V1 1 Complete

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/235687178
Quantifying Operability of Working Machines
Thesis · November 2011

DOI: 10.13140/RG.2.1.4702.0004
CITATIONS READS
48 1,563
1 author:
Reno Filla
Scania
37 PUBLICATIONS 408 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Optimal Working Cycle View project
All content following this page was uploaded by Reno Filla on 26 May 2014.
The user has requested enhancement of the downloaded file.

LINKÖPING STUDIES IN SCIENCE AND TECHNOLOGY.
DISSERTATIONS, NO. 1390
Quantifying Operability
of Working Machines
Reno Filla
DIVISION OF FLUID AND MECHATRONIC SYSTEMS

DEPARTMENT OF MANAGEMENT AND ENGINEERING
LINKÖPING UNIVERSITY
SE-581 83 LINKÖPING, SWEDEN
LINKÖPING 2011
Copyright © Reno Filla, 2011
“Quantifying Operability of Working Machines”
Linköping Studies in Science and Technology. Dissertations, No. 1390
ISBN 978-91-7393-087-1
ISSN 0345-7524
Printed by: LiU-Tryck, Linköping
Distributed by:
Linköping University
Division of Fluid and Mechatronic Systems
Department of Management and Engineering
SE-581 83 Linköping, Sweden
Tel. +46 13 281000
http://www.liu.se
To Maria
Only a few people in this world were lucky enough

to run into their true partners - it took outrageous
luck for it to happen, then the sense to recognize it,
and the courage to act. Few could be expected to
have all that, and then to have things go well.
The rest had to make do.
(from “Blue Mars” by Kim Stanley Robinson)
To my parents Inge & Peter
There is one way to understand another culture.

Living it. Move into it, ask to be tolerated as a
guest, learn the language. At some point under-
standing may come. It will always be wordless.
The moment you grasp what is foreign, you will
120
lose the urge to explain it. To explain a phenome-
non is to distance yourself from it.
(from
1 “Smillas’ sense of snow” by Peter Høeg)
Perhaps we are asking the wrong questions.
(Agent Brown in the film "The Matrix")

Abstract
IN WORKING MACHINES the human operator is essential for the performance of the
total system. Productivity and energy efficiency are both dependent not only on inherent
machine properties and working place conditions, but also on how the operator ma-
noeuvres the machine. In order to operate energy-efficient the operator has to experi-
ence the machine as harmonic. This is important to consider during the development of
such working machines.
It is necessary to quantify operability and to include this interaction between the hu-
man operator and the machine in both the later stages of a development project (where
physical prototypes are evaluated by professional test operators) as well as in the earlier
stages like concept design (where only virtual prototypes are available).
The influence of the human operator is an aspect that is traditionally neglected in dy-
namic simulations conducted in concept design, because the modelling needs to be ex-
tended beyond the technical system. In the research presented in this thesis we show
two approaches to rule-based simulation models of a wheel loader operator. Both opera-
tor models interact with the machine model just as a human operator does with the ac-
tual machine. It is demonstrated that both operator models are able to adapt to basic
variations in workplace setup and machine capability. A “human element” can thus be
introduced into dynamic simulations of working machines, providing more relevant
answers with respect to operator-influenced complete-machine properties such as pro-
ductivity and energy efficiency.
While the influence of the human operator is traditionally ignored when evaluating
machine properties in the early stages of the product development process, later stages
are very reliant on professional test operators using physical prototypes. The presented
research demonstrates how the traditional subjective evaluation of a machine’s operabil-
ity can be complemented by a calculated measure for the operator’s control effort, de-
rived from the recorded control commands of the machine operator. This control effort
measure can also be calculated from the control commands of an operator model in a
simulation, such as those presented in this thesis. It thus also allows for an assessment
of operability already in the concept design phase.
In addition, the results of a study of quantifying operator workload by means of
measuring psychophysiological data such as heart rate variability and respiration rate
are presented as the first step towards realising workload-adaptive operator assistance
functions. Once fully developed, the method itself can also be used as another comple-
ment to the traditional subjective evaluations of operability. This approach can then be
applied not only in testing of physical prototypes, but also earlier in the product devel-
opment process in studies on human-in-the-loop simulators.
Sammanfattning
”Harmoniska arbetsmaskiner”
I ARBETSMASKINER spelar föraren en avgörande roll för maskinens prestanda. Såväl

produktivitet som energieffektivitet beror inte enbart av maskinens egenskaper och
arbetsomgivningen, utan beror också av sättet på vilket föraren manövrerar maskinen.
För att främja ett bränslesnålt körsätt ska maskinerna upplevas som harmoniska och det
är viktigt att beakta detta vid utvecklingen.
Det är nödvändigt att kvantifiera maskinharmonin och att ta hänsyn till interaktionen
mellan föraren och maskinen i alla steg av ett utvecklingsprojekt. Detta gäller såväl sena
faser, när fysiska prototyper redan har tagits fram och utvärderas av professionella
provförare, såväl som tidiga faser som konceptutveckling, när endast virtuella
prototyper finns tillgängliga.
Förarens inflytande beaktas traditionellt inte i prestandasimuleringar i konceptfasen,
eftersom detta innebär att mer än enbart det tekniska systemet måste modelleras. I den
forskningen som presenteras här visas två olika regelbaserade modeller av hjullastar-
förare. Båda förarmodellerna använder maskinmodellen på samma sätt som en verklig
förare använder en verklig maskin. Det visas att båda förarmodellerna kan anpassa sig
till förändringar både i arbetsomgivningen och i maskinens egenskaper. I och med detta
kan man utöka dynamiska simuleringar av arbetsmaskiner med ”ett mänskligt element”.
Detta ger bättre resultat vad gäller produktivitet, energieffektivitet och liknande
egenskaper som föraren påverkar i kompletta maskiner.
Medan man i tidiga faser av produktutvecklingsprocessen traditionellt bortser från
förarens inflytande, så är man i senare faser mycket beroende av att professionella
provförare testar fysiska prototyper. Den här presenterade forskningen visar hur den
traditionella subjektiva förarbedömningen av en maskins körbarhet kan kompletteras
med ett mått på förarens ”spakarbete”, som beräknas utifrån en mätning på hur föraren
använder sina kontroller för att styra maskinen. Detta mått på ”spakarbete” kan också
beräknas utifrån de spaksignaler som genereras av förarmodellerna i en simulering. I
och med detta kan en maskins körbarhet undersökas redan under konceptutvecklingen.
I avhandlingen redovisas också resultaten från en studie som gjorts i syfte att
kvantifiera förarens arbetsbelastning genom att mäta psykofysiologiska mått som
variationer i hjärtfrekvens och andningsfrekvens. Studien är ett första steg mot att
förverkliga en vision av stödfunktioner i arbetsmaskiner vilka anpassar sig efter förarens
momentana arbetsbelastning. En sådan metod att mäta förarens arbetsbelastning kan
också användas som ett komplement till den traditionella subjektiva förarbedömningen
av en maskins körbarhet. Metoden kan inte bara användas vid provning av fysiska
prototyper utan också tidigare i produktutvecklingsprocessen vid studier i avancerade
körsimulatorer.
Zusammenfassung
I N ARBETSMASCHINEN spielt der Fahrer eine entscheidende Rolle für die Leistung
des gesamten Systems. Produktivität und Energieeffizienz sind nicht nur abhängig von
den Grundeigenschaften der Maschine und den Bedingungen am Einsatzort, sondern
auch von der Art und Weise wie der Fahrer die Maschine manövreriert. Für eine
kraftstoffsparende Fahrweise muss der Fahrer die Maschine als harmonisch erleben.
Dies muss bei der Entwicklung beachtet werden.
Das Erfassen der Fahrbarkeit und die Berücksichtigung des Zusammenspiels
zwischen Fahrer und Maschine ist in allen Phasen der Entwicklung notwendig, sowohl
in den späteren Phasen, wenn Prototypen von Erprobungsfahrern ausgewertet werden,
als auch in den frühen Phasen wie dem Konzeptentwurf, wenn nur virtuelle Prototypen
vorhanden sind.
Der Fahrereinfluss wird traditionell in den dynamischen Simulationen während des
Konzeptentwurfs vernachlässigt, denn er erfordert die Ausweitung der Modellierung
über das technische System hinaus. In dieser Dissertation werden zwei Herangehens-
weisen zur Erstellung regelbasierter Modelle eines Radladerfahrers aufgezeigt. Beide
Fahrermodelle interagieren mit dem Maschinenmodell gleich einem menschlichen
Fahrer mit einer realen Maschine. Es wird gezeigt, dass beide Fahrermodelle in der
Lage sind, sich auf Änderungen des Einsatzortes und der Maschineneigenschaften
anzupassen. Somit kann „ein menschliches Element“ in die dynamische Simulation von
Arbeitsmaschinen eingeführt werden, was zu qualitativ besseren Resultaten bezüglich
Produktivität, Energieeffizienz und ähnlicher fahrerbeeinflusster Eigenschaften führt.
Während man in den frühen Phasen der Produktentwicklung traditionell vom Fahrer-
einfluss absieht, ist man später sehr auf die Erprobung physischer Prototypmaschinen
durch professionelle Testfahrer angewiesen. In dieser Dissertation wird aufgezeigt, wie
die traditionell subjektive Bewertung der Fahrbarkeit einer Maschine mit einem Maß
der „Steuerungsarbeit“ komplettiert werden kann, berechnet aus der gemessenen
Betätigung der dem Fahrer zur Verfügung stehenden Bedienelemente. Dieses Maß der
„Steuerungsarbeit“ kann auch aus den Signalen der von uns vorgestellten Fahrermodelle
in einer Simulation berechnet werden. Damit kann man die Fahrbarkeit bereits in der
Konzeptentwicklung abschätzen.
Weiterhin werden die Resultate einer Studie zur Quantifizierung der Fahrerbelastung
mithilfe psychophysiologischer Daten wie Veränderungen der Herzfrequenz und
Atmungsfrequenz vorgestellt. Diese Studie ist ein erster Schritt zur Entwicklung eines
Assistenzsystemes, dass sich an die aktuelle Fahrerbelastung anpasst. Eine solche
Messmethode der Fahrerbelastung kan auch zusätzlich zur traditionellen subjektiven
Fahrbarkeitseinschätzung angewendet werden – nicht nur bei der Erprobung physischer
Prototpyen, sondern auch schon frühzeitig bei Studien auf Fahrsimulatoren.
It's a great thing when you realize that you still have the
ability to surprise yourself. Makes you wonder what else
you can do, you have forgotten about...
(from the film “American Beauty”)

