 Multi-stage Feedback Design for Other Classes of Systems . . . . . 193
Appendix 9.1: Summary of the Three-Stage Feedback Controller
Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
Appendix 9.2: Summary of the Four-Stage Continuous-Time
Feedback Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Chapter 1
Introduction

The design of linear continuous- and discrete-time state feedback controllers is well
documented in control engineering literature; see, for example, Franklin et al.
(1990), Ogata (1995), Sinha (2007), and Chen (2012). The authors of this mono-
graph have recently developed new algorithms for the design of two- and three-stage
feedback controllers for both linear discrete- and continuous-time dynamic systems
(Radisavljevic-Gajic and Rose 2014; Radisavljevic-Gajic 2015a, b; Radisavljevic-
Gajic et al. 2015, 2017) that have been efficiently applied to two- and three-time-
scale models of fuel cells (Radisavljevic-Gajic and Rose 2014; Radisavljevic-Gajic
et al. 2015, 2017; Radisavljevic-Gajic and Milanovic 2016; Milanovic et al. 2017;
Milanovic and Radisavljevic-Gajic 2018). In general, the results of these new
multistage and multi-time-scale feedback controller design algorithms are applicable
under mild conditions to almost all linear continuous- and discrete-time time-invari-
ant linear systems.
Fuel cells produce electricity from hydrogen-rich fuels via chemical reactions
without burning fuel (Larminie and Dicks 2001; Barbir 2005; Nehrir and Wang
2009; Gou et al. 2010; Hoffmann and Dorgan 2012; Eikerling and Kulikovsky
2014). The type of fuel cells considered in this book is the proton-exchange
membrane (PEM) fuel cells, also known as polymer exchange membrane fuel
cells. A PEM fuel cell is a triode composed of an anode, membrane, and cathode.
Hydrogen is pumped from the anode side, and oxygen is pumped from the cathode
side. They are the most developed and the best understood type of fuel cells used for
both mobile and stationary applications. PEM fuel cells are devices that rely on the
chemical production of electric energy and water by mixing in a specific manner
hydrogen and oxygen (a process reverse to water electrolysis). This process was
discovered in the middle of the nineteenth century (almost 200 years ago), but due to
its multidisciplinary nature complexity, it took a long time to become a mature
technology. It is interesting to point out that in the 1960s within the Apollo Space.
Program, NASA used fuel cells to provide water for astronauts from the tanks of
oxygen and hydrogen.

© Springer Nature Switzerland AG 2019 1

V. Radisavljević-Gajić et al., Multi-Stage and Multi-Time Scale Feedback Control
of Linear Systems with Applications to Fuel Cells, Mechanical Engineering Series,
https://doi.org/10.1007/978-3-030-10389-7_1
2 1 Introduction

Modeling, control, and simulation of PEM fuel cells have been a very active
research area; see, for example, Pukrushpan et al. (2004a, b), Nehrir and Wang
(2009), Gou et al. (2010), Wang et al. (2011, 2013), Barelli et al. (2012), Matraji
et al. (2012, 2013, 2015), Bhargav et al. (2014), Jiao (2014), Wang and Guo (2015),
Li et al. (2015a, b), Naghidokht et al. (2016), Wu and Zhou (2016), Zhou et al.
(2017), Hong et al. (2017), Tong et al. (2017), Daud et al. (2017), Reddy and Samuel
(2017), Majlan et al. (2018), Sankar and Jana (2018a, b), and references therein. The
importance of mathematical modeling for studying fuel cell dynamics is emphasized
in Fuhrmann et al. (2008): “The operation of fuel cells with polymer electrolyte
membranes (PEMs) is based on complex interactions of physical, chemical, and
electrochemical processes on multiple time scales. A quantitative and qualitative
understanding of this complex matter is possible only on the base of mathematical
models.” Mathematical modeling of fuel cells should be done very carefully com-
bining knowledge from several scientific and engineering disciplines such as math-
ematics, physics, chemistry, system analysis, and control engineering.
One of the most important applications of PEM fuel cells is for electric cars. A
single plate PEM fuel cell, whose width is very narrow, 1 mm, produces the voltage
of only 0.7 V. It has the current density of
0.8 A/cm2 so that a plate of only
10 cm  10 cm produces 56 W. To get much higher voltage and electric power,
n fuel cell plates are connected in series to form a fuel cell stack. To drive a fuel cell-
powered electric car, a nominal electric power of ~40 kW (sufficient to provide
electricity for eight average houses) is needed that requires PEM fuel cell electric
power of at least 80 kW, which can be provided by 228 fuel cell plates of dimension
25 cm  25 cm connected in series. For comparison, to produce ~80 kW, Tesla
battery-powered electric cars use 8000 batteries of 1 V. Note that Honda Clarity,
Hyundai Tucson already on our roads, and Mercedes-Benz B-Class (still in the
testing phase) use fuel PEM cells of 100 kW. Toyota Mirai (also commercially
available) has a PEM fuel cell of 114 kW, as reported in Rojas et al. (2017). A
very recent article by Samuelsen (2017) published in IEEE Spectrum was entitled
“The automotive future belongs to fuel cells: Range, adaptability, and refueling time
will ultimately put hydrogen fuel cells ahead of batteries.” Several research publi-
cations study modeling and control issues of fuel cells (particularly PEM fuel cells)
for automotive applications; see, for example, zur Megede (2002), Pukrushpan et al.
(2004a, b), Mitchell et al. (2006), Wang and Peng (2014), Haddad et al. (2015),
Reddy and Samuel (2017), Han et al. (2017), and Zhou et al. (2017, 2018).
The two- and three-stage designs of feedback controllers were simplified and
specialized in Radisavljevic-Gajic and Rose (2014), Radisavljevic-Gajic (2015a, b),
and Radisavljevic-Gajic et al. (2017) for two- and three-time-scale systems (also
known as singularly perturbed systems) that have a natural decomposition into slow
and fast subsystems, and, hence, they are very well suited for the two- and three-
stage feedback controller design techniques. Dynamic systems with slow and fast
state space variables play important roles in control engineering; see, for example,
the books and overview papers by Kokotovic et al. (1999), Naidu and Calise (2001),
Gajic and Lim (2001), Liu et al. (2003), Dimitriev and Kurina (2006), Zhang et al.
(2014), Kuehn (2015), and references therein.
1 Introduction 3

Several time scales are present in many real physical systems that have compo-
nents of different nature (electrical, mechanical, chemical, thermodynamic, electro-
chemical). For example, advanced heavy water reactor (Shimjith et al. 2011a, b;
Munje et al. 2014) has three time scales. Dynamics of fuel cells evolves in at least
three, possibly four, time scales (Zenith and Skogestad 2009). It was shown in Zenith
and Skogestad (2009) that a proton-exchange membrane fuel cell (PEMFC) system
has three subsystems operating in three different time scales corresponding to three
different time constants: electrochemical subsystem operating in seconds, chemical
part of the PEMPC system (energy balance and mass balance) operating in minutes,
and electrical part of the PEMFC system operating in milliseconds. Road vehicles
possess the multi-time scale dynamics as demonstrated in Wedig (2014). Chemical
reaction networks can be modeled using the multi-time-scale analysis (Lee and
Othmer 2010). It is interesting to point out that the Hodgkin-Huxley equation that
models nerve electric conductivity has a singularly perturbed form as reported in
Cronin (2008). The paper by Jalics et al. (2010) showed that the neuron dynamic
model can be studied in three time scales. In power electronics, many devices
operate in three time scales (Umbria et al. 2014), and in general, power systems
composed of electrical, mechanical, and electronic components possess several time
scales due to the presence of several time constants of different order of magnitudes.
Three time scales can be found in helicopter dynamics (Esteban et al. 2013) and five
times scales are needed to model electron dynamics (Kummrow et al. 1999).
The power of the two- and three-stage (and in general multistage) feedback
design techniques is summarized in the following:
(a) Different types of controllers (eigenvalue assignment, optimal, robust, reliable,
etc.) can be designed for different parts of the system (subsystems) using
corresponding feedback gains obtained by performing calculations (design)
only with subsystem (reduced-order) matrices.
(b) Local subsystem feedback gains, for example, in the case of the two-stage
feedback design, G1 and G2, which control local subsystems are compounded
into one full-state feedback gain via a simple formula Geq ¼ Geq(G1, G2), leading
to the standard full-state feedback controller. The corresponding block diagram
for the two-stage design is presented in Fig. 1.1.
(c) Computational requirements are drastically reduced (especially for two- and
three-time scale linear systems) since all numerical operations are done with
matrices of the reduced-order corresponding to the subsystems.