Acknowledgements
T HIS WORK, which is a result of many discussions and much practical work, was
undertaken at the Department for Research and Development at Volvo Construction
Equipment in Eskilstuna and at the Division of Fluid and Mechatronic Systems at
Linköping University. I would first like to thank my main advisor Jan-Ove Palmberg
and my academic co-advisors Kjell Ohlsson and Jonas Larsson, as well as my (maybe
unofficial but certainly crucial) industrial co-advisor Allan Ericsson for their very much
appreciated guidance and support with knowledge and wisdom, as well as reflections
and encouragement throughout the years, all of which proved really invaluable.
I would also like to thank Ulf Peterson and Bo von Schéele for their involvement in
my research projects as references and, like the people mentioned above, as active con-
tributors with ideas, visions, and sometimes deeply philosophical thoughts.
Many more have provided valuable insight, thoughts, ideas, and positive criticism,
mainly at Volvo and at Linköping University, but also elsewhere. In fact, there are so
many of you, whom I feel a deep obligation to thank, that I am afraid to even begin list-
ing names, because the space available on this page will not suffice. Instead, please ac-
cept my collective “Thank you!” and rest assured that if we know each other well
enough for having discussed about (but not necessarily agreed on) virtual & physical
testing, hydraulic and electric hybrids, autonomous machines, advanced machine con-
trol, human-machine interaction, development processes & innovation, knowledge
management, environmental & political issues, personal development & the meaning of
life, the flawed patent system, and the joy of our respective hobbies, then you have
made a much appreciated contribution to my journey.
Founding has over the years been provided by Volvo Construction Equipment, the
Swedish Program Board for Automotive Research (PFF), the Swedish Agency for Inno-
vation Systems (VINNOVA), the Swedish Energy Agency (Energimyndigheten), and
the Energy & Environment programme within the Swedish Vehicle-Strategic Research
and Innovation programme (FFI) – all of which is hereby gratefully acknowledged.
Finally, I would like to express my deepest thanks to my fiancée Maria for our inspir-
ing life outside work. As they say in the wonderful film The Bridges of Madison
County: “This kind of certainty comes but once in a lifetime.” And not only have we
helped each other to struggle along with our respective research, however far apart the
academic disciplines, it also just so happened that we both had to write our respective
theses at the same time – for which her family’s old-style summer house on Väddö,
some 1.5 hours north of Stockholm, was the perfect location. <3
Reno Filla
Väddö, July 2011
Papers
THE FOLLOWING SEVEN PAPERS are appended and will be referred to by their
Roman numerals. The papers are printed in their originally published or submitted state
except for some changes in formatting and the correction of minor errata.
[I] Filla, R. and Palmberg, J.-O. (2003) “Using Dynamic Simulation in the Devel-
opment of Construction Machinery”. The Eighth Scandinavian International
Conference on Fluid Power, Tampere, Finland, Vol. 1, pp 651-667.
http://www.arxiv.org/abs/cs.CE/0305036
[II] Filla, R., Ericsson, A. and Palmberg, J.-O. (2005) “Dynamic Simulation of Con-
struction Machinery: Towards an Operator Model”. IFPE 2005 Technical Con-
ference, Las Vegas (NV), USA, pp 429-438.
[III] Filla, R. (2005) “An Event-driven Operator Model for Dynamic Simulation of
Construction Machinery”. The Ninth Scandinavian International Conference on
Fluid Power, Linköping, Sweden.
[IV] Filla, R. (2009) “A Methodology for Modeling the Influence of Construction
Machinery Operators on Productivity and Fuel Consumption”. Proceedings of
HCII 2009: Digital Human Modeling, LNCS 5620, pp 614-623.
http://dx.doi.org/10.1007/978-3-642-02809-0_65
[V] Filla, R. (2009) “Hybrid Power Systems for Construction Machinery: Aspects of
System Design and Operability of Wheel Loaders”. Proceedings of ASME
IMECE 2009, Vol. 13, pp 611-620.
http://dx.doi.org/10.1115/IMECE2009-10458
[VI] Filla, R. (2011) “Study of a Method for Assessing Operability of Working Ma-
chines in Physical and Virtual Testing”. Submitted in July 2011 for publication in
International Journal of Vehicle Systems Modelling and Testing.
[VII] Filla, R., Olsson, E. M. G., von Schéele, B. H. C. and Ohlsson, K. (2011) “A
Case Study on Quantifying the Workload of Working Machine Operators by
Means of Psychophysiological Measurements”. Submitted in August 2011 for
publication in International Journal of Human Factors and Ergonomics.
The following papers are not included in the thesis, but constitute part of the back-
ground. Most of the main ideas expressed in these papers have been re-introduced in
later publications that are appended to this thesis.
[VIII] Filla, R. (2003) “Anläggningsmaskiner: Hydrauliksystem i multidomäna

miljöer”. Hydraulikdagar 2003, Linköping, Sweden.
http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-13371
[IX] Filla, R. (2005) “Operator and Machine Models for Dynamic Simulation of Con-
struction Machinery”. Licentiate thesis, Department of Management and Engi-
neering, Linköping University, Sweden.
[X] Filla, R. (2008) “Alternative system solutions for wheel loaders and other con-
struction equipment”. 1st International CTI Forum Alternative and Hybrid Drive
Trains, Berlin, Germany.
(Internet links refer either to the original papers or to Technical Reports based on them.
Verified on August 17, 2011)
Contents
1 Introduction and research background 1

1.1 Industrial objectives 2
1.2 Academic need 2
1.3 Research questions and hypotheses 3
1.4 Scope and delimitations 5
1.5 Research approach 7
1.6 Thesis outline 8
2 Operability of wheel loaders 9
2.1 General 9
2.2 Working cycles 11
2.2.1 Short loading cycle 11
2.2.2 Load & carry cycle 13
2.3 System architecture 14
2.4 Operability aspects 15
2.4.1 Harmonic working machines 15
2.4.2 Requirements 16
2.4.3 General operability aspects in short loading cycles 17
2.4.4 Operability aspects in bucket filling 21
2.4.5 Operability aspects in reversing 25
2.4.6 Operability aspects of bucket emptying 26
2.5 Summary 28
3 Simulation models 29
3.1 State of the art 29
3.1.1 General 29
3.1.2 Machine models 30
3.1.3 Environment models 30
3.1.4 The need for driver/pilot/operator models 31
3.1.5 Pilot models 33
3.1.6 Driver models 34
3.1.7 Cognitive architectures 37
3.1.8 Operator models 38
3.1.9 Autonomous excavation 41
3.2 Models developed in our research 43
3.2.1 Machine and environment model 43
3.2.2 Methodology for developing operator models 43
3.2.3 Operator model #1 45
3.2.4 Operator model #2 46
3.2.5 Summary and discussion 49
4 Operator workload 51
4.1 Literature research 51
4.1.1 Self-report measures 52
4.1.2 Psychophysiological measures 52
4.1.3 Simulation approaches 55
4.2 Empirical study #1 56
4.3 Empirical study #2 57
4.3.1 Setup 57
4.3.2 General results 60
4.3.3 Discussion of possible measures for operability 61
4.3.4 Control effort 62
4.3.5 Psychophysiological measurements 65
4.3.6 Summary 68
5 Adaptive aid 71
5.1 General 71
5.2 Implementation of adaptive aid 72
5.3 Applications for adaptive aid in working machines 73
5.4 Real-time assessment of operator workload 75
6 Review of appended papers 77
Paper I 78
Paper II 78
Paper III 79
Paper IV 79
Paper V 79
Paper VI 80
Paper VII 80
7 Scientific contributions 83
8 Conclusions and outlook 87
References 89
Appended papers
I Using Dynamic Simulation in the Development of 101
Construction Machinery
II Dynamic Simulation of Construction Machinery: 121
Towards an Operator Model
III An Event-driven Operator Model 137
for Dynamic Simulation of Construction Machinery
IV A Methodology for Modeling the Influence of Construction 157
Machinery Operators on Productivity and Fuel Consumption
V Hybrid Power Systems for Construction Machinery: 171
Aspects of System Design and Operability of Wheel Loaders
VI Study of a Method for Assessing Operability of 193
Working Machines in Physical and Virtual Testing
VII A Case Study on Quantifying the Workload of 225
Working Machine Operators by Means of
Psychophysiological Measurements
1
Introduction and
research background
W ORKING MACHINES in construction, mining, agriculture, and forestry have re-

placed the traditional tools of trade (Figure 1). With the help of these machines humans
were able to reduce their physical workload – or, as Hollnagel and Woods put it in [1],
the human operators of working machines can amplify their physical abilities. But oper-
ating these machines still requires much effort (mentally, as well as physically) and ex-
perience, as they are becoming ever more sophisticated in terms of digital control.
Figure 1. Construction tools, past and present
Many working machines are also complex in architecture. They consist of at least
two working systems that are used simultaneously and the human operators are essential
to the performance of the machines in their working place. Their productivity and en-
ergy efficiency is linked to the human operators’ workload; and it is therefore important
to make working machines easier to use.
2 Quantifying Operability of Working Machines
1.1 Industrial objectives

The focus on a machine’s operability, i.e. “the ease with which a system operator can
perform the assigned mission with a system when that system is functioning as de-
signed” [2], has to be established early in the product development process. As with all
system properties that arise out of the complex interplay between components and sub-
systems within a machine, freedom of design is greatest in the beginning, in the concept
development phase. This is also were the potential impact on both development and
product cost is greatest, while the actual cost of exploring various scenarios is lowest. It
is therefore necessary to improve our methods so that we can evaluate the operability of
a working machine and operator-influenced properties like productivity and energy effi-
ciency early in a development project. The tool of choice here is simulation.
In addition to this, we also need to be able to better assess operability later in the
process, where traditionally a physical prototype’s operability is subjectively evaluated
by asking the test operator to pass a judgement, sometimes involving rating the machine
according to a specific scale. Here the idea is to complement these subjective evalua-
tions with a less subjective measure based on the operator’s workload, which according
to one definition is “the portion of the operator’s limited capacity actually required to
perform a particular task” [3].
There are also a great many opportunities to improve operability due to extensive
computer control of the machine and its sub-systems. New human-machine interfaces
will include augmented cognition, but also new ways in which a working machine is
controlled by the operator. One obvious idea is to automate repeatedly used sequences
of actions and movements that are manually controlled today. Another promising idea is
to make the whole system adaptive to the operator’s workload and offer assistance in
critical situations. The verification of the effectiveness of such advanced systems again
needs a more precise method than is presently available. In the case of workload-
adaptive functions, the operator’s workload also needs to be determined on-line and
close to real time in order to be available as an input to the assistive system.
1.2 Academic need

The original industrial objectives leading to this thesis revolved around wheel loaders.
Taking the position that research is either the creation of new knowledge, or the applica-
tion of elsewhere existing knowledge in a new area, or even the combination of existing
area knowledge with new insights gained as a result, the needs formulated in the previ-
ous section are already research problems, since they are not restricted to wheel loaders
(which would be a specific application) but formulated for working machines in gen-
eral.
But we can also widen our perspective a little further. A great many research results
can be found concerning the performance of complex technical systems and ways to
automate work tasks. More research is also being done in the field of human factors and
mental workload. However, with working machines we have a human operator who is
Introduction and research background 3
not merely a supervisor with mostly cognitive tasks to perform, but is an essential part
of the total system.
This is of course also true of on-road vehicles and aircrafts, at least the ones where
the human pilot still plays a role. A quote attributed to Arlen Rens, test pilot for Lock-
heed Martin, comes to mind: “Airplanes are now built to carry a pilot and a dog in the
cockpit. The pilot's job is to feed the dog, and the dog's job is to bite the pilot if he
touches anything.”
But while there is an abundance of research into driver models for vehicles, as well
as pilot models for aircraft (especially military ones where the performance of the total
system is often limited only by the human pilot), little documentation exists on research
into operator models for working machines. Though similarities exist in certain re-
spects, there is no simple general analogy that can be made, no ready solutions that can
easily be transferred. The tasks required of a human driver to control a car on a road and
a human pilot to control a fighter plane in air combat are significantly different to the
tasks required of a human operator using a wheel loader, a tree harvester, a tractor or an
LHD-loader (Load-Haul-Dump) in a working place.
There is therefore a general need to describe the influence of the human operator on
total system performance in the area of working machines. Specifically, there is a need
to derive methods to model the human-machine interaction and evaluate the operability
of working machines in order to guide product development in all phases of the devel-
opment process.
The term operator workload has already been used and it has been mentioned that the
human operator is part of the total system. There is a need to see working machines
from this perspective, which will be done in this thesis with a wheel loader as an exam-
ple. It can be argued that this aspect is an industrial problem rather than an academic
one, but it is tied to quantifying operator workload, which in turn certainly is a topic of
ongoing academic research.
1.3 Research questions and hypotheses

Condensing the problems stated in the previous sections, we can formulate one major
research question, followed by the main hypothesis:
RQ. How to quantify operability of working machines in all phases of the de-
velopment process?
H. We can assess operability comparatively by comparing calculated or meas-
ured operator workload for different machines. Higher workload means
lower operability and vice versa. Which method of quantifying operator
workload is best will vary depending on the current phase of the product
development process.
Guided by the problems formulated earlier we arrive at the following sub-questions,
each accompanied by a hypothesis that shaped this part of the research. The order in
which they are listed here does not necessarily reflect the order in which the questions
were dealt with.
RQ1. How to simulate productivity, energy efficiency and similar operator-
influenced properties of working machines, where the human operator is
essential for the performance of the total system?
H1. By extending the simulation models to not only cover the necessary ma-
chine sub-systems, but also include models of the working environment and
the human operator, qualitatively better answers can be found regarding the
above mentioned complex product properties. Initially there is no need to
model the human operator in detail; sufficiently good results can be ob-
tained by implementing principal operator decision models and strategies.
Considerable effort was needed to answer the next question, which is why it is not part
of the previous one:
RQ2. How can operator models be used to also evaluate a machine’s operability
in dynamic simulations?
H2. The way the human operator uses the controls at his/her disposal is indica-
tive of the operator’s workload and thus of the machine’s operability. Using
recorded data from real working cycles, a measure of the operator’s control
effort can be found that correlates well with the operator’s subjective as-
sessment of the operability of the machine in use. This measure can also be
calculated from the command signals generated by an operator model in a
dynamic simulation.
It can be argued whether the next question should rather be a sub-question of RQ1 or if
it is complex enough to stand alone:
RQ3. What methodology to use to derive operator models for dynamic simulation
that adaptively model human-machine interaction in working machines in a
transparent and flexible way?
H3. Adaptive (in the sense of the operator model itself being able to adapt to
basic variation in the scenario), flexible (in the sense of easy to manually
adapt to a new scenario), and transparent (in the sense of easy to understand
and validate) operator models can be derived by avoiding the lure of path
tracking (less adaptive and not flexible) and artificial neural networks (less
flexible and not transparent), and instead describing the actions of the op-
erator model in a generic, rule-based way, using rules extracted from inter-
view studies with professional operators.
Simulation is the tool of choice in the concept development phase. The next question
moves on to later stages:
RQ4. How can operability and operator workload be assessed in later phases of
the product development process, which today relies on subjective evalua-
tions of physical prototypes by professional test operators?
H4. One way is to use the solution proposed in hypothesis H2, i.e. the measure
of the operator’s control effort, calculated from recorded control command
signals. Another way is the measurement of psychophysiological data such
as heart rate, heart rate variability, finger temperature and respiratory rate
(as well as several others), which is an established way to assess mental
workload of, for example, fighter pilots during flight missions. The work-
load of an operator in a working machine is not limited to the cognitive
domain, but neither is the workload of a fighter pilot. Thus, the above men-
tioned parameters, together with others, should be possible to use. In the fu-
ture, testing of physical prototypes will also be complemented by testing on
human-in-the-loop simulators, but the same method of quantifying oper-
ability and operator workload can be used.
Judging from the title of this thesis the next question goes beyond the research scope,
but has been dealt with prior to RQ1 and leads on to RQ6:
RQ5. How is operability affected by system architecture and are there ways to
minimise operator workload and thus enhance operability?
H5. A good analogy is a pyramid with the operator at the apex and the ma-
chine’s hardware as the base. Traditionally, the choice of hardware compo-
nents (the base of the pyramid) determined operability of a working ma-
chine and thus the operator’s workload (the pyramid’s apex). There are two
layers between base and apex: software and automation. With the introduc-
tion of electronic control systems, software could be used to enhance the in-
teraction between sub-systems and thus the way the machine reacted to the
operator’s commands. Nowadays there are also a great many opportunities
in new human-machine interfaces, workload-adaptive operator assistance
functions, and for automation of repeatedly used sequences of actions and
movements that are today manually controlled by the operator.
The last part of hypothesis H5 leads to RQ6:
RQ6. How can operator workload be quantified such that workload-adaptive op-
erator assistance functions can be realised?
H6. Provided techniques exist for measuring with little or no artefacts and
methods for evaluating the measurements on-line and in real time, the
measurement of psychophysiological data proposed in H5, e.g. heart rate,
heart rate variability, finger temperature and respiratory rate (as well as
several others) will meet the requirements.
1.4 Scope and delimitations

The research questions formulated in the previous section cover quite a broad area.
Even though the author has worked for a little more than nine years on the research pre-
sented in this thesis, not every aspect has been dealt with equally much; and some of
what has been worked on can not be made public at this point in time.
This research was originally initiated by the Wheel Loader division of Volvo Con-
struction Equipment. It is therefore natural that the presented figures and examples con-
cern wheel loaders, even though the scope of this research encompasses all working
machines of similar complexity. These can be found not only in construction but also
agriculture, forestry, mining, and others. Common factors are that these machines con-
sist of at least two working systems that are used simultaneously and that the human
operator is essential to the performance of the total system.
The research presented in this thesis goes across boundaries and covers several tradi-
tional research areas. It has been performed at the division of Fluid and Mechatronic
Systems (which has a strong emphasis on advanced hydraulics and their control), and
was to begin with focused on evaluating complete machine systems by dynamic simula-
tion (with an emphasis on productivity, energy efficiency and operability). The focus
soon shifted to operator models, though without being concerned with vehicle dynam-
ics, but rather machine-internal distribution of power (which has its home in the re-
search area of Machine Design). The latter interest led to more than just theoretical
musings on energy efficiency and hybrids – and how such advanced machine systems
could also improve the operability of working machines. This again required an answer
to the question of how to measure operability; and the increased interest in human-
machine interaction in general finally tipped the balance in favour of the research area
of Human Factors / Cognitive Systems Engineering (CSE) with a dash of Psychophysi-
ology. Experts in all these areas will surely find that this thesis and the appended papers
lack several of the usual “must have” references. This is not due to ignorance or dis-
agreement, but is rather a necessary consequence of limited capacity on the part of the
author – which brings us back to workload.
Mental and physical workload, human-machine interaction and cognitive processes
have only been researched to such an extent that the research questions RQ1-RQ6 could
be answered. We do not claim to have performed research in the area of Human Factors
or CSE, we rather consider our horizon to have been a somewhat extended engineering
one (Figure 2).
Figure 2. Research area of this thesis

Throughout the research, the terms operability and operator workload have been
used as described in the main hypothesis: higher workload (both the mental and physi-
cal workload of the operator) means lower operability, and vice versa. We exclude op-
erator comfort with aspects like exposure to vibration, ergonomics etc, and instead con-
centrate on the part of the workload of the operator, which is affected by the operator’s
cognitive and control efforts.
In all theoretical and practical work the technical system is considered to work as de-
signed, which excludes the effects of component or system failures on the machine.
Later on, methods such as engineering design optimisation and probabilistic design
(analytical or experimental) might be applied in order to judge the robustness of a cho-
sen property (such as productivity, energy efficiency or operability) but these have been
neither researched nor used in the work covered in this thesis.
Co-simulation of models from various technical domains has been performed, but
has not been at the focus of this research. The simulation tools used have been enhanced
by subroutines written by the author when necessary, but this has not been a research
goal and no research has been done into computational algorithms.
Paper [VI] presents an algorithm for piecewise linear approximation of a curve. This
algorithm is presented “as is” without proof of optimality (or even any claim of novelty,
even though a literature survey did not produce any sign of it having existed before).
The psychophysiological measurements presented in paper [VII] were reviewed by
three experts in the field, who compared their individual assessments and agreed on a
consensus value for workload of respective operator in respective condition. The au-
thor’s contribution in this regard consisted of everything else, i.e. planning and conduct-
ing the study, data acquisition, pre-processing, post-processing, drawing of conclusions
and writing the paper.
1.5 Research approach

As in all research, the study of relevant literature has had an important part to play, even
though, as warned earlier, experts in all touched-upon research areas will surely find
this thesis and the appended papers lacking of possibly crucial references.
At least equally important has been the study of available knowledge and experience
at the industrial partner, which was many times found to be undocumented and thus
only accessible via personal interviews. The author has participated in several develop-
ment projects, in some of them in a more leading role, which gave the opportunity to
broaden and document knowledge about conceptual design of working machines. In this
aspect, the research approach has had a certain descriptive component.
Comprehensive test series of physical prototypes and production machines were
specified, conducted/supervised, and analysed. In addition to this practical work, theo-
retical calculations and simulations have been performed.
1.6 Thesis outline

The following chapters in this thesis will give the reader an overview of the state of the
art and achieved results in the four main areas of our research:
- Operability aspects of wheel loaders as complex systems
(representative for all similar working machines)
- Simulation models
- Operator workload
- Adaptive aid
(which, even though a little beyond the original scope, provides motivation
for part of our later research)
These chapters are followed by a summary of the scientific contributions of our re-
search, and then discussion, conclusions, and some suggestions for future work. For
more thorough descriptions of our results, the reader is referred to the seven appended
papers, of which summaries are provided towards the end of the thesis.
2
Operability of
wheel loaders
WHEEL LOADERS are complex working machines. The interconnections of sub-

systems of various technical domains (mechanics, hydraulics, controls...) cause a non-
trivial behaviour of the total system, of which the human operator is a part.
This chapter will focus on these aspects of wheel loaders as a background to our re-
search into operability of working machines in general. The following text can be seen
as the result of the descriptive part of our research approach. It might occasionally ap-
pear rather detailed, but in successive chapters we will refer to many of the facts and
phenomena described in the following.
2.1 General
Wheel loaders are highly versatile machines and are built in vastly varying sizes. The
smallest have an operating weight of just 2 tons while the world’s largest wheel loader
weighs more than 260 tons. Naturally, not only are there technical differences between
these extremes, but also differences in their application. Smaller wheel loaders are often
built with a hydrostatic drive train and are used for all kinds of service jobs at smaller
construction sites, farms, etc. A large number of attachments exist, for example buckets,
grapples, forks, material handling arms and cutting aggregates to mention just a few.
Machines of this size are purchased for their immense versatility. The opposite is true of
larger wheel loaders, which are often bought for one single application in a quarry or
similar setting. The largest wheel loaders feature a diesel-electric drive train with an
electric motor and a planetary gear in each wheel hub.
The research presented in this thesis focuses on medium to large-sized wheel loaders.
Figure 3 shows a Volvo L180E with a nominal operating weight of 26 to 29 tons. Such
wheel loaders are usually assigned to one specific application as part of a production
chain. Several special attachments exist; there even is a machine variant with a high-lift
loading unit for timber loading.
Figure 3. Wheel loader [4]
Due to their versatility, wheel loaders need to fulfil a great many requirements, which
are often interconnected and sometimes contradictory. Leaving aspects such as total cost
of ownership, availability, legislation compliance, etc. aside, the most important ma-
chine properties are
- Productivity (expressed for example in ton/hour)