Fig. 1.1 Two-stage full-

state feedback
u = −Geq x(t) System x(t )
Geq ¼ Geq(G1, G2) with G1 (t) = Ax(t) + Bu(t)
controlling a part of the
system and G2 controlling
‐
the rest of the system
Geq = Geq (G1 ,G2 )
4 1 Introduction

(d) Very high accuracy can be achieved since numerical ill-conditioning of higher-
order matrices can be eliminated and computations performed with well-
conditioned lower-order matrices (especially for two- and three-time scale
systems).
(e) The design can be extended for the development of corresponding two- and
three-stage observers and the Kalman filter, as well observer- and Kalman filter-
driven controllers (hence, it can be also extended to stochastic systems) includ-
ing their two- and three-time-scale counterparts.
(f) The design is independent for each local subsystem so that it provides flexibility
for the development of partial full-state feedback (for the subsystem for which all
state variables are available for feedback) and partial output feedback (for
subsystem for which only the output signal is available for feedback).
(g) Robustness and reliability can be facilitated by using the two- and three-stage
designs, as well as the feedback control-loop security can be improved, which
appears to be very important these days, especially for cyber physical systems.
(h) The design of desired local controllers, observers, and filters might be feasible
even in the cases when the global system is not controllable (observable), but the
local subsystems are controllable (observable), in which cases those controllers,
observers, and filters could have been designed for particular subsystems of the
linear dynamic system.
The statement made in item (h) can be demonstrated by using the following
simple example.
Example 1.1 Consider the linear continuous-time system
2 3
dx1 ðt Þ     
dxðt Þ 6 dt 7 a1 0 x1 ð t Þ b
¼4 ¼ þ 1 uðt Þ, b1 6¼ 0, b2 6¼ 0
dt dx2 ðt Þ 5 0 a2 x 2 ð t Þ b2
dt
Finding the controllability matrix, that is,
 
b a1 b1
CðA; bÞ ¼ ½B AB ¼ 1
b2 a2 b2

it can be easily seen that this system is not controllable for a1 ¼ a2. However, the
local subsystems possess local controllability for b1 6¼ 0, b2 6¼ 0 so that the local
linear-quadratic optimal controllers can be designed or the local closed-loop system
eigenvalues can be placed in the desired location. A similar example may be
constructed for local observability.
The two-, three-, and four-stage feedback designs are applicable to almost all
classes of discrete- and continuous-time linear time-invariant systems as well as to
linearized time-invariant systems. The feasibility conditions will be spelled out in the
corresponding chapters when we present these multistage feedback design
techniques.
1 Introduction 5

In this book, we will discuss the possibilities of extending the two-, three-, and
four-stage feedback design techniques to general multistage feedback designs of
large-scale (complex) systems such that good features (a)–(g) of the two- and three-
stage feedback design are preserved. Some of the above features, specially feature
(g), appear to be extremely important these days for large-scale systems such as
smart power grids, Internet, communication networks, and networks in systems
biology and chemistry. In Fig. 1.2, we present a symbolic schematics of the
multistage feedback controller design of linear systems with different controllers
designed independently for different parts (subsystems) of the system.
The linear time-invariant systems to be subjected to two-, three-, and four-stage
and in general multi-stage feedback designs should be first appropriately partitioned
and their subsystems identified. The partitioning can be done using several criteria:
(a) Based on the physical nature of the subsystem parts (system natural
decomposition)
(b) According to the conditions that must be satisfied such that the partitioned
system is feasible for the multistage feedback design
(c) Based on mathematical conditions that must be satisfied to solve the
corresponding design equations
(d) Control needs (which parts of the system should be independently controlled via
local feedback controllers)
(e) Grouping the state space variables such that the subsystems satisfy control-
oriented assumptions (conditions) needed for the design of local controllers,
observers, or filters such as controllability (stabilizability) and/or observability
(detectability)

GOptimal Gev _ assigned

GH 2

GObs

G0 = 0

GLQ
G H∞ G??

Fig. 1.2 Symbolic schematics of the multistage feedback controller design: GLQ, optimal linear-
quadratic; GH2 , H2-optimal; GH1 , H1-robust optimal; GObs, observer-based controller; Gev _ assigned,
eigenvalue assignment; G0 ¼ 0, no subsystem feedback control is applied; G??, any linear feedback
controller
6 1 Introduction

1.1 General Remarks

The results to be presented in the first four chapters on the two- and three-stage
feedback controller designs for continuous- and discrete-time linear dynamic sys-
tems represent already the mature controller system design techniques published by
the authors in the journals and presented at the conferences. Presently, the authors
are finalizing the work on three-stage feedback controller design for discrete-time
systems and the four-stage feedback controller design for continuous-time systems.
These results are also specialized to corresponding three- and four-time-scale linear
systems with applications to proton-exchange membrane fuel cells. From our expe-
rience, the more design stages are involved, the more efficient the controllers
become, but conditions for the applicability of these controllers and solutions of
corresponding nonlinear algebraic equations become more and more difficult. The
main ideas and complete derivations of discrete-time three-stage feedback design
and four-stage continuous-time feedback controller design are presented, respec-
tively, in Chaps. 5 and 6, but these areas still remain open for research, especially the
applications to the three-time-scale discrete-time linear systems and to the four-time-
scale continuous-time linear dynamic time-invariant systems.

1.2 Applications to Other Classes of Dynamic Systems

The results presented in this research monograph are for linear continuous- and
discrete-time time-invariant dynamic systems. The results obtained can be directly
extended to the linearized models of nonlinear dynamic systems. The authors believe
that this study can be also extended to linearized models of distributed parameter
systems described in the modal coordinates by infinite sets of second-order ordinary
differential equations, models of flexible space structures, and second-order
nonclassically damped linear mechanical systems. Linearized models of distributed
parameter systems (infinite dimensional systems described by partial differential
equations) can be represented in the modal coordinates by infinite sets of second-
order ordinary differential equations (Meirovitch and Baruh 1983; Baruh and Choe
1990). Flexible structures, especially large space flexible structures, can be modeled
by an infinite series of pure oscillators (having eigenvalues on the imaginary axis)
and lightly damped oscillators (having eigenvalues in the stable half complex plane
very close to the imaginary axis) (Gawronski and Juang 1990; Gawronski 1994,
1998). These techniques will facilitate conditions for applications of two-, three-,
and four-stage feedback controller design techniques to the corresponding classes of
dynamic systems.
1.4 Book Organization 7

1.3 Improved System Robustness, Reliability, and Security

Designing individual controllers for each local subsystem will improve the system
robustness, reliability, and security, which are particularly important these days for
large (complex)-scale systems such as power grids, water supply networks, Internet,
communication networks, biological networks, chemical networks, and in general
cyber physical systems. Note that in the multistage feedback design techniques, the
total number of the feedback loops remains the same as for the original full-state
feedback systems, but the feedback loops are grouped according to the subsystems
and feedback algorithms used, hence naturally less prompt to global failures. If one
group of the feedback loops fails, the remaining groups will be working properly,
hopefully still providing a satisfactory system performance. We expect that, in the
future, the multistage feedback design can be successfully extended to control of
cyber physical systems and corresponding linear-quadratic dynamic games
(Pasqualetti et al. 2015; Zhu and Basar 2015). For example, some feedback gains
can be specifically designed to reject external malicious attacks on particular
subsystems.

1.4 Book Organization

After the introductory chapter, in Chap. 2, the continuous-time two-stage feedback

controller design technique of linear time-invariant dynamic systems is presented in
detail. All design algebraic and differential equations, expressions for the newly
derived matrix formulas needed for the design, as well as the conditions under which
the two-stage design is applicable are presented. The results obtained are specialized
to the two-time scale singularly perturbed linear systems, emphasizing the design
simplification for this class of linear time-invariant systems. To that end, the
presentation follows the fundamental results of Radisavljevic-Gajic and Rose
(2014). The design is first demonstrated on a tenth-order hydrogen gas reformer
(gas processing system) that produces hydrogen from hydrogen-rich fuels. The
produced hydrogen is needed for the PEM fuel cell operation. The eigenvalue
assignment controller design for both slow and fast subsystems is presented. In
addition, a hybrid linear-quadratic optimal slow and eigenvalue assigned fast con-
troller are designed for this two-time scale system.
In the second part of this chapter, we apply the two-stage design methodology to
the eight-order model of a PEM fuel cell following the results of Radisavljevic-Gajic
et al. (2015). We present also efficient fixed-point and Newton method type numer-
ical algorithms for solving efficiently the design algebraic equations as systems of
linear algebraic equations. At the end of this chapter, we design an observer for the
considered fuel cell model to estimate all state variables needed for the full-state
feedback controller design (either eigenvalue assigned or linear-quadratic optimal
controllers). Like in the case of the hydrogen gas reformer, we independently design
8 1 Introduction

for this fuel cell the eigenvalue assignment controllers for both the slow and fast
subsystems, as well hybrid optimal slow and eigenvalue assigned fast controller for
the overall PEM fuel cell model.
Chapter 3 considers the discrete-time two-stage feedback controller design. The
presentation follows the results obtained in Radisavljevic-Gajic (2015a, b). Like in
the continuous time, all design algebraic and differential equations, expressions for
the newly derived matrix formulas needed for the design, and the conditions under
which the two-stage design is applicable are derived and established. In addition, the
obtained results are specialized and simplified to the two-time scale discrete time-
invariant systems. Both the slow time scale formulation and the fast time scale
formulation of this class of systems are considered. The design efficiency was
demonstrated using examples of an electric power system (slow time scale formu-
lation) and a steam power system (fast time scale formulation).
Chapter 4 presents the continuous-time three-stage feedback controller design.
We have first indicated specifics of the three-stage feedback controller design and
differences comparing to the two-stage feedback controller design. We have
presented all design formulas, derived corresponding algebraic equations, and
established conditions under which such a design is possible. Like in Chaps. 2 and
3, we have specialized and simplified the design to the three-time-scale singularly
perturbed systems and shown how to solve all required nonlinear equations as
systems of linear algebraic equations using the fixed-point iterations. The efficiency
of the presented three-stage three-time feedback controller design is demonstrated on
the example of an eight-order PEM fuel cell model. Several controller types were
designed: (1) eigenvalue assignment controllers for all three subsystems; (2) optimal
linear-quadratic controller which is designed only for the slow subsystem with no
controllers used for the fast and very fast subsystems (partial optimization); and
(3) optimal linear-quadratic controllers which are designed independently for each
subsystem (slow, fast, and very fast) using the subsystem data only. The presentation
of this chapter follows closely the recent papers of the authors Radisavljevic-Gajic
and Milanovic (2016) and Radisavljevic-Gajic et al. (2017).
Chapter 5 parallels the continuous-time derivations of Chap. 4 and completely
derives the three-stage feedback controllers for the general class of linear time-
invariant discrete-time systems. The chapter also discusses the main ideas and
potential derivations of the discrete-time three-stage three-time scale feedback
controller design. Presently, the problem formulation for this class of systems is
not settled down, and there are at least four different formulations that potentially can
be used for the three-stage three-time-scale feedback controller design of linear
discrete-time systems. Hence, this topic remains an interesting open area for future
research.
Chapter 6 presents the complete derivations for the continuous-time four-stage
feedback controller design of linear time-invariant systems. Even though the results
reported in this chapter are an extension of the results reported in Chap. 4 for the
continuous-time three-stage feedback controller design, these derivations are differ-
ent than those in Chap. 4. The derivations here are much more involved, and the
obtained algebraic equations needed for the design are much more complex. No
1.4 Book Organization 9

obvious method exits at this time how to solve these algebraic equations in general.
Hence, they will be the subject of the future research. We hope that the algebraic
equations derived will be considerably simplified in the case of four-time scale
singularly perturbed systems, where the small singular perturbation parameters
present in these algebraic equations might provide ideas how to solve these equa-
tions either using the fixed-point iterations or the Newton method. This formulates
another open research problem in this control systems area.
Chapter 7 discusses modeling, system analysis, and control issues with potential
limitations and constraints for proton-exchange membrane fuel cell mathematical
models. We start with the linear third-order model of El-Sharkh et al. (2004) and the
bilinear third-order model of Gemmen (2003) and Chiu et al. (2004) that consider
three fundamental fuel cell state space variables: hydrogen pressure, oxygen pres-
sure, and the cathode side water vapor pressure. We show that these third-order
models have some fundamental controllability problems. The presentations of the
third-order fuel cell mathematical models follow closely the author’s papers
Radisavljevic (2011) and Radisavljevic-Gajic and Graham (2017). In addition, in
the first part of this chapter, we discuss also the fifth-order nonlinear model devel-
oped by Na and Gou (2008) and considered in detail in Gou et al. (2010).
In the second part of this chapter, we discuss also the nonlinear fifth-order fuel
cell model derived in Milanovic et al. (2017) for the Greenlight Innovation G60
Testing station fuel cell used in the corresponding Villanova University laboratory.
State space variables of this model are mass of oxygen in cathode, mass of nitrogen
in cathode, mass of hydrogen in anode, mass of water vapor in anode, and mass of
water vapor in cathode. The model shows very good agreement with the experimen-
tal results. At the end of this chapter, an eight-order model of a PEM fuel cell used in
automotive applications is presented based on the results reported in Pukrushpan
et al. (2004a, b). In addition to mass of oxygen in cathode, mass of nitrogen in
cathode, mass of hydrogen in anode, mass of water vapor in anode, and mass of
water vapor in cathode, the model of Pukrushpan et al. (2004a, b) as the state
variables have the dynamics of the gas in intake (supply) and outtake (return)
manifolds due to the fuel cell particular application for electric cars. This model
can be studied in two, three, and eventually four time scales, and it is considered in
several case studies in this monograph.
Chapter 8 presents some fundamental control strategies for a hydrogen natural
gas processing system (known also as the hydrogen gas reformer) that provides pure
hydrogen from the natural gas via simple physical processes with efficiency of
roughly 50%. That hydrogen is then used in PEM fuel cells. The chapter is based
on the author’s research paper Radisavljevic-Gajic and Rose (2015). A reduced-
order observer is designed first to estimate the state variables of the tenth-order
hydrogen gas reformer. The controller proposed has two feedback controllers and
one feed-forward controller. The feedback controllers are an integral controller that
copes with constant disturbances and the linear-quadratic optimal controller. The
feed-forward controller cancels the disturbance caused by the fuel cell current. The
results are derived via a rigorous dynamic linear-quadratic optimization. The simu-
lation results show that the controller designed performed very well rejecting a large
10 1 Introduction

disturbance very quickly. Even more, the proposed controller outperforms the
corresponding full-order observer-based controller used for the same hydrogen
processing system.
Chapter 9 discusses ideas and formulates research problems for extensions of the
presented methodology to the general multistage feedback controller design for
linear dynamic systems composed of N subsystems, including linear dynamic
systems that operate in N time scales. N-stage and N-time scale linear feedback
controller design is the final research goal. This research monograph is the first step
in that direction, and we believe that it will take a couple of years before the general
multistage feedback controller design problem is completely solved. Presently, it is
an interesting and challenging control engineering research area.

1.5 Notes

Chapters 2, 3, 4, 7, and 8 use in parts the material from our previous published
journal and conference papers. The remaining four chapters are our original contri-
butions. Permissions for the use of such material in this research monograph were
granted to us by the American Society of Mechanical Engineers (ASME) for three
journal and five conference papers and by Elsevier for two journal papers. Acknowl-
edgments of granted permissions for each particular paper are done at the end of each
chapter clearly identifying the particular book sections, papers, and the permission
granting publisher.
Chapter 2
Continuous-Time Two-Stage Feedback
Controller Design

In this chapter, we first present a general algorithm for a two-stage feedback

controller design for linear continuous-time, time-invariant, dynamic systems fol-
lowing the results of Radisavljevic-Gajic and Rose (2014), Sect. 2.1. The proposed
design significantly reduces the computational requirements and provides flexibility
of designing different types of controllers for different dynamic parts of the system –
subsystems that form the given system.
The newly proposed design is further simplified and specialized for linear
dynamic systems with slow and fast modes (singularly perturbed linear systems)
in Sect. 2.2. The corresponding algorithm is applied efficiently for design of
feedback controllers for a hydrogen gas reformer that produces hydrogen (from
hydrogen rich fuels like natural gas or methanol) to be used in fuel cells, Sect. 2.3. In
Sects. 2.4 and 2.5, we demonstrate the use of the presented algorithm for design of
feedback controllers for a proton-exchange membrane fuel cell, including the design
of observer-based controllers.

2.1 Two-Stage Design of Linear Feedback Controllers

Design of linear state feedback controllers plays an important role in engineering

applications (Ogata 1995; Sinha 2007; Chen 2012). In this section, a two-stage
design algorithm developed by Radisavljevic-Gajic and Rose (2014) is presented.
Consider a linear time-invariant dynamic system represented in its partitioned
form by

© Springer Nature Switzerland AG 2019 11

V. Radisavljević-Gajić et al., Multi-Stage and Multi-Time Scale Feedback Control
of Linear Systems with Applications to Fuel Cells, Mechanical Engineering Series,
https://doi.org/10.1007/978-3-030-10389-7_2
12 2 Continuous-Time Two-Stage Feedback Controller Design

2 3
dx1 ðt Þ     
dxðt Þ 6
6 dt 7
7
A11 A12 x 1 ðt Þ B11
¼4 ¼ þ uðt Þ ¼ Axðt Þ þ Buðt Þ ð2:1Þ
dt dx2 ðt Þ 5 A21 A22 x 2 ðt Þ B22
dt

where x(t) 2 Rn, x1 ðt Þ 2 Rn1 , x2 ðt Þ 2 Rn2 , and n ¼ n1+n2 represent state space vari-
ables, u(t) 2 Rm is the system control input vector, and Aij and Bii and i, j ¼ 1,2, are
constant matrices of appropriate dimensions. Matrices A11 and A22 define subsystems
of dimensions n1 and n2, respectively, corresponding to the state variables x1(t) and
x2(t). Matrices A12 and A21 define couplings between the subsystems.
In the follow-up of the section, it will be shown how to simplify design of
independent controllers that operate independently on different subsystems
(represented by A11 and A22) despite strong coupling among subsystems (represented
by A12 and A21) and both subsystems having a common input u(t).
In the following, we will show how the subsystem feedback gains can be found
independently and then implemented as a compounded full-state feedback gain; see
Fig. 1.1.
The new two-stage linear feedback algorithm is based on finding appropriate
forms of two similarity transformations (Chen 2012), applied successively to system
defined in (2.1) to achieve a system block triangular form. At the end of the design,
the linear feedback gain is obtained in the original system coordinates defined in
(2.1).
The first results about the two-stage design of linear feedback controllers in both
continuous- and discrete-time domains appeared in the classic works of Philips
(1980a, b, 1983). The work of Phillips was simplified, and an algorithm that reduces
the computational requirements was presented in Radisavljevic-Gajic and Rose
(2014). The two-stage linear feedback controller design of Radisavljevic-Gajic and
Rose (2014) is presented in five steps given below.
Stage 1
Step 1: Apply the following change of variables to (2.1)