- Fuel efficiency (expressed for example in ton/litre) or better
Energy efficiency (expressed for example in ton/kWh)
- Operability
Varying with the working task, other important parameters are:
- Geometric parameters
(e.g. lift height, digging depth, dump reach, parallel alignment)
- Loads, torques and forces
(e.g. tipping load, break-out torque, lift force, traction force)
- Speeds and cycle times (complete machine and sub-systems)
- Controllability (precision, feedback, response)
It has to be pointed out that while some of the items are clearly determined by more
than just one sub-system (e.g. lift force, which is determined by hydraulics and loading
unit), others seem to be possible to attribute to one single sub-system (e.g. traction
force). One might thus, wrongly, be tempted to leave such aspects out of the optimisa-
tion loop when it comes to trading off product targets against each other when choosing
technical solutions. As stated before, everything is connected and interdependent – and
there are therefore seldom any simple solutions to be found. Paper [I] gives a more de-
tailed break-down of the design process and proposes some adjustments regarding how
to integrate dynamic simulation.
Operability of wheel loaders 11
2.2 Working cycles

Wheel loaders are versatile machines, as stated before, and essentially each working
place is unique, yet common features can nonetheless be found.
The knowledge regarding wheel loader operation and working cycles presented in the
following has been derived from the author’s own wheel loader experience and to a
large extent from discussions with colleagues, test engineers, product specialists, and
professional operators at Volvo. Most of these discussions were not formally structured,
but rather conducted in an ad-hoc manner. However, some unstructured research inter-
views were carried out and one interview, with a professional test operator, was re-
corded in the form of a semi-structured research interview. In addition, many measure-
ments were performed and the results and implications discussed. Most of these reports
are internal, but some Master theses [5][6][7][8] and academic papers [VIII] are avail-
able in the public domain.
In [9], Gellerstedt published wheel loader operators’ thoughts and reasoning and also
documented some typical working cycles with photos and test data. Non-academic pub-
lications like operating manuals and instruction material available from machine manu-
facturers also contain useful information.
2.2.1 Short loading cycle
This cycle, sometimes also dubbed V-cycle or Y-cycle for its characteristic driving pat-
tern, is highly representative of the majority of applications.
Figure 4. Short loading cycle

Typical for this cycle is bucket loading of some kind of granular material (e.g. gravel,
sand, or wood chips) on an adjacent load receiver (e.g. a dump truck, conveyor belt, or
stone crusher) within a rather tight time frame of 25-35 seconds. For this reason, the
short loading cycle has been established as the main test cycle for operability during
development of wheel loaders (to be discussed in more detail later).
As can be seen in Figure 4, several phases can be identified in such a loading cycle.
Table 1 gives a brief description of these.
Table 1. Phases of the short loading cycle
# Phase Description
1 Bucket filling Bucket is filled by simultaneously controlling the
machine speed and lift and tilt functions.
2 Leaving bank Operator drives backwards towards the reversing
point and steers the machine to achieve the charac-
teristic V-pattern.
3 Retardation Is started some time before phase 4 and can be
either prolonged or shortened by controlling the
gas pedal and the service brakes.
4 Reversing Begins when the remaining distance to the load
receiver will be sufficient for the lift hydraulics to
achieve the bucket height necessary for emptying
during the time it takes to get there.
5 Towards The operator steers towards the load receiver, thus
load receiver completing the V-pattern. The machine arrives
perpendicular to the load receiver.
6 Bucket emptying The machine is driven forward slowly, the loading
unit being raised and the bucket tilted forward at
the same time.
7 Leaving Operator drives backwards towards the reversing
load receiver point, while the bucket is lowered to a position
suitable for driving.
8 Retardation Not necessarily executed at the same location as in
and reversing phases 3 and 4, because lowering an empty bucket
is faster than raising a full one.
9 Towards bank The machine is driven forward to the location
where the next bucket filling is to be performed,
the bucket being lowered and aligned with the
ground at the same time.
10 Retardation Often combined with the next bucket filling by
at bank using the machine’s momentum to drive the
bucket into the gravel pile.
Actually, eleven phases could be identified, if phase 8 were divided in two, compara-
ble to phases 3 and 4. However, the latter have been defined as separate phases, because
unlike phase 8 they reveal critical aspects of a wheel loader’s design and operability.
In developing this definition of the short loading cycle, the focus has been on testing
physical prototypes and giving guidance to both operators and test engineers. The start
and end of each of these phases have been defined so that they can easily be identified
in measurements. We have found it practical to use the same definition in our work with
simulation and operator models.
However, in other circumstances it might be meaningful to separate or combine sev-
eral phases, as for example in [5] (six phases), [8] (where an attempt is made to consoli-
date different definitions in order to standardise) and [10] (four phases, one per leg).
It can also be mentioned that the load receiver does not necessarily have to stand per-
pendicular to the place of bucket filling. Both [8] and [10] show slightly different vari-
ants with the advantage of less steering involved for the wheel loader operator.
2.2.2 Load & carry cycle
Load & carry cycles, sometimes also called long loading cycles, are also widely used.
They resemble short loading cycles, but involve two transport phases of up to 400 m in
forward gear (Figure 5).
Figure 5. Load & carry cycle
The definition of several phases (Table 2) is similar to the short loading cycle. Here
too, some phases can be combined as done for example in [5], where only four phases
have been defined.
It is also quite common in load & carry cycles for the load receiver to be a conveyor
belt or a stone crusher, possibly even standing elevated so that for the last metres the
wheel loader has to climb a steep inclination (which presents a special challenge in
terms of machine performance and operability).
Table 2. Phases of the load & carry cycle
# Phase Description
1 Bucket filling (see short loading cycle)
2 Leaving bank (…)
3 Retardation (…)
4 Reversing (…)
5-7 Towards The machine is driven forward at as high a
load receiver speed as inclination and stability allow. The
bucket is in carry position.
8 Towards load The machine’s speed is gradually reduced as it
receiver + lifting comes closer to the load receiver, the loading
unit being raised at the same time.
9 Bucket emptying The machine is driven forward slowly, the
loading unit being raised even higher and the
bucket tilted forward at the same time.
10 Leaving load Operator drives backwards and then reverses,
receiver + reversing while the bucket is lowered to carry position.
11-13 Return to bank The machine is driven forward at as high a
speed as inclination and stability allow. At the
end of phase 13, the bucket is aligned with the
ground to prepare for the next filling.
2.3 System architecture

A modern wheel loader consists of tightly coupled, non-linear sub-systems that interact
even in seemingly simple cases. Medium to large sized wheel loaders are usually
equipped with a conventional drive train that includes a hydrodynamic torque converter
and an automatic power-shift transmission. Many modern wheel loaders feature load-
sensing hydraulics with variable displacement pumps. The linkage of the loading unit
acts as a non-linear transmission between hydraulic cylinders and attachment. In addi-
tion to auxiliary systems such as cooling systems, there may also be external systems
connected via PTO’s (power take-offs).
Lifting +
Hydraulics Linkage Bucket
Breaking/Tilting
Auxiliaries
Gravel
Σ Engine
pile
External (+/-)
Travelling/
Drive train Wheels
Penetration
Figure 6. Simplified power transfer scheme of a wheel loader during bucket loading
The drive train and hydraulics in a wheel loader are both equally powerful sub-
systems and compete for the limited engine torque. Figure 6 shows how in the case of
bucket loading the primary power from the diesel engine is essentially split up between
hydraulics and drive train to create lift/tilt movements of the bucket and deliver traction
to the wheels. This presents an operability challenge, which will be discussed in the
following.
2.4 Operability aspects

2.4.1 Harmonic working machines
While there are usable definitions for productivity (for a wheel loader, material loaded
per unit of time) and energy efficiency (preferably material loaded per required unit of
energy/fuel, rather than fuel consumption per time unit), a generally agreed definition of
the human operator’s difficulty in working with the machine has been lacking. For vehi-
cles (where the drive train is the only major power-consuming system), the concept of
drivability has been the main focus for quite some time. In working machines, however,
drive train and hydraulics constitute two parallel and equally powerful sub-systems. We
thus are concerned with operability instead.
[1] offers a definition that also works well for working machines: “Operability is the
ease with which a system operator can perform the assigned mission with a system
when that system is functioning as designed”. The limitation to states where the system
is functioning as designed effectively excludes somewhat related yet separate properties
such as robustness and reliability.
The challenge in designing a wheel loader, from a manufacturer’s point of view, is to
find an appropriate, robust, and maintainable balance between productivity, energy effi-
ciency, and operability over the whole area of use. At Volvo, the term harmonic wheel
loader has been coined by some engineers to describe an ideal machine, possessing a
high degree of machine harmony which makes it intuitively controllable and able to
perform the work task in a straightforward manner without much conscious thought or
strategy. The lower the degree of machine harmony, the more effort the operator has to
put in in order to perform the work task and thus the higher the workload (Figure 7).
Operator
workload
y
Secondary harmony
on
rm
(added control systems)

ha
e
hin
ac
M
Primary harmony
(matched and appropriately specified hardware)
Figure 7. Machine harmony in a working machine and operator workload

While in the past only a certain level of machine harmony could be achieved by a
delicate combination of carefully specified hardware (and leaving the rest to the opera-
tor), working machines are nowadays equipped with at least one electronic control unit
(ECU), often several, each dedicated to one specific component or sub-system, and all
ECUs connected to a data bus. “X by wire” is also a trend in the off-road equipment
industry, which implies that operators communicate their wishes to the machine via
electronics, rather than controlling it directly. As shown in Figure 8, this opens new
possibilities to reduce operator workload by controlling components and sub-systems in
context, rather than stand-alone.
Op.
workl.
Operator
workload Automation
Operator
y
on
workload
rm
ha
y
on
Secondary harmony Secondary harmony
e
rm
hin
(added control systems) (added control systems)
ha
ac
e
M
hi n
ac
M
Machine harmony Primary harmony Primary harmony