ηðt Þ ¼ Lx1 ðt Þ þ x2 ðt Þ ð2:2Þ

where L satisfies a nonsymmetric, nonsquare, Riccati algebraic equation defined by

LA11  A22 L  LA12 L þ A21 ¼ 0 ð2:3Þ

The nonsymmetric algebraic equation (2.3) appears in some areas of systems and
control theory. It has been extensively studied by control systems and applied
mathematics researchers; see, for example, Medanic (1982), Gao and Bai (2010),
and references therein. It can be solved in general using the eigenvector method,
Medanic (1982).
2.1 Two-Stage Design of Linear Feedback Controllers 13

The original and new state variables are related via the following linear
transformation
      
x1 ð t Þ x ðt Þ I 0 x1 ð t Þ
¼ T 1new 1 ¼ ð2:4Þ
η ðt Þ x2 ð t Þ L I x2 ð t Þ

and its inverse

      
x1 ðt Þ 1 x1 ð t Þ I 0 x1 ð t Þ
¼ T 1new ¼ ð2:5Þ
x2 ðt Þ ηðt Þ L I ηðt Þ

The application of transformation (2.2) to (2.1) produces

2 3
dx1 ðt Þ     
6 dt 7 A  A12 L A12 x1 ð t Þ B11
6 7 ¼ 11 þ uð t Þ
4 dηðt Þ 5 0 A22 þ LA12 ηðt Þ B22 þ LB11
ð2:6Þ
dt
    
A1 A12 x1 ðt Þ B11
¼ þ uð t Þ
0 A2 η ðt Þ B2

where

A1 ¼ A11  A12 L, A2 ¼ A22 þ LA12 , B2 ¼ B22 þ LB11 ð2:7Þ

Step 2: Use feedback control

uðt Þ ¼ G2 ηðt Þ þ vðt Þ

to set up the closed-loop eigenvalues of the η-subsystem, which leads to the

following linear system
2 3
dx1 ðt Þ     
6 dt 7 A1 A12  B11 G2 x1 ð t Þ B11
6 7¼ þ vð t Þ ð2:8Þ
4 dηðt Þ 5 0 A2  B2 G2 ηðt Þ B2
dt

Stage 2
Step 3: Apply another change of state variables as

ξðt Þ ¼ x1 ðt Þ  Pηðt Þ ð2:9Þ

where P satisfies the linear Sylvester algebraic equation (Chen 2012)

14 2 Continuous-Time Two-Stage Feedback Controller Design

A1 P  PðA2  B2 G2 Þ þ A12  B11 G2 ¼ 0 ð2:10Þ

with matrices A1 , A2 , B2 previously defined in (2.7). This defines another similarity

transformation
" # " # " #" #
ξðt Þ x1 ð t Þ I P x1 ð t Þ
¼ T 2new ¼ ð2:11Þ
η ðt Þ ηðt Þ 0 I ηðt Þ

whose inverse is
" # " # " #" #
x1 ð t Þ ξðt Þ I P x1 ð t Þ
¼ T 1
2new ¼ ð2:12Þ
ηðt Þ ηðt Þ 0 I ηðt Þ

which leads to
2 3
dξðt Þ " #" # " #
6 dt 7 A1 0 ξðt Þ B11  PB2
6 7¼ þ vð t Þ ð2:13Þ
4 dηðt Þ 5
0 A2  B2 G2 η ðt Þ B2
dt

Step 4: Use state feedback

vðt Þ ¼ G1 ξðt Þ

for the ξ-subsystem, which produces

2 3
dξðt Þ " #" #
6 dt 7 A1  ðB11  PB2 ÞG1 0 ξðt Þ
6 7
4 dηðt Þ 5 ¼ ð2:14Þ
B2 G1 A2  B2 G2 ηðt Þ
dt

Since (2.14) is a lower triangular matrix, its eigenvalues are the union of the
eigenvalues of A2  B2G2 and A1  (B11  PB2)G1, both set up at the subsystem
levels.
Step 5: The feedback gains in the original coordinates are obtained using the
similarity transformations as follows
2.1 Two-Stage Design of Linear Feedback Controllers 15

" # " #
x1 ð t Þ ξðt Þ
uðxðt ÞÞ ¼ ½0 G2   ½ G1 0 
η ðt Þ ηðt Þ
" # " #
x1 ð t Þ x1 ð t Þ
¼ ½ 0 G2   ½ G1 0 T 2new
ηðt Þ η ðt Þ
" #
x1 ð t Þ
¼ f½ 0 G2  þ ½ G1 0 T 2new gT 1new
x2 ð t Þ
( " # )" #" #
I P I 0 x1 ð t Þ ð2:15Þ
¼  ½ 0 G2  þ ½ G1 0 
0 I L I x2 ð t Þ
" #" #
I 0 x1 ð t Þ
¼ ½ G1 G2  G1 P 
L I x2 ð t Þ
" #
x1 ðt Þ
¼ ½ G1 þ ðG2  G1 PÞL G2  G1 P 
x2 ðt Þ
¼ G1eq x1 ðt Þ  G2eq x2 ðt Þ ¼ Geq xðt Þ ¼ ½ G1eq G2eq xðt Þ

with the equivalent feedback gains from the original state space variables equal to

G1eq ¼ G1 þ ðG2  G1 PÞL, G2eq ¼ G2  G1 P ð2:16Þ

The proposed designs will require solutions of systems of nonsquare

nonsymmetric algebraic equations. Note that the nonsymmetric, nonsquare, alge-
braic equation has many real solutions (Medanic 1982) and any real solution in
general will serve the purpose of the considered two-stage design of linear feedback
controllers. The proposed technique in general is not unique (due to nonuniqueness
of L ) which is an advantage so that the presented design can provide several
decomposition pairs of subsystems, and for its implementability, it is sufficient
that only one of the subsystem pairs satisfies the design conditions (assumptions).
A solution of the nonsymmetric, nonsquare algebraic equation defined in (2.3) can
be obtained via the eigenvector method from the generalized eigenvectors of the
corresponding n  n matrix H; see, for example, Medanic (1982) and Bingulac and
Van Landingham (1993).
 nn
A11 A12
H¼ ð2:17Þ
A21 A22

Using, for example, the algorithm of Bingulac and Van Landingham (1993), it is
required that a matrix V is formed from corresponding real eigenvectors of H, and,
for all complex-conjugate eigenvectors of H, we put in matrix V both its real and
16 2 Continuous-Time Two-Stage Feedback Controller Design

imaginary parts and discard their complex-conjugate pairs. Partitioning the matrix
V as
 
  V n111 n1 V n211 n2
V nn
¼ V nn1
V nn2
¼ ð2:18Þ
1 2
V n122 n1 V n222 n2

any solution for L can be obtained using the formula (Medanic 1982; Bingulac and
Van Landingham 1993)

L ¼ V 12 V 1
11 ð2:19Þ

Hence, any collection of n1 eigenvectors of matrix H that provides invertible matrix

V11 will serve the purpose of the presented two-stage linear feedback design. Since
there are many permutations of the eigenvectors of H, in general, there are many
feasible solutions, and hence there are no problems with the existence of a solution of
equation (2.3).

2.2 Two-Stage Feedback Design for Systems with Slow

and Fast Modes

Linear systems with slow and fast modes (singularly perturbed systems (Kokotovic
et al. 1999; Naidu and Calise 2001, also known as multi-time scale systems)) are
particularly well suited for the considered two-stage feedback design. For this class
of systems, in general, numerical ill-conditioning appears if one attempts to design
linear feedback controller using the entire (full-order) system. Singularly perturbed
systems have numerous applications in engineering and sciences (Kokotovic et al.
1999; Naidu and Calise 2001) and play an important role in mechanical and
aerospace engineering (Hsiao et al. 2001; Naidu and Calise 2001; Chen et al.
2002; Shapira and Ben-Asher 2004; Demetriou and Kazantzis 2005; Wang and
Ghorbel 2006; Amjadifard et al. 2011; Kuehn 2015).
In this section, further design simplifications will be achieved by specializing the
proposed design from Sect. 2.1 to singularly perturbed linear systems so that only
solutions of linear algebraic equations will be required. The digital implementation
of the corresponding controllers will allow different sampling rates to be used for the
slow (large sampling rate) and the fast (small sampling rates) controllers. Otherwise,
without the two-stage design, the whole system digital controller will require the
small sampling rate. Moreover, for the class of linear dynamic systems, the
corresponding two-stage design algebraic equations have unique solutions for suf-
ficiently small values of the singular perturbation parameter.
The corresponding time-invariant linear continuous-time singularly perturbed
system is defined by
2.2 Two-Stage Feedback Design for Systems with Slow and Fast Modes 17

2 3
dx1 ðt Þ     
6 dt 7 A11 A12 x1 ðt Þ B11
6 7¼ þ uð t Þ
4 dx2 ðt Þ 5 A21 A22 x2 ðt Þ B22 ð2:20Þ
ε
dt
yð t Þ ¼ C 1 x1 ð t Þ þ C 2 x2 ð t Þ

where ε is a small positive singular perturbation parameter that indicates separation

of state space variables into slow ones, x1(t), and the fast state variables x2(t). The
dimensions of state variables, control input, and constant matrices are defined in
Sect. 2.1. It is a standard assumption in the theory of singularly perturbed systems
that matrix A22 is nonsingular (Kokotovic et al. 1999; Naidu and Calise 2001).
Hence, the following assumption is imposed in the follow-up of this chapter.
Assumption 2.1 Fast subsystem matrix A22 is nonsingular.
In the case of singularly perturbed systems, the proposed simplified two-stage
feedback design of Radisavljevic-Gajic and Rose (2014) has the following steps
grouped in two stages.
Stage 1
Step 1: Solve the algebraic Riccati-type equation (2.3), which in the case of
singularly perturbed systems due to presence of the small positive singular pertur-
bation parameter ε has the form

εLA11  A22 L  εLA12 L þ A21 ¼ 0 ð2:21Þ

A unique solution of (2.21) exists for sufficiently small values of ε under Assump-
tion 2.1. Under that assumption, (2.21) can be efficiently solved by performing
fixed-point iterations on a system of linear algebraic equations as follows

A22 Lðiþ1Þ ¼ A21 þ εLðiÞ A11  εLðiÞ A12 LðiÞ , Lð0Þ ¼ A1
22 A21 , i ¼ 1,2,::, k ð2:22Þ

It can be shown that this algorithm has the rate of convergence of O(ε), meaning that
after i iterations the accuracy of O(εi) is achieved, with O(εi) defined as O(εi) < cεi,
where c is a bounded constant and i is a real number. Moreover, if ε is not sufficiently
small, the eigenvector method (Medanic 1982) can be used to solve (2.21).
Apply the change of variables to (2.20)

x f ðt Þ ¼ Lx1 ðt Þ þ x2 ðt Þ ð2:23Þ

which leads to the following upper block triangular system

18 2 Continuous-Time Two-Stage Feedback Controller Design

2 3
dx1 ðt Þ     
6 dt 7 A  A12 L A12 x1 ð t Þ B11
6 7 ¼ 11 þ uð t Þ
4 dx f ðt Þ 5 0 A22 þ εLA12 x f ðt Þ B22 þ εLB11
ε
dt
    
As A12 x1 ðt Þ B11
¼ þ uð t Þ
0 Af x f ðt Þ Bf
ð2:24Þ

where

As ¼ A11  A12 L, A f ¼ A22 þ εLA12 , B f ¼ B22 þ εLB11 ð2:25Þ

The original and new state variables are related via a similarity transformation as
follows
      
ξðt Þ x1 ð t Þ I P x1 ð t Þ
¼ T 2new ¼ ð2:26Þ
ηðt Þ ηðt Þ 0 I η ðt Þ

whose inverse is
      
x1 ð t Þ ξðt Þ I P x1 ð t Þ
¼ T 1
2new ηðt Þ ¼ 0 ð2:27Þ
η ðt Þ I η ðt Þ

Step 2: Apply feedback control

uðt Þ ¼ G f x f ðt Þ þ vðt Þ

to the fast subsystem, which leads to

2 3
dx1 ðt Þ     
6 dt 7 As A12  B11 G f x1 ð t Þ B11
6 7¼ þ vðt Þ ð2:28Þ
4 dx f ðt Þ 5 0 Af  Bf Gf x f ðt Þ Bf
ε
dt

Stage 2

Step 3: Apply another change of state variables as

xs ðt Þ ¼ x1 ðt Þ  εPx f ðt Þ ð2:29Þ

where P satisfies the Sylvester algebraic equation

Another random document with
no related content on Scribd:
similar places. The two latter occur in chains formed by fission; but
the sexual individuals (which are of distinct sexes, contrary to the
usual hermaphrodite condition of Flat Worms) only appear at stated
times and are not well known. A large number of genera are purely
marine, and one family, the Proboscidae (distinguished by having the
anterior end invaginated by special muscles and converted into a
sensory organ), is entirely so. The most cursory examination of
littoral weeds reveals species of Macrorhynchus, Acrorhynchus,
Promesostoma, Byrsophlebs, and Proxenetes, the character of
which may be gathered from von Graffs great monograph, or from
Gamble's paper on the "British Marine Turbellaria."[65] Much,
however, still remains to be done before we possess an adequate
idea of the occurrence of this group on our coasts.

Fig. 19.—Forms of Rhabdocoelida. A, Mesostoma tetragonum O. F. M.

(Rhabdocoela), × 10; B, Convoluta paradoxa Oe. (Acoela), × 10;
C, Vorticeros auriculatum O. F. M., × 6; D, Monotus fuscus Oe.
(Alloeocoela), × 4. ap, Adhesive papillae; d, intestine; m, pharynx;
ot, otolith; rh, rhabdites; te, testes; ut, uterus with eggs; yg, yolk-
glands; ♂, male, ♀, female genital pores. (A after Braun.)

Some Rhabdocoels are parasitic. Fecampia erythrocephala, which

occurs in the lacunar spaces and alimentary canal of young shore
crabs (Carcinus maenas), is a white cylindrical animal ¼ inch long,
with a red snout. After attaining maturity it works its way out of the
crab and encysts under stones, forming a pyriform mass in shape
like a "Prince Rupert's drop." Within this case the eggs develop, and
the young probably emerge through the open narrow end of the hard
white tube, but how they reach the crab is not known. Graffilla
muricicola is found in the kidney of Murex brandaris and M.
trunculus, at Naples and Trieste; G. tethydicola in the foot of Tethys.
Anoplodium parasiticum occurs among the muscles which attach the
cloaca of Holothuria tubulosa to the body-wall; and A. schneideri
occurs in the sea-cucumber, Stichopus variegatus. These are truly
parasitic forms, constituting a special sub-family. They have no
rhabdites in the skin; the nervous system and sense-organs are only
slightly developed; and the pharynx has undergone a notable
reduction in relation to the simpler mode of obtaining nourishment.
Other cases of association between certain Rhabdocoels (closely
allied to, if not identical with, certain free-living species) and
Lamellibranchs or Sea-urchins, are, however, of another kind. Thus
on the gills or in the mantle cavity of species of Mytilus, Cyprina,
Tellina, and upon the test of Clypeaster, such forms as Enterostoma
mytili, Acmostoma cyprinae, and Provortex tellinae have been found.
But it is probable that these Turbellaria here obtain merely a
temporary shelter and possibly a supply of the food of the mussel or
sea-urchin.

The Alloeocoela afford a well-established case of association.

Monotus fuscus (Fig. 19, D), an abundant, active, elongated animal,
lives on our coasts in the upper part of the littoral zone among
Patella, Balanus, and sometimes Chiton. When the tide is low, the
Monotus, to obtain moisture and darkness, creeps between the
mantle-folds of these animals, where it may readily be found. Upon
the return of the tide it leaves its retreat and creeps or swims about
freely. Other Alloeocoela collect in great numbers in tufts of red-
seaweeds (Florideae). By placing such tufts in vessels, the sea-
water, especially as darkness sets in, begins to swarm with
Cylindrostoma 4-oculatum, species of Enterostoma and
Plagiostoma; P. vittatum, with three violet bands across the white
body, being a particularly obvious form. Vorticeros auriculatum (Fig.
19, C), another abundant species, is remarkable for the long
tentacles which can be completely withdrawn, and in this condition it
completely resembles a Plagiostoma.
The presence of a species (P. lemani) of the characteristically marine
genus Plagiostoma, in the Lake of Geneva, and in one or two other
Swiss lakes, at depths varying from 1 to 150 fathoms, is very
interesting, and is perhaps the only well-established case of the
survival of a once marine Rhabdocoelid under changed conditions.
Plagiostoma lemani is by far the biggest of the group to which it
belongs, being over half an inch in length. It is usually found in fine
mud, sometimes among Chara hispida, and has the general
appearance of an inactive white slug. We are indebted to Forel and
Duplessis for the discovery of this species, and also of
Otomesostoma morgiense, a Mesostoma with an otolith, dredged in
10 to 50 fathoms in the Lake of Geneva, the Lake of Zürich, and
found recently also by Zacharias in the Riesengebirge. The genus
Bothrioplana, first found by Braun in the water-pipes of Dorpat, has
been carefully investigated by Vejdovsky,[66] who places it in a
special family, Bothrioplanidae, among the Alloeocoela. One species
has recently been found near Manchester.

A comprehensive survey of the Rhabdocoelida shows that, with the

chief exception of the Proboscidae, the more lowly organised forms,
the Acoela and Alloeocoela, are marine, whereas the fresh-water
forms are in most cases the most highly organised genera
(Mesostoma, Vortex). But Macrorhynchus helgolandicus, though
minute (1.5-2 mm. long), has a more complex structure[67] than any
other species of the specialised marine genus to which it belongs,
and is a remarkable instance of great complexity being associated
with small size.