(matched and appropriately specified hardware) (matched and appropriately specified hardware) (matched and appropriately specified hardware)
Figure 8. The evolution of the harmonic working machine: past, present and future
In the future we will see an increasing number of automatic functions, which will re-
duce operator workload even further. Already today there are some examples, for in-
stance automatic return of the bucket to a previously defined angle and height. Automa-
tion of bucket filling has also been shown to work satisfactory in certain conditions.
This is just the beginning of an exciting development towards automation.
2.4.2 Requirements
In general, the assumption made in this research is that the operator’s impression of the
working machine’s operability stems from the amount of workload the operator is sub-
jected to. We exclude operator comfort with aspects like exposure to vibration, ergo-
nomics etc, and instead concentrate on the part of the operator’s workload, which is
affected by his or her cognitive and control efforts. The following requirements can be
found (the fourth probably being the most prominent):
- fast response of each function without perceivable lack of power
- high precision in machine and bucket positioning
- several operator controls to be actuated simultaneously in order to achieve a
specific effect
- several machine functions to be synchronized.
In a short loading cycle as shown in Figure 4, as well as in a load & carry cycle ac-
cording to Figure 5, there are three phases in which the operator is challenged to a
higher degree:
- Bucket filling (phase #1)

- Reversing (phase #4)
- Bucket emptying (phase #6 in Figure 4, #9 in Figure 5)
In addition to this, the operator needs to maintain a certain balance of motions while
also being aware of the distribution of engine power, as mentioned previously. Each of
these aspects will be examined in more detail with regard to operability and operator
workload in the following sections.
2.4.3 General operability aspects in short loading cycles

After bucket filling (phase 1 in Figure 4), the operator drives backwards towards the
reversing point and steers the wheel loader to achieve the characteristic V-pattern of a
short loading cycle. The lifting function is engaged the whole time. The operator
chooses the reversing point such that having arrived at the load receiver and starting to
empty the bucket (phase 6 in Figure 4), the lifting height will be sufficient to do so
without delay. In the remaining phases, the bucket needs to be lowered and the operator
steers the wheel loader back to the initial position in order fill the bucket again in the
next cycle.
This pattern in a short loading cycle is comprehensibly visualised in Figure 9, which
shows a graph representing bucket height over integrated machine speed (i.e. the ma-
chine’s travelling distance, provided no tyre-slip occurred).
Figure 9. Machine harmony diagram with identified phases of a short loading cycle
The ratio of lift speed to machine speed, which is crucial to how long the loader
needs to be driven backwards until reversing, can here be seen as the graph’s slope. The
graph is fairly straight from the beginning of phase 2 (where the loader leaves the bank)
to the end of phase 5 (where the loader arrives at the load receiver). This indicates that
the operator judged well when to reverse. Otherwise, the slope would have been steeper
at the end of phase 5, indicating that the operator needed to slow down or even stop in
order for the bucket to reach sufficient height for emptying.
The distance between the point of reversing and the bank (or load receiver) is an in-
dication of how good a balance has been achieved between the two motions of lifting
and driving. Introduced in late 2002, engineers at Volvo call this a Machine harmony
diagram, because it visualises aspects that are important for a machine’s operability.
Another interesting aspect is the visualisation of the operator’s style of bucket filling
and bucket emptying. The chosen bucket filling style also indirectly influences the posi-
tion where the loader needs to reverse: Leaving the bank at the beginning of phase 2
with a higher bucket means that it will take less time to raise it from there to a sufficient
height for emptying. This means that in theory the point of reversing will be nearer to
the bank when leaving with a higher bucket position, because the operator adapts to the
machine. It can be shown that this is also the case in practice:
Figure 10 shows three selected short loading cycles, originating from a non-stop re-
cording, during which the operator changed his bucket filling style. As can be seen, the
bucket height when leaving the bank varies, but the operator succeeded very well in
choosing a reversing point in order to reach the load receiver with the bucket at a suffi-
cient height for emptying.
Together with time plots, such Machine harmony diagrams are a useful tool for
evaluating how good a motion balance has been achieved in the machine (in a proper
short loading cycle revealed by the start of reversing, i.e. phase 4). To a certain extent
they also show the human operator’s operating style.
Figure 10. Recorded short loading cycles with different bucket filling styles
Another ingredient of machine harmony is the availability of power. Here, of course,

we have a delicate balance with energy efficiency to achieve. In a short loading cycle,
phase 1 (bucket filling) and phase 4 (reversing) are the most challenging from an en-
gine’s point of view. Both drive train and hydraulics have high power demands. High
traction, lifting, and break-out forces are needed during bucket filling, while during re-
versing, the engine is run at a lower speed and needs to simultaneously cope with in-
creased load from the torque converter and with hydraulic power demand due to contin-
ued lifting. The operator’s impression of the wheel loader as a powerful machine
depends on either the engine’s ability to satisfy all power demands, the overall machine
control system’s ability to govern each sub-system, or a combination of the two.
The diagram in Figure 11 displays data recorded during a short loading cycle. The
distribution of engine power to hydraulics and drive train seems on average to be ap-
proximately equal, but with large momentary deviations from that rule (as Figure 12
shows). Most engine power is required during bucket filling (the first phase in Figure 11
and solid grey dots in Figure 12). The oscillations in power and in distribution ratio are
a reflection of the operator’s bucket filling style and the conditions at the work place.
Figure 11. Power distribution to hydraulics and drive train in a short loading cycle
(total engine power = top of the black area)
Figure 12. Power ratio of hydraulic power to drive train power (solid grey = bucket filling)
As Figure 12 shows, on average more than half of the engine power was transferred
to hydraulics in the short loading cycle displayed. This ratio varies with working task
and operator, but Figure 11 and Figure 12 give a good indication of the general dynamic
situation within the system. Figure 13 shows the corresponding engine diagram, which
arguably reveals a more dynamic situation than for any on-road vehicle’s engine.
Figure 13. Typical engine load duty in a short loading cycle (solid grey = bucket filling)
Again, the solid grey dots show that the highest power requirement is in the bucket
filling phase. Figure 14 shows that the fuel consumption rate is also highest in this phase
of a short loading cycle. For phases 2 to 6 (including bucket emptying) the fuel con-
sumption rate is close to the mean fuel consumption rate of the complete cycle. The
remaining phases 7-10 are less energy-demanding and the fuel consumption rate is
therefore lower than the average for the complete short loading cycle.
Figure 14. Fuel consumption during short loading cycle

When it was stated earlier that the engine in a wheel loader has a fairly dynamic
situation to deal with compared to its counterpart in an on-road vehicle, this is also valid
for the human operator. Figure 15 shows the operator’s use of gas pedal, lift and tilt
lever, steering joystick and brake pedal during a calm short loading cycle of about
33 seconds (the diagram shows a section from a continuous recording).
Figure 15. Operator control commands in a calm short loading cycle
In the following we will examine some prominent loading cycle phases in more de-
tail with regard to operability and operator workload.
2.4.4 Operability aspects in bucket filling

Extending the schematics of the wheel loader’s system architecture (previously visual-
ised in Figure 6), Figure 16 below shows how the human operator interacts with the
machine during bucket filling.
Lifting +
Breaking/Tilting
ECU
Auxiliaries
Gravel
Σ Engine ECU ECU Operator
pile
External (+/-)
ECU
Travelling/
Drive train Wheels
Penetration
Figure 16. Power transfer and control scheme of a wheel loader during bucket loading
In order to fill the bucket, the operator needs to control three motions simultaneously:
a forward motion of the complete machine that also exerts a force (traction), an upward
motion of the bucket (lift) and a rotating motion of the bucket to fit in as much material
as possible (tilt). This is similar to how a simple manual shovel is used. However, in
contrast to a manual shovel, the operator of a wheel loader can only observe, and cannot
directly control these three motions. Instead, he or she has to use different subsystems
of the machine in order to accomplish the task. The gas pedal controls engine speed,
while lift and tilt lever control valves in the hydraulics system that ultimately control
movements of the linkage’s lift and tilt cylinders, respectively.
One difficulty is that no operator control directly affects only one single motion. The
gas pedal controls engine speed, which together with the selected transmission gear af-
fects the machine’s longitudinal motion. But engine speed also relates to the speed of
the hydraulic pumps, which in turn affects the lift and tilt cylinder speed. The linkage
between the hydraulic cylinders and the bucket acts as a non-linear planar transmission
and due to its design a lift movement will also change the bucket’s tilt angle and a tilt
movement affects the bucket edge’s height above the ground. Furthermore, when simul-
taneously using the lift and tilt functions, a reduction in the lift cylinder speed can be
expected as the displacement of the hydraulic pump is limited and the tilt function is
given priority over lift due to its lower pressure demand. The more the tilt lever is actu-
ated, the higher the tilt cylinder speed, while the speed of the lift cylinder decreases.
Finally, there is also a strong interaction between traction and lift/tilt forces: penetrat-
ing the gravel pile with the bucket requires traction force, which is transferred through
the wheels to the ground (Figure 17). In accordance with Newton’s Third Law of Mo-
tion, the Law of Reciprocal Actions, the gravel pile exerts an equal and opposite force
upon the loader bucket. A typical sequence for actually filling the bucket is then to
break material by tilting backwards a little, lifting, and penetrating even further.
Figure 17 below shows how these two efforts work against each other: in order to
achieve a lifting force, the cylinders have to create an anti-clockwise moment around
the loading unit’s main bearing in the front frame. At the same time, the reaction force
from the traction effort creates a clockwise moment that counteracts the lifting effort.
The ground and the gravel pile thus connect drive train and hydraulics, forcing them
to interact with each other, and to perceive each other’s effort as an external load. Since
both working systems are already mechanically connected via the engine (see Figure 16
above), an interesting situation arises in which engine torque is transferred to the wheels
to accomplish traction, but at the same time counteracts that part of the engine torque
that has been transferred through hydraulics to accomplish lifting – in turn requiring
even more torque to be transferred.
Figure 17. Force balance during bucket filling (simplified)
The normalised diagram in Figure 18 visualises that depending on lifting height hlift
and applied traction force Ftrac, the lifting force Flift can be cancelled out totally, which
prevents the bucket from being moved upwards any further. The operator then has to
either reduce the traction force by reducing the engine speed (which is rather counter-
intuitive) or apply the tilt function.
Figure 18. Reduction of lifting force as function of lifting height and applied traction force
The momentary torque or power distribution to drive train and hydraulics is specific
to the working task at hand. It is controlled by the operator, who ultimately balances the
complete system and actively adapts to both the machine, the task at hand, and the
working place.
A design is needed that carefully balances traction and lift/tilt. But as mentioned be-
fore, trivial fixes can seldom be found. The lift and tilt function can not be optimised
without influencing other functions, e.g. machine speed and lifting time. In addition to
this, loading unit geometry and bucket design are of great importance for a smooth
bucket filling.
The traditional way of mitigating the severity of these phenomena has been to design
the wheel loader with a high load margin by utilizing the internal combustion engine at
a high speed (often near governed speed), backed up by a steep torque rise, and to em-
ploy weak torque converters that require a high rate of slip in order to transfer torque.
This creates a certain decoupling effect in the drive train, but it is unfortunately prohibi-
tively costly in terms of fuel efficiency.
In modern working machines the main idea is to utilize the engine at as low speeds as
possible by employing a stiffer torque converter and hydraulic pumps with larger dis-
placement. Also, major components and subsystems are controlled by electronic control
units (ECUs), all connected within a network, as shown schematically in Figure 16. This
makes it possible to give the operator support in controlling the wheel loader (see mid-
dle and right pyramid in Figure 8).
However, it is important to observe that not all limitations that arise because of con-
ventional wheel loader design are perceived as negative by the operator. During their
work with a pump-controlled hydraulic system with a separate hydraulic circuit each for
the lift and tilt function, the authors of [11] and [12] noticed that professional operators
use the aforementioned prioritisation of the tilt function over the lift function to their
advantage:
Instead of operating the lift and tilt lever separately, the operators pull the lift lever to
a certain level, hold it there and then use the tilt lever to control the speed of both tilt
and lift cylinder. By using this procedure during bucket filling, the operator has only to
actively handle two controls (tilt lever and gas pedal) instead of three (because the lift
lever has been locked in a fixed position). This behaviour has been confirmed in the
author’s own interview studies [IV].
During measurements on a research prototype with the pump-controlled hydraulic
system described above, it was observed that the fuel consumption in short loading cy-
cles was unusually high during the bucket filling phase. It could be shown that the op-
erators tried, but were not able, to use the new system just like the old one. With the
speeds of the lift and tilt cylinders now independent of each other, they could no longer
be controlled by just one lever. No record exists as to whether the operators themselves
commented on this and complained about the new functionality (in the author’s own
studies on a similar machine one did), but it is clear that the prototype’s operability was
reduced and the operator’s workload thus greater. To solve this, a similar behaviour was
implemented by software, artificially decreasing the flow demand for the lift function
when the tilt lever was activated (Figure 19).
1.0
0.8
0.6
Lift speed
0.4
0.2
1.0
0.8 0
0.6 0
0.2
0.4 0.4
0.2 0.6
Lift 0 0.8
command 1.0 Tilt command
Figure 19. Reduction of lift speed due to prioritisation
Learning from the example above, the question arises of how many changes one can
make in the design of the human-machine interface and the behaviour of a working ma-
chine without professional operators experiencing this as actually increased workload –
and then, how long it would take for the operators to adapt.
2.4.5 Operability aspects in reversing

In this phase of a short loading cycle the operator changes the direction of the machine
from reverse to forward and aims at the load receiver, while at the same time continuing
to raise the bucket using the lift function. As examined in [I], there is again a conflict
between the working systems:
In order to reverse the machine, the operator lowers the engine speed (otherwise gear
shifting will be jerky and the transmission couplings might wear out prematurely). Less
torque is available at lower speeds. Engine response is also worse at lower speeds,
mainly due to the inertia of the turbo-charger and smoke limiter settings.
When the operator shifts from reverse to forward the loader is still rolling backwards.
This forces an abrupt change in the rotational direction of the torque converter’s turbine
wheel, greatly increasing the slip, which leads to a sudden increase in torque demand
from the already weakened engine.
Continuing to lift the bucket while reversing requires high oil flow, which is propor-
tional to the pump’s displacement and shaft speed (and thus engine speed). In order to
satisfy the flow requirement at lower speed, pump displacement is usually at maximum
in this phase, which together with a high hydraulic pressure due to a full bucket and the
loading unit’s geometry leads to a large demand for driving torque.
Thus, both drive train and hydraulics apply high load to the engine, which at lower
speed has less torque available and a longer response time. In order to lower fuel con-
sumption, operators are usually advised to release the gas pedal and use the brakes in
order to lower machine speed almost to stand-still before shifting into forward. This
greatly reduces torque converter slip and thus saves fuel, but increases workload.
Related to this, the author witnessed another example of how productivity, energy ef-
ficiency and operability (in the form of operator workload) are intertwined. In a research
prototype machine a function had been implemented that automatically lowered engine
speed and applied the brakes when the operator commanded a change of direction the
traditional way by just shifting the transmission gear lever from reverse into forward.
The initial expectation was to save fuel, according to the chain of reasoning as shown
below in Figure 20.
Machine automatically
Reduced slip Reduced
uses brakes when
over torque converter fuel consumption
beginning to reverse
Figure 20. Expected effect of the automatic reversing feature
Comparing fuel consumed per time unit with and without the automatic reversing
feature, it showed that the new feature actually led to higher fuel consumption. The rea-
son for this was the operator using the automatic reversing feature to increase productiv-
ity, which also increases fuel consumption per time unit. Figure 21 shows the actual
course of events.
Reduced slip Increased
uses brakes when
Reduced Increased
operator work load productivity
Operator compensates
for lower work load
and drives harder
Figure 21. Actual effect of the automatic reversing feature
Besides showing how productivity, energy efficiency, and operability are connected
in a working machine, the example above also proves how clearly inappropriate it is to
use fuel consumption measured per time unit in such comparative testing. Fuel effi-
ciency expressed in loaded mass per consumed fuel unit (or the inverse) would have
been a much better choice.
2.4.6 Operability aspects of bucket emptying

Having driven the machine from the gravel pile to the load receiver and having raised
the full bucket to dumping height at the same time, the operator now needs to coordinate
four controls in order to successfully empty the bucket‘s contents:
- Tilt lever in order to dump the material

- Lift lever, gas pedal and brake pedal in order to safely guide the bucket
over the edge of the load receiver and to evenly distribute the material
In this phase, the aforementioned difficulty of no operator control affecting just one
motion is again apparent. The connection between desired increase in hydraulic flow
and thus commanded increase in engine speed now leads to a possibly unwanted in-
crease in traction force, propelling the machine forward at a faster rate. With the wheel
loader already standing in front of the load receiver and only small forward movements
required, this increases the operator’s workload since the machine may risk driving into
the load receiver, which can only be prevented by also applying the brake pedal.
In a series of measurements on a research prototype with a stiffer torque converter
(i.e. one that can create the same torque at a lower slip rate than that of a baseline ma-
chine) the author again witnessed the connection between productivity, energy effi-
ciency, and operability. Originally a decrease in fuel consumption was expected due to
the stiffer torque converter in the research prototype machine, which theoretically leads
to less slip and thus lower drive train losses (Figure 22).
Stiffer torque converter, Reduced slip Increased

yet same traction demand over torque converter fuel efficiency
Figure 22. Expected effect of a stiffer torque converter
Comparing fuel efficiency, the operator, when using the research prototype machine
with the stiffer torque converter, was shown to require more time per loading cycle and
thus consumed more fuel in total and per loaded unit of material (but not per time unit,
as the additional consumption per time unit was close to the average and thus led to
approximately the same mean fuel consumption figure). It was concluded that the main
reason for the longer cycle time was difficulties in bucket filling and bucket emptying
due to the stiffness of the torque converter. Figure 23 shows the course of events.
However, same pump

Stiffer torque converter, Reduced
speed and thus same
yet same traction demand fuel efficiency
engine speed demand
Higher traction force Increased cycle

propels machine faster time and reduced
for same pedal angle productivity
by applying brakes
during bucket emptying
Figure 23. Actual effect of a very stiff torque converter
Especially during bucket emptying the operator found it difficult to handle the higher
traction force at lower torque converter slip rates (and thus lower engine speeds if com-
pared for the same propeller shaft speed). In order to increase hydraulic oil flow, opera-
tors apply the gas pedal to increase pump speed, which due to a conventional wheel
loader’s system design also increases converter speed and thus increases traction force.
To avoid colliding with the load receiver, the operator had to apply the brake pedal,
which resulted in both greater workload and higher drive train losses (because the
torque converter essentially acted as a retarder brake).
2.5 Summary
In summary, there are many interdependencies between functions, components and sub-
systems; and it thus takes a certain amount of training to be able to use a wheel loader
efficiently.
For a manufacturer like Volvo it is therefore of great interest to be able to quantify
operability in order to achieve a working machine with a high degree of machine har-
mony. As with all system properties that arise out of the complex interplay between
components and sub-systems within a machine, the freedom of design is greatest in the
beginning, in the concept development phase. This is also were the potential impact on
both development and product cost is greatest, while the actual cost for exploring vari-
ous scenarios is lowest.
As mentioned before, wheel loaders are versatile machines used in a variety of appli-
cations, which makes it hard to find a globally optimal setup. This problem is even more
accentuated when it comes to hybridisation of these working machines. Customers will
expect great gains in terms of energy efficiency when using the hybrid wheel loaders for
the same variety of applications they used their conventional machines for.
Naturally, the previously described operability challenges will be impacted, too. For
example, the power distribution visualised in Figure 11 and Figure 12 will benefit from
the hybrid’s ability to store energy and thus move power supply and demand in time.
Figure 16 describes not only a conventional machine, but also a parallel hybrid, when
realising the box “External (+/-)“ as a motor/generator (electric hybrid) or motor/pump
(hydraulic hybrid) with a mechanical connection to the engine’s flywheel. However, in
a parallel hybrid topology the sub-systems engine, drive train and hydraulics will be still
coupled to each other, meaning that a change in engine speed will result in a change of
input speed of both the transmission and the hydraulic pump(s).
An even higher degree of freedom in solving challenges of both energy efficiency
and operability can be achieved by using a series hybrid topology, thus further de-
coupling the system. Some of our work in this respect is being discussed in [V].
3
Simulation models
V IRTUAL PROTOTYPING has been generally adopted in product development in

order to minimise the traditional reliance on testing of physical prototypes. It thus con-
stitutes a major step towards solving the conflict between actual increasing development
cost and time due to increasing customer demands on one side, and the need to decrease
development cost and time due to increasing competition on the other.
3.1 State of the art

3.1.1 General
Computer simulation of complete vehicles and machines can today be considered to be
the state of the art in the conceptual design phase. Especially in the on-road sector, be it
on the consumer side (cars) or on the commercial side (trucks and buses), the trend has
progressed from simulation of sub-systems to the simulation of complete systems,
among other things to evaluate performance, energy efficiency, handling, comfort, and
durability.
Particularly challenging for the off-road equipment industry is that its products,
working machines, are complex in architecture, where, as we have seen in the previous
chapter, sub-systems of various domains and their interconnections cause a non-trivial
behaviour of the total system. In addition to this, as we conclude in [I], for working ma-
chines the modelling needs to be extended beyond the technical system and include the
working environment and the human operator in a suitable way. This challenge requires
significant investments, at first in licenses for powerful simulation software, but above
all in building up knowledge of how to use simulation in one’s product development
process. At the very least, methods for modelling and verification need to be developed.
But it is also important to generally define what simulation shall be used for, i.e. analys-
ing virtually what is not (or cannot) be analysed by physical testing – or to complement
or even replace established physical tests. Once fully implemented, simulation will lead
to a reduction in both development time and cost, but the way towards this goal is a
time-consuming and costly process itself. Many smaller manufacturers of working ma-
chines might thus be more reluctant to invest in simulation capabilities, even though
their products will ultimately benefit.
3.1.2 Machine models