Reproduction.—The Rhabdocoelida present the greatest diversity

in the development of the reproductive system. The Acoela and
Alloeocoela have the simplest arrangement. Scattered testes, often
without a distinct membrane, form the spermatozoa, which in most
cases wander into parenchymatous spaces, but in Monoporus
rubropunctatus and Bothrioplana, into distinct vasa deferentia. In
both groups a protrusible penis opens independently to the exterior,
and may be simply muscular or provided with a chitinous armature.
Two ovaries are present, and the oviducts, if distinct, are
continuations of the ovarian membrane. In most forms a "bursa
seminalis," which receives the spermatozoa of another individual, is
appended to the female genital canal. In many of the Alloeocoela,
however, a portion of the ovary is sterile, and its cells, forming a yolk-
gland, feed the fertile portion, the whole structure being then spoken
of as a germ-yolk-gland. In many others (Monotidae) this sterile part
has become an independent yolk-gland, which communicates by
yolk-ducts with the oviducts. The Acoela form no egg-case, the body
of the parent becoming a bag for the ova, which elaborate their own
food-yolk. The Alloeocoela lay hard-shelled eggs, which are
produced in Bothrioplana and Automolos by the activity and
interaction of reproductive organs, resembling closely those of
certain Triclads.[68]

The Rhabdocoela exhibit every stage in the development of a

complex reproductive system, from the simple ovaries and testes of
a Microstoma or Macrostoma, to the intricate system of ducts and
glands of a Macrorhynchus (Proboscidae), in which there is still
much to be made out. The complications of the copulatory organs
chiefly arise from the way in which the spermatozoa are brought into
contact with a nutritive prostatic fluid, or are formed into
spermatophores; and also from the penial armature, which is often
very complex, and may consist of a curved chitinoid hook or a coiled
loop (Promesostoma), of hooks (Proboscidae), or of an intricate
arrangement of plates (Proxenetes); or the penis may take on a
complex corkscrew-like form (Pseudorhynchus). The (frequently
armed) female genital canal usually possesses a bursa seminalis for
the fertilisation of the eggs, but a receptaculum seminis or
spermatheca may serve for the reception, the bursa, for the
lodgment of the spermatozoa of another individual. The fertilised
ovum is provided with a supply of food-yolk and with a shell, which
may be formed in a special diverticulum, the "uterus." The
development of these organs strains the resources of the animal to
the utmost, and in some Proboscidae the alimentary canal is
squeezed out and disintegrates, in order to make room for them.
A few Mesostoma (M. ehrenbergii, M. productum, M. lingua) produce
two kinds of eggs—thin- and thick-shelled. The latter are laid
throughout the summer, and lie dormant through winter. The young
which hatch in spring out of these "winter" eggs develop rapidly, and
when only 7 to 8 mm. long (i.e. one-third the size of the parent)
already possess functional genital organs; the penis, however, is
rudimentary, and incapable of being used for copulation. Hence it is
probable that this stunted progeny self-fertilise their thin-shelled or
"summer" eggs. After the formation of these eggs the same parent is
said (Schneider[69]) to produce thick-shelled or winter eggs, but
however that may be, the first young which hatch from the thin-
shelled ova are produced in great numbers at a time (April to May)
when food is abundant. These grow rapidly to the full size, and then
having attained maturity, cross-fertilise one another's ova, which
become encased in a thick brown shell; and it is these numerous
"winter" eggs that lie dormant throughout the autumn and winter.
Many Mesostoma, and practically all other Rhabdocoela, however,
produce only thick-shelled eggs, and in all cases it is probable that to
these many species owe their wide distribution, the exact range of
which is, however, unknown, as is also the means of dispersal.

Classification of Rhabdocoelida.

ACOELA.
Family. Genus and British species.
Proporidae Proporus venenosus O. Sch. Plymouth.
Monoporus rubropunctatus O. Sch.
Plymouth.
Haplodiscus.
Aphanostomatidae Aphanostoma diversicolor Oe. Common.
A. elegans Jen. Plymouth.
Convoluta saliens Grff. Plymouth,
Millport.
C. paradoxa Oe. (Fig. 19, B). Common.
C. flavibacillum Jen. Plymouth, Port
 Erin, Millport.
Amphicoerus.
Polychoerus.

RHABDOCOELA.
Macrostomatidae Mecynostoma.
Macrostoma hystrix Oe. Stagnant water.
Omalostoma.
Microstomatidae Microstoma lineare Oe. Fresh water.
M. groelandicum Lev. Plymouth, among
Ulva.
Stenostoma (Catenula) lemnae Dug.
Near Cork.
S. leucops O. Sch. Common in fresh
water.
Alaurina claparedii Grff. Skye.
Prorhynchidae Prorhynchus stagnalis M. Sch. In
Devonshire rivers.
Promesostoma marmoratum M. Sch.
Common.
P. ovoideum O. Sch., P. agile Lev.
Plymouth. P. solea O. Sch. Plymouth,
Port Erin. P. lenticulatum O. Sch. Port
Erin.
Mesostomatidae Byrsophlebs graffii Jen. Plymouth,
Millport.
B. intermedia Grff. Millport, Port Erin.
Proxenetes flabellifer Jen. Millport,
Plymouth, Port Erin.
P. cochlear Grff. Millport.
Otomesostoma.
Mesostoma productum Leuck., M. lingua
O. Sch., M. ehrenbergii O. Sch., M.
tetragonum O. F. M. (Fig. 19, A). All at
Cambridge.
 M. rostratum Ehr. Widely distributed. M.
viridatum M. Sch. Manchester. M.
robertsonii Grff., M. flavidum Grff. Both
at Millport.
Bothromesostoma personatum O. Sch.
Preston.
Castrada.
Proboscidae Pseudorhynchus bifidus M‘Int. Millport,
St. Andrews, Port Erin.
Acrorhynchus caledonicus Clap.
Generally distributed.
Macrorhynchus naegelii Köll., M. croceus
Fabr. Plymouth, Millport.
M. helgolandicus Metsch. West coast.
Gyrator hermaphroditus Ehrbg. St.
Andrews. Also common in fresh water.
Hyporhynchus armatus Jen. Plymouth,
Port Erin.
H. penicillatus O. Sch. Plymouth.
Vorticidae Schultzia. Provortex balticus M. Sch.
Generally distributed.
P. affinis Jen., P. rubrobacillus Gamb.
Plymouth.
Vortex truncatus Ehrbg. Abundant in
fresh water.
V. armiger O. Sch. Millport (fresh water).
V. schmidtii Grff., V. millportianus Grff.
Millport. V. viridis M. Sch. Generally
distributed.
Jensenia.
Opistoma.
Derostoma unipunctatum Oe. Edinburgh.
Graffilla.
Anoplodium.
Fecampia erythrocephala Giard.
Plymouth, Port Erin.
Solenopharyngidae Solenopharynx.

ALLOEOCOELA.
Plagiostomatidae Acmostoma. Plagiostoma dioicum
Metsch., P. elongatum Gamb., P.
pseudomaculatum Gamb., P. sagitta
Ulj., P. caudatum Lev., P.
siphonophorum O. Sch., P.
ochroleucum Grff. All at Plymouth.
P. sulphureum Grff. Port Erin. P. vittatum
F. and Leuck. Millport, Plymouth, Port
Erin. P. koreni Jen. Plymouth, Millport.
P. girardi O. Sch. Plymouth, Port Erin,
Valencia.
Vorticeros auriculatum O. F. M. (Fig. 19,
C). Port Erin, Plymouth.
V. luteum Grff. Plymouth.
Enterostoma austriacum Grff. Plymouth,
Port Erin.
E. fingalianum Clap. Skye, Plymouth. E.
coecum Grff. Millport.
Allostoma pallidum van Ben. Millport.
Cylindrostoma 4-oculatum Leuck. Skye,
Millport, Plymouth.
C. inerme Hall, C. elongatum Lev.
Plymouth.
Monoophorum striatum Grff. Plymouth.
Bothrioplanidae Bothrioplana.
Bothrioplana sp.? Manchester.
Otoplana.
Monotidae Monotus lineatus O. F. M., M. fuscus Oe.
(Fig. 19, D). Both common littoral
forms.
M. albus Lev. Plymouth.
Automolos unipunctatus Oe. Skye, St.
Andrews, Plymouth.
 A. horridus Gamb., A. ophiocephalus O.
Sch. Plymouth.

CHAPTER II

TREMATODA

CHARACTERS OF TREMATODES—HABITS AND STRUCTURE OF

TREMATODA ECTOPARASITICA (MONOGENEA)—LIFE-HISTORIES OF
POLYSTOMUM INTEGERRIMUM, DIPLOZOON PARADOXUM, AND
GYRODACTYLUS ELEGANS—TREMATODA ENDOPARASITICA (DIGENEA)—
OCCURRENCE AND HABITS OF DIGENEA—LIFE-HISTORY OF DISTOMUM
MACROSTOMUM—DISTOMUM HEPATICUM AND ITS EFFECTS—BILHARZIA
HAEMATOBIA—BISEXUAL TREMATODES—TABLE OF HOSTS—
CLASSIFICATION.