Naturally, in order to simulate complete working machines, most of the machines’ sub-
systems and components need to be modelled in a modular way and at various levels of
detail. This is usually a collaborative activity. As noted by many researchers, engineers
have often already chosen one domain-specific simulation program that they are famil-
iar with. Instead of forcing migration to one monolithic simulation system that can be
used in several domains (but offers only limited functionality in the individual domain),
a better approach is to couple the specialised, single-domain tools. This has the advan-
tage that both pre- and post-processing are done decentralised, in the engineer’s do-
main-specific tools. In [13] Larsson examines how interoperability in modelling and
simulation can be supported. Part of his work is the development of a co-simulation
technique and application to a model of a complete wheel loader.
A conventional hierarchical and modular approach gives the engineer flexibility re-
garding creation of models with various levels of detail (which parts of papers [I] and
[II] elaborate on). Another aspect of modular modelling is that flexibility is gained re-
garding input data procurement. Planning to reuse an existing model by parameter up-
date implies that exactly the same type of input data is available for the update as it was
for the original model. If a diesel engine has been modelled in great detail with hun-
dreds of parameters, then an update is difficult if all the engineer has got is a maximum
torque curve. “The more detailed a model is, the more time must be spent on defining
the parameters, based on measurements or theoretical engineering reasoning”, notes
Makkonen in [14]. Test data might not be compatible due to different measurement set-
ups, and theoretical reasoning requires the engineer to reason in a similar (if not identi-
cal) way to the creator of the original model. In practice, it might prove more time- and
cost-efficient to just develop a new module from scratch, but with an identical interface.
Despite issues with interoperability and input data procurement, modelling the tech-
nical system is relatively straightforward but time-consuming if model validation is
taken seriously. Recent publications on simulation model-based Systems Engineering
[15][16] reflect the existing challenges.
3.1.3 Environment models

Looking again at the simplified wheel loader system in Figure 6, we see that the ma-
chine interacts with its environment in two places: tyres and bucket.
Quite a few research papers can be found that explicitly focus on wheel-ground inter-
action in off-road situations. We will provide no references here, because neither study-
ing the effects on the ground nor developing a new tyre model was the focus of our re-
search. We used what was available, given a reasonable match with our needs.
In our application we required a tyre model that was computationally efficient yet
sophisticated enough to result in approximately correct traction and wheel loads. Virtu-
ally all commercially available tyre models were initially developed for on-road vehi-
cles with relatively small wheels and medium or high speeds. In order to analyse phe-
nomena at lower speeds these models have later been revised or new ones introduced.
However, all of these tyre models need comprehensive input data, regardless of whether
Simulation models 31
they are based on a mechanical model or on scaling of test results. When our research
began almost ten years ago, the global manufacturers of large off-road tyres were un-
able, unwilling, or simply unused to provide these data. This has now changed to a cer-
tain degree, even though there is still no streamlined process for acquiring even the few
basic data needed for simplified finite-element models (with which in turn proper tyre
parameters can be estimated).
The other place of interaction between a working machine and its environment is at
the working tool. For most work scenarios, a proper environment model might be trivial
to achieve. However, this is not the case for bucket loading of gravel or other granular
material.
Lumped models are usually the first approach. Hemami [17] studies the motion tra-
jectory during bucket loading of an LHD-loader (Load-Haul-Dump). A review of resis-
tive force models for earthmoving processes is made by Blouin et al. in [18]. The sim-
plified model we used for predicting digging forces was presented by Ericsson and
Slättengren in [19]. (Again, this was not in the focus of our research.) A more recent
work is reported by Worley and Saponara in [20].
Autonomous excavation is the motivation behind many works on lumped environ-
ment models (including most of the above). The work of Cannon [21] and Tan et al.
[22] are two such examples.
Over the decades there has been a lot of research on particle models for complex and
granular materials, Discrete Element Modeling (DEM) being one of the main ap-
proaches. There are several variants on DEM, of which some of the more complex bor-
der on rigid multi-body simulation (MBS) with handling of contact and friction. Per-
forming full 3D DEM simulations takes a lot of time and 2D and 2.5D approximations
are therefore also used.
Furthermore, there are the continuum methods with field-based description of the
material including elasticity and plasticity. The resulting equations are solved with ei-
ther mesh-free particle methods like Smoothed Particle Hydrodynamics (SPH) or Finite
Elements Methods (FEM).
There are many publications available in these fields, see [23] and [24] for an intro-
duction and overview. A recent publication that reports the co-simulation of MBS,
simulating the working machine, and DEM, handling the bucket-soil interaction, is [25].
3.1.4 The need for driver/pilot/operator models

Traditionally, simulation engineers who model complete vehicles or machines do not
include any driver or operator model, but control their virtual experiments by relatively
simple algorithms that prescribe a specific time sequence of either input signals, system
states or output signals.
The latter, called inverse (or backward) simulation is used predominantly in the con-
ceptual design phase. This method takes a fixed output (e.g. vehicle speed) and calcu-
lates backwards in order to obtain the results. Such simulations are fast and often used
for optimisation or exploration of the design space. They are also frequently used when
comparing the energy efficiency of different vehicles, for instance using the synthetic
on-road New European Driving Cycle (NEDC). The problem is not only that cycles like
the NEDC are simplistic to the point of actually being unrealistic, but also that often
mere quasi-static calculations are performed, ignoring any dynamic effects. Research is
being done with the aim of including dynamics into inverse simulation, e.g. [26], but it
still ignores the human operator’s ability to adapt. Forcing a simulated vehicle or ma-
chine to behave precisely according to a specific output will in many cases result in
forced input signals that do not reflect how a human operator would have used the sys-
tem. In some cases, it may even result in unphysical results, especially when more than
one path for power transfer exists – as is the case for the working machines we consider
in our research.
Choosing instead to precisely track a time series of system states (for instance engine
speed and engine torque) does not change that much, it rather makes the results more
dubious. This is because a real driver uses a car according to a strategy or tactic, i.e. the
inputs are controlled, according to the output. For example, forcing a simulated car to
precisely match a time series of engine speed and engine torque means that we still
track the output signals of a reference car (i.e. its speed), and still with all the dynamics
of that reference car (e.g. engine dynamics) hardcoded into the required output signal,
but now we have also for example implicitly hardcoded values for axle ratio, wheel di-
ameter, and losses due to friction and air resistance.
The last simple option then would be to track a time series of input signals in a for-
ward simulation. A first approach might be to simply record a human operator’s control
signals, adjust them by filtering, scaling etc, and input them into the model. This method
is applicable in situations where a technical system’s reaction to a specific time se-
quence of input signals is of interest. One example might be a brake test or an avoidance
manoeuvre to test car stability. This will give correct results with respect to the dynamic
behaviour of the system, but it replaces the adaptive closed-loop control of a human
driver or operator with open-loop control. We again have the problem of implicitly
hardcoded values, as already mentioned above. The simulation results may be correct,
but irrelevant because the output does not match the desired driving or working cycle.
A combination of these principle modelling approaches is also used. Simulation en-
gineers sometimes want to avoid modelling the complete system and give the load from
one sub-system as a reference in time or position. In [27] and [28] for example, the au-
thors did not want to model the hydraulic system of a backhoe / wheel loader and thus
load the engine according to a defined power/torque profile, which has been determined
earlier in either a measurement or simulation. The problem with this is that these loads
can become unsynchronized with the simulation model (will be applied too early or too
late, time-wise or position-wise) and will thus make such comparisons invalid.
What we need are therefore simulation models of the human operator that use the
system model like a human operator would use the physical system in reality. Just as the
requirement on the system model’s level of detail is dependent on the question we want
the simulation to answer, so is the requirement on the operator model’s level of detail.
In some cases it might actually be enough to let a simple PI-controller with anti-windup
track a time-series of vehicle speed values, like in [29] and more or less also in [30]. But
in all cases where the behaviour and limitations of the operator affect the behaviour and
limitations of the entire system, we clearly need operator models with a greater level of
detail, models that experience feedback and can make more complex decisions rather
than just slavishly follow an arbitrary path or trajectory. If we are interested in the Joint
System perspective, i.e. how the human operator and the machine combined form a joint
cognitive system [1], we need yet other types of models – which we will see have been
developed in the aerospace and automotive fields.
Any operator model to be used in the product development process for working ma-
chines needs to be
- adaptive (in the sense of the operator model itself being able to adapt to ba-
sic variation in the scenario)
- flexible (in the sense of easy to manually adapt to a new scenario)
- transparent (in the sense of easy to understand and validate)
3.1.5 Pilot models

Modelling the behaviour of human pilots in aircraft was an active research area years
before vehicle driver modelling. MacAdam elaborates briefly on this in [31] and gives
literature references. In the beginning, the focus was on the human pilot as a controller
of dynamical plants; and the approach of choice was classic control theory. Much prac-
tical research was conducted to measure transfer functions of human pilots performing
specific manoeuvres, e.g. [32][33][34].
However, it was soon realised that to be able to assess handling qualities, the aircraft
and the human pilot needed to be regarded as a combined system. In [35], Cooper and
Harper defined handling as “those qualities or characteristics of an aircraft that govern
the ease and precision with which a pilot is able to perform the tasks required in support
of an aircraft role.” In [36], from 1984, the same authors note that “the quality of han-
dling is intimately linked to the dynamic response characteristics of the airplane and
human pilot acting together to perform a task.” They conclude that since the pilot is
difficult to describe analytically with accuracy, one has to resort to experiments in order
to experience the real system dynamics.
In [36], Harper and Cooper describe the pilot-vehicle dynamic system in detail and
acknowledge that the ability of the pilot to adapt to the airplane. They continue by de-
scribing methods for assessing pilot-vehicle system quality, but only briefly mention
analytical ones, accompanied by a quote from a 1930 paper by Cowley and Skan [37]:
“A mathematical investigation of the controlled motion is rendered almost impossible
on account of the adaptability of the pilot. Thus, if it is found that the pilot operates the
controls of a certain machine according to certain laws, and so obtains the best perform-
ance, it cannot be assumed that the same pilot would apply the same laws to another
machine. He would subconsciously, if not intentionally, change his methods to suit the
new conditions and the various laws possible to a pilot are too numerous for a general
analysis.”
In light of this quote, that was still considered to be true in 1984 when [36] was pub-
lished, more recent developments such as cognitive architectures show the impressive
amount of progress that has been made in the field [38].
When the focus (like ours) is more on the engineering part, i.e. analysis aiming at
improving the aircraft, then pilot models tend to become simpler, lose the cognitive per-
spective and act more as advanced controllers. Simulation engineers then employ opti-
mal control [39][40] and fuzzy logic [41], to name just a few approaches.
3.1.6 Driver models