From the Turbellaria we now pass on to a consideration of the

second great subdivision of the Platyhelminthes, the Trematodes or
"flukes," of which the "liver-fluke" is the best known, since it is one of
the most dangerous parasites that infest domestic animals.

It has been pointed out that the Polyclads, Triclads, and

Rhabdocoels are carnivorous, and that in each of these groups
sporadic cases of parasitism occur. In other words, when the prey is
much larger than the Turbellarian, the latter tends to become a
parasite, and we can trace the development of the parasitic habit
from the gradual association of Turbellaria with Ascidians,
Crustacea, Molluscs, and Polyzoa merely for protective purposes,
through the adoption, not only of the body of the host for shelter, but
of its flesh for food; though it is only in some Rhabdocoels (Graffilla,
etc.) that there exists a degeneration corresponding to the easier
mode of nutrition and simpler life. The Trematodes,[70] however, are
wholly parasitic, either on the outer surface, the gills, or internal
organs of their host, which is almost always a Vertebrate. Some
Trematodes lodge in the mouth; others wander down the
oesophagus into the stomach or intestine, where they fix themselves
to the mucous membrane. Again, others work their way into the
digestive glands by the ducts, and thus become further and further
removed from the external world, and more adapted to live in the
particular organs of that host in which they best flourish. The most
important result of the adoption of this internal habitat by
endoparasitic Trematodes is, however, seen in their life-history. If a
liver-fluke were to deposit its million or so of eggs in the bile-ducts of
the sheep, and these were to develop in situ, the host could not
withstand the increased drain upon its vital resources, and host and
parasites would perish together. Hence it is clear that the infection of
a second host by Trematodes is highly necessary, whether they be
ectoparasitic, in which case the infection is easily effected, when two
hosts are in contact, by the adult worms, as well as when they are
apart, by free-swimming larvae. In endoparasitic Trematodes it is
brought about by the migration of the young to the outer world, their
entrance into a, usually, Invertebrate host and their asexual
multiplication within it, and the capture and deglutition of this
"intermediate host" by the final Vertebrate one. Within the latter the
immature parasites find out the organ in which their parents
flourished, and here they too grow and attain maturity. The chances
of any one egg of an endoparasitic Trematode producing eventually
an adult are, therefore, far less favourable than in the case of an
ectoparasitic form. In other words, while the former must lay a great
number of small eggs, the latter need only deposit a (comparatively)
few large ones, and this fact has a corresponding influence on the
structure of the genitalia in the two cases. The Digenea, which
employ two hosts in a lifetime, have accordingly a different
generative mechanism from that of the Monogenea. The great need
of the latter is a powerful apparatus for adhering to the surface of the
body of its host; while the adaptations which the endoparasite
requires are, in addition, (1) protection against the solvent action of
the glands of its host, (2) the power of firm adhesion to a smooth
internal surface, and (3) the ability not only to produce a large
quantity of spermatozoa and ova, but in the absence of a fellow-
parasite, to fertilise its own ova; and we find these conditions
abundantly satisfied.

Trematoda monogenea (ectoparasitica).

There are four subdivisions of the Monogenea:—

I. Temnocephalidae, with four to twelve tentacles, and one sucker

posteriorly (Fig. 20).

II. Tristomatidae, with two lateral, anteriorly-placed suckers. Oral

suckers are absent, a large posterior sucker is constant, and is often
armed with hooks (Fig. 22, C).

III. Polystomatidae, with, usually, two oral suckers and a posteriorly-

placed adhesive disc armed with suckers and hooks (Figs. 23 and
24).

IV. Gyrodactylidae (Fig. 29).

Habits and Structure of Ectoparasitic Trematodes.

I. Temnocephalidae.—These interesting forms, of which a good

account has lately been written by Haswell,[71] occur on the surface
(rarely in the branchial chamber) of fresh-water crayfish and crabs in
Australasia, the Malay Archipelago, Madagascar, and Chili. Others
have been found on the carapace of a fresh-water tortoise, and in
the branchial chamber of the mollusc Ampullaria from Brazil. Wood-
Mason discovered others, again, in bottles containing spirit-
specimens of Indian fish. Temnocephala is rarely more than a
quarter of an inch long, and looks like a minute Cephalopod or a
broad flattened Hydra. By the ventral sucker each species adheres
to its own particular host, the tentacles being used as an anterior
sucker for "looping" movements. The food, consisting of
Entomostraca, Rotifera, and Diatoms, is first swallowed whole by the
large pharynx (Fig. 20, ph), which can be protruded through the
ventrally-placed mouth, and is then received into a simple lobed
intestine (d). The skin, especially on the surface of the tentacles, is
provided here and there with patches of cilia borne by the cellular
epidermis,—the only undoubted case of external cilia occurring in an
adult Trematode. Minute rhabdites formed in special gland-cells,
occur plentifully on the tentacles, and are another distinctly
Turbellarian feature. The excretory system is peculiar (Fig. 21). Fine
ducts proceed from the various organs of the body, and open to the
exterior by means of a pair of contractile sacs placed on the dorsal
surface. Each sac is a single cell, and within it not one merely, but
several "flames," or bunches of rhythmically contractile cilia, are
present. These are placed on the course of excessively fine canals,
which perforate the protoplasm of this cell. The terminal branches of
the excretory canals end in branched cells, apparently devoid of
"flames."

Fig. 20.—Temnocephala novae-zealandiae Has. × 10. Ventral view to

show the digestive and reproductive systems. (After Haswell.)
Fig. 21.—The same from the dorsal surface, to show the excretory
system (double line), and the nervous system (black and shaded).
(After Haswell.)
d, Intestine; dln, dorso-lateral nerve; dn, dorsal nerve; ex.o, excretory
aperture on dorsal surface; ex.s, terminal excretory sac; m,
mouth; ov, ovary; ovd, oviduct; ph, pharynx; rh, rhabdites; rh.c,
 cells in which the rhabdites are formed; rv, yolk receptacle; sc,
sucker; sh, shell-gland; te, testes; ut, uterus; vg, vagina; vn,
ventral nerve; vs, vesicula seminalis; yd, yolk-duct; yg, yolk-gland.
♀, ♂, common genital pore.

The reproductive system is very similar to that of certain

Rhabdocoels. An armed penis and the female genital duct open into
a genital atrium, and this by a single aperture (♀, ♂, Fig. 20) to the
exterior. The fertilised ovum and yolk are enclosed in a stalked shell
formed in the uterus.

The interest and importance of the Temnocephalidae lies in the fact

that they are almost as much Turbellaria as Trematodes. In habits, in
the character of the skin, the muscular, digestive, and reproductive
systems, they find their nearest allies in Rhabdocoels (Vorticidae).
But in the excretory and nervous systems, the latter composed of
two dorsal, two lateral, and two ventral trunks all connected together
(Fig. 21), they are Tristomid Trematodes. Thus they may fitly connect
an account of the two great groups.

Fig. 22.—A, Nematobothrium filarina van Bened. Nat. size. Two

individuals (a and b) are found together, encysted on the branchial
chamber of the Tunny. B, Udonella caligorum Johns. A Tristomid,
several of which are attached to the ovary of a Copepod
(Caligus), itself a parasite on the gills of the Hake. × 8. C,
Epibdella hippoglossi O. F. M. A Tristomid found on the body of
the Halibut. Nat. size. m, Mouth; ms, lateral suckers; ov, ovary;
ps, posterior sucker; te, testes. (All after P. J. van Beneden.)
II. Tristomatidae and III. Polystomatidae.[72]—The members of these
families are found on the body, or attached to the gills, of fresh-water
and marine fishes. The edible and inedible fish of our coasts have
each their particular ectoparasitic Trematodes; while the Minnows,
Sticklebacks, and Miller's Thumbs of streams and ponds are
attacked by Diplozoon, Gyrodactylus, and other forms. The aquatic
Amphibia also harbour a number. Polystomum integerrimum is
common in the bladder of Frogs, where it leads a practically aquatic
life. Other species of Polystomum inhabit the buccal and nasal
cavities of certain Chelonia, but naturally no terrestrial Vertebrates
are infested externally by these Trematodes. The blood and epithelia
of the host are sucked, and to this end the pharynx has frequently a
chitinous armature to aid in the abrasion or inflammation of the
tissues upon which the parasite feeds. In the case of a Sturgeon
attacked by Nitzschia elongata, a Tristomid, the mouth of the host
appeared to be highly inflamed by these attacks (v. Baer).

Fig. 23.—Octobothrium merlangi Kuhn, from the gills of the whiting, ×

8. int, Intestine; ms, mouth; sc suckers with chitinoid armature; yk
yolk-glands. (After v. Nordmann.)

The suckers, in the two families under consideration, vary in number

and complexity. There is always a powerful apparatus at the hinder
end of the body securing the Trematode firmly to the slimy body or
gills of its host, and, usually in the Polystomatidae, a pair of suckers
at the sides of the mouth accessory to the pumping action of the
pharynx. In Axine, and to a less extent in Octobothrium (Fig. 23), the
suckers are strengthened by a complex hingework of chitinoid bars
or hooks, which serve as insertions for the muscles of the suckers,
and thus increase their efficiency.

The mouth is invariably present just beneath the anterior end of the
body. It leads into a muscular, pumping pharynx (Fig. 24, ph), and
this into a bifurcated intestine which ends blindly. The two openings
of the excretory system lie on the dorsal surface (as in
Temnocephala), and the excretory canals branch through the
substance of the body, ending usually in "flame-cells." The nervous
system is highly developed, and resembles that of Temnocephala
(Fig. 21) in detail. Upon the brain one or even two pairs of eye-spots
are present in the larvae, and may persist throughout life. Tactile
setae occur in Sphyranura, a parasite of the North American
Amphibian Necturus, but a cellular epidermis is apparently rendered
impossible, perhaps from the nature of the mucus in which the body
is bathed, or to the attempts of the host to free itself from these
parasites; and hence an investing membrane is present, which
morphologically is either a modified epithelium, or a cuticle formed
by the glandular secretion of the parenchyma.

Fig. 24.—Polystomum integerrimum Fröh., from the bladder of the

Frog, and seen from the ventral surface. The alimentary canal is
black, the white dots upon it being the yolk-glands, dvi, Ductus
vitello-intestinalis (probably homologous with the Laurer's canal or
"vagina" of Digenea); eh, hooks of sucking disc; int, intestine; m,
mouth; ov, ovary; pe, penis; ph, pharynx; sc, suckers with an
 embryonic hook persisting in each; te, testes; ut, uterus with eggs;
vag, left vagina; vd, vas deferens; yd, yolk-duct; yg, yolk-glands; ♂
♀, common genital aperture. (Modified from Zeller.) × 8.

The reproductive organs of the Polystomatidae may be understood

from Figs. 24, 27, and 28. At the point of union of the oviduct (Fig.
28, ovd), the vitelline ducts (yd), and the commencement of the
uterus (ut), a slender duct is given off which opens into the intestine,
and is known as the "vitello-intestinal canal" (Fig. 24, dvi; Fig. 28,
gic). This duct has apparently the same relations as the "canal of
Laurer" of Digenea,[73] except only that the latter opens to the
exterior directly. In connexion with this vitello-intestinal canal a
"vagina" is present, which in Polystomum and most Monogenea is
paired (Fig. 24, vag), in Diplozoon and in one or two other forms,
however, unpaired. The vagina receives the penis of another
individual during copulation (Fig. 26), and does not appear to have
an homologue in the liver-fluke or other Digenea.

Fig. 25.—Eggs of Monogenea. A, Eggs of Encotylabe pagelli v. Ben.-

Hesse; B, eggs of Udonella pollachii v. Ben.-Hesse (with young
forms just hatching out); C, egg of Microcotyle labracis v. Ben.-
Hesse. (After van Beneden and Hesse.) × 50.

Life-Histories of the Polystomatidae.[74]—Polystomum

integerrimum. After the mutual fertilisation of two individuals, the
eggs are laid in the water by the protrusion of the body of the parent
through the urinary aperture of the Frog. About 1000 eggs are laid in
the spring at the rate of 100 a day for ten days. After about six
weeks, the larva (.3 mm. long) hatches out, and swims about freely
by means of bands of large ciliated cells (Fig. 26, A); but if it does
not meet with a tadpole within twenty-four hours, it dies. Should it,
however, encounter one, the larva creeps along it in a looping
fashion until it approaches the opercular spout, or opening of the
branchial chamber, on the left side; into this it darts suddenly, fixes
itself, and throws off its cilia. Here it remains eight or ten weeks,
feeding, increasing in size, and forming the suckers from behind
forwards. At the time of the tadpole's metamorphosis, the young
Polystomum works its way down the pharynx into the oesophagus
and along the intestine, till it reaches and enters the opening of the
bladder. Three years afterwards it becomes mature.

Sometimes, however, Polystomum experiences another fate. The

larvae settling down on the external gills of a young, recently-
hatched tadpole, and obtaining a richer supply of blood than in the
previous case, grow far more rapidly, so that in five weeks they are
mature, although still in the branchial chamber of the tadpole. They
do not then wander into the alimentary canal, but usually, having
discharged their eggs, die at the time of the tadpole's
metamorphosis. Still more interesting, however, is the difference
between the genitalia in these and in the normal Polystomum. In
contrast with the latter, these possess (1) one testis and a
rudimentary penis; and their spermatozoa differ in structure and
shape from those of the normal Polystomum. (2) The vaginae are
absent, a fact connected with the absence of a functional copulatory
organ. (3) In compensation for the loss of these, a duct connects the
single testis and the point of union of oviduct and yolk-ducts, and by
this self-fertilisation occurs. (4) The uterus is absent; the "ootype" or
duct into which the shell-gland opens, communicating directly with
the exterior. In (1) and (4) these aberrant Polystomum resemble P.
ocellatum, from the Tortoise Emys europaea.
Fig. 26.—Polystomum integerrimum. A, Free-swimming larva, seen
from the ventral surface. × 80. B, Two mature individuals in
mutual coition attached to the bladder of a Frog. × 5. (After Zeller.)
d, Intestine; ex.o, excretory pore, dorsal in position, seen here by
transparency; ey, eye-spots; gl, frontal glands; m, mouth; ph,
pharynx; sd, adhering disc; vag, vagina.

Fig. 27.—A, Egg of Diplozoon paradoxum v. Nord., consisting of a shell

enclosing ov, the actual ovum, surrounded by yc, the yolk-cells; B,
larva just hatched (× 125); C, two Diporpa (I and II) about to unite;
D, conjugation in progress but not yet complete. dt, Dorsal papilla;
e, eye; g, intestine; m, mouth; sc, ad-oral sucker; th, spirally-
wound thread attaching the egg to the gill of the Minnow; vs,
ventral sucker; (in D) I, I, one Diporpa, ventral view; II, II, the
other, dorsal view. (After Zeller.)
 Fig. 28.—Hinder part of the body of Diplozoon paradoxum. The fusion
of the two Diporpa, where they come into contact, is now
complete. They now cross each other like an X, and are twisted,
so that Diporpa I, in front of the point of fusion, is seen from the
dorsal surface; behind, from the ventral surface; and the reverse
is the case with Diporpa II. The compound animal is seen from the
opposite surface to that shown in Fig. 27, D. The digestive and
excretory organs are omitted. (After Zeller.) I Ant. dorsal, dorsal
surface of Diporpa I, facing the anterior end; I Post. ventral,
ventral surface of Diporpa I, posterior end; and similarly for II Ant.
ventral and II Post. dorsal. d, Piece of the intestine showing
opening of, gic, vitello-intestinal canal; ov, ovary; ovd, point of
union of female genital ducts; sc, suckers; te, testis; ut (in Diporpa
I), "ootype" or chamber into which shell-glands open. This is
continuous with the uterus (ut) of Diporpa I; uto, ventral opening of
uterus; vag, vagina, with vd, vas deferens, permanently inserted
into it through the genital pore; yd, yolk-ducts; yg, yolk-glands.

Diplozoon paradoxum.—The life-history of Diplozoon is unique. For

whereas the larvae of most animals grow up, each into a single
adult, in Diplozoon, of the few larvae that survive the dangers of their
free-swimming existence, only those become mature which
conjugate permanently with another individual. But although there
are thus only half as many adult Diplozoon as there were
conjugating larvae (or Diporpa, as they were called when they were
considered distinct forms), yet the total number of eggs produced is
probably as great as if each larva became individually mature.
 Fig. 29.—Gyrodactylus elegans v. Nord., from the fins of the
Stickleback. (After v. Nordmann.) × 125. emb, Embryo.

Diplozoon paradoxum lays its eggs on the gills of the Minnow, which
it frequently infests in great numbers. The ovum divides rapidly at the
expense of the yolk-cells, and in a fortnight a larva (.2 mm. long) of
the shape and complexity shown in Fig. 27, B, hatches out, which,
however, succumbs if it does not meet with a Minnow in five or six
hours. Should it survive, a dorsal papilla, a median ventral sucker,
and a second pair of posterior suckers develop. Thus the Diporpa
stage is attained. These Diporpa may acquire a third and even a
fourth pair of suckers, and continue to live three months, but they
only develop and mature their reproductive organs, if each
conjugates with another Diporpa (Fig. 27, C, D), and this only occurs
in a small percentage of instances. Each grasps the dorsal papilla of
the other by its own ventral sucker, thus undergoing a certain
amount of torsion. Where the two bodies touch, complete fusion
occurs, and, as shown in Fig. 28, the united Diporpa (or Diplozoon,
as the product is now called) decussate, each forming one limb of
the X-shaped Diplozoon, within which the two sets of complex
genitalia develop (Fig. 28).

IV. Gyrodactylidae.—Gyrodactylus (Fig. 29), the structure of which is

in many ways peculiar, produces one large egg at a time. An
embryo, in which the large and smaller hooks of the adhesive disc
can be seen (emb), develops from this egg while still within the body
of the parent, and may give rise to yet another generation within