We have already mentioned that driver modelling has its roots in early human-machine
studies related to airplanes. MacAdam starts with this fact in his comprehensive over-
view [31], focusing on common lateral and longitudinal control activities like steering
and braking/accelerating of a vehicle.
The same insight of having to consider technical system and human operator as one
combined system was quickly won also in the automotive field, and MacAdam quotes
from an early paper by Rashevsky [42]: “The car and driver constitute a complex feed-
back system. The behavior of the car results in certain reactions by the driver. Inversely,
the behavior of the driver affects the behavior of the car. This ‘man-machine’ system
cannot, in many instances, be separated into the purely ‘mechanical’ and the purely
‘human’ components. The system must be treated as a whole.”
MacAdam characterises the human driver by his/her physical limitation and attrib-
utes. He discusses physical limitations such as human time delay and threshold limita-
tions, visual characteristics, motion influences, auditory information, as well as tactile
and haptic information. MacAdam mentiones that the human controller’s difficulties
increase with increased control order and therefore with increased complexity of the
controlled system. While zero-order plants (position control) required one correction in
a quoted study, first order systems (speed control) required two corrections. An increase
in control order from first order to second order (acceleration control) reportedly leads
to a significant increase in control errors. In this respect it is interesting to point out that
a wheel loader operator on some occasions during bucket filling and bucket emptying
has to control the jerk, i.e. the time derivative of acceleration, of the drive train and the
hydraulic functions, thus controlling a third order system.
Continuing with physical attributes, MacAdam lists preview utilisation as an essen-
tial part of human driver control behaviour, as well as the human controller’s ability to
adapt to changing dynamics of the controlled vehicle, the internal model of the human
driver, and also learning, anticipation and planning abilities.
In conclusion MacAdam compiles a list of the minimal set of features required to
build a driver model:
- transport time delay
- use of preview by the driver for sensing upcoming lateral and longitudinal
control requirements
- driver adaptation capability, reflecting knowledge or understanding of the

controlled vehicle’s dynamics (and detects changes that might have oc-
curred)
- linear “crossover model” behaviour when close to the crossover frequency
(describing that as quasi-linear transfer function can be derived for the
combined system of vehicle and driver)
- internal vehicle model
The following features are listed by MacAdam as additional and desirable:
- incoming sensory signals subjected to neural delays, thresholds, rate limit-
ers, and dynamical properties of individual sensory channels
- output channels such as steering, braking and accelerator pedal subjected to
neuromuscular filtering
- path preference of the model or ability to adjust or alternate path
- ability to adjust speed based on upcoming lateral path requirements in order
to enhance directional stability (e.g. in curves or obstacle avoidance ma-
noeuvres)
- surprise or situational awareness features to allow for discrimination be-
tween driver reaction in familiar vs. unfamiliar scenarios
- inclusion of driver skill.
MacAdam gives extensive literature references for published driver models, which
we will not reproduce here. Instead, we refer to [31], and for some additional references
also to also Erséus’ dissertation [43], a more recent publication in this field.
Another excellent starting point for a review of driver models, their development and
applications is paper [44] by Plöchl and Edelmann. The authors acknowledge that even
though vehicle and driver form a couple, the aim of investigation is often focused on
either the vehicle or the driver, although applications exist where the couple is not split.
They group and present various driver models by the main focus of the analysis:
- focus on the vehicle
• design and tuning of vehicle components or the overall vehicle
• vehicle dynamics
• active safety matters with respect to on-board systems
- focus on the driver
• understanding driver and (individual) driving behaviour in general
• path and speed planning features
• optimised driver behaviour
• subjective driving perception
- focus on the overall (combined) system
• reconstruction of accidents and accident analysis/prevention
• learning tasks
• evaluation of driveability and handling

• driver support systems
• active safety
- focus on the environment/traffic
• influence on/of other drivers in simulator or in traffic
• traffic flow and general understanding of the “system” traffic
The overall focus in [44] seems to be more on mechanistic/controller-oriented driver
models and less on behavioural driver models, even though the more important recent
developments are mentioned.
In [45], a well-cited publication from 1983, Rasmussen identifies three levels of be-
haviour in an operator's interaction with a system: skill-based, rule-based and knowl-
edge-based. While a skill-based interaction requires very little or no conscious control
to perform or execute an intended action, in a rule-based interaction an operator acts
according to an acquired or provided set of instructions. A knowledge-based interaction
represents the highest level of control, which is employed in novel or unexpected situa-
tions. Here, the operator has to devise an own strategy according to his/her understand-
ing of the current situation and of the system to be controlled. The operator's mental
workload is thus highest in knowledge-based interactions.
For an experienced wheel loader operator, how to fill a bucket in a gravel pile would
be a pure skill-based interaction, while loading shot rock is rule-based due to the inher-
ently higher difficulty of this task. The experienced operator will follow a learned set of
rules. For a novice, however, without any instructions loading shot rock might be
knowledge-based and bucket filling will be rule-based, following the instructions pro-
vided by a senior operator or supervisor.
Another classic paper is [46], written by Michon in 1985. It contains a critical review
of the contemporary state of the art in driver behaviour modelling, starting with the
harsh lines: “There appears to be a lack of new ideas in driver behavior modeling. Al-
though behavioral research is under some pressure, it seems too facile to attribute this
deplorable state of affairs only to a lack of research funds. In my opinion the causal
chain may well run in the opposite direction. An analysis of what is wrong has led me to
the conclusion that human factors research in the area of driver behavior has hardly
been touched by the ‘cognitive revolution’ that swept psychology in the past fifteen
years. A more cognitive approach might seem advisable and the ‘promise of progress’
of such an approach should be assessed.”
Michon wants to focus on a general theory of driving and towards an intelligent,
knowledge and rule-based model of the driver that will be capable of dealing with a
wide variety of realistic, complex situations. He misses the cognitive perspective, which
by then had been adopted by Cognitive Psychology, but was largely ignored in the field
of Human Factors. “Much of the work in the area of driver behavior modeling is not
relevant for a cognitive approach to the driving task.” Michon writes. After a detailed
review of the more relevant behavioural driver models, he advises an alternative, cogni-
tive approach, based on state machines (also called “discrete-event systems” or “produc-
tion systems”). Michon mentions the work done by others, who have built information
flow models, but these are not cognitive and thus “once the parameters for a particular
run of the model have been determined the program will run strictly on the basis of
fixed algorithms and further external inputs. If the outcome of a program run will be at
all surprising, it is simply because of the complexity of the computation.”
We can add here that technically speaking the models presented by us in [II]-[IV] fall
into the same group; our focus, however, was not to model operator behaviour and the
operator’s cognitive processes, but rather to assess operability in order to improve the
analysed working machine. Following the classification made in [44], our models are
thus in another analysis category than those criticised by Michon in [46].
With respect to the general problem of driving, Michon divides the task into three
levels of skills and control: strategical (planning), tactical (manoeuvring), and opera-
tional (control). The operational processes involve the use of the control inputs (such as
steering wheel, accelerator pedal and brake) for stable driving, while the tactical proc-
esses concern safe interaction with other vehicles and the environment. Strategic proc-
esses involve higher-level reasoning and planning.
3.1.7 Cognitive architectures

Not only does Michon in [46], published in 1985, demand the development of rule-
based driver behaviour models, for which he gives several smaller examples of typical
IF-THEN sentences, he also demands “a flexible, general theory of human cognition
and a calculus which does embody that theory”. Michon notes that the focus needs to
move “from adaptive control of behaviour to adaptive control of thought”. It is certainly
no coincidence that one of the most prominent cognitive architecture available today,
many years later, is called “Adaptive Control of Thought – Rational” (ACT-R).
As explained in [38] and [47], executable cognitive models are used to describe the
mental processes of human beings in the assessing situations and choice of actions. The
first part of such cognitive models is the cognitive architecture, which is a computa-
tional framework that integrates task-independent cognitive processes. It incorporates
built-in, well-tested parameters and constraints on cognitive and perceptual-motor re-
sponses. By modelling for example a human driver in a cognitive architecture all these
parameters and constraints are inherited, which results in more psychologically plausi-
ble models, especially compared to modelling the driver as a continuous controller –
which disregards the discrete nature of a driver’s control actions, as well as the multi-
tasking nature of the driving task itself. The second part of cognitive models is a formal
model of task-specific knowledge (for instance flight procedures or traffic regulations)
that is executed by the cognitive architecture in a psychologically plausible way.
In [38], a more recent publication from 2009, Lüdtke et al. present their own cogni-
tive architecture for simulating human-machine interaction in aeronautics and automo-
tive research, ultimately supporting the design of assistance systems for pilots and vehi-
cle drivers.
The most commonly used frameworks seem to be ACT-R and possibly SOAR, but
[48] examine a great many more systems in their extensive effort to compare results
from various models realised in different frameworks both with each other and with
actual human operators performing the same task in reality.
While [38] focuses more on human error during task execution, [47] presents an ac-
tual driver model that is integrated into ACT-R. Salvucci et al. model a human driver
navigating a four-lane highway with traffic and demonstrate that the model is able to
correctly predict both lower-level control such as steering actions and higher-level cog-
nitive tasks such as decision making and task management. Many of the above predic-
tions are not explicitly built into the model, but rather come as a result of the model
being implemented into the cognitive architecture – showing the strength of this ap-
proach. With [47] published in 2001, five years later Salvucci reports on progress of his
work in [49][50]. These papers also include a good overview of the problem and an
introduction to ACT-R.
Just as Salvucci and others work with driver modelling within the framework of a
cognitive architecture, others are working on using and enhancing cognitive architecture
for their field of research. In [51], Robbins et al. present their efforts, and those of oth-
ers, to enhance ACT-R with spatial awareness in order to perform ergonomics-focused
analyses that would traditionally would fall within the area of Digital Human Modelling
(DHM).
3.1.8 Operator models

Surprisingly little could be found in a literature review focused on working machines
and designing operator models for the purpose of simulation. We could find even less
with a focus on the combined system of working machine and operator; rather the pre-
dominant focus is on the machine itself.
We certainly expected to find several operator models based on artificial neural net-
works. In our work we decided against such an approach, because it is less flexible (in
the sense of easy to manually adapt to a new scenario), and certainly not transparent (in
the sense of easy to understand what happens inside the model and easy to validate), but
we thought this approach would be used by several others due to similarities with con-
trol strategies within the field of robotic/autonomous excavation.
However, only one such attempt has been found, reported in [52] back in 1993. In
this paper Novak et al. develop and use a back-propagating Neural Network to resemble
how a human operator uses the steering function in a wheel loader. They state their mo-
tivation for using this approach as follows: “Mathematical functions have been used in
the past in an attempt to model a human in their operation of many types of vehicles.
Since such functions can typically only be derived for situations in which the problem
domain is thoroughly understood, research continues in an effort to develop a complete
‘operator model’.”
This statement is interesting to reflect on, as this kind of reasoning is exactly our mo-
tivation for not using neural networks. It is our strong conviction that no modelling and
simulation effort should be undertaken if the problem domain is not properly under-
stood. We acknowledge the usefulness of simulation as a tool to gain insights by virtual
experiments, sometimes even involving different models of the same object in order to
use the structure of the best fitting model as a working theory to derive deeper knowl-
edge on perhaps otherwise not accessible inner states. In some way, therefore, it is le-
gitimate to start with less domain knowledge and use simulation to increase it through
experimenting with various modelling approaches. However, some level of domain
knowledge is required to begin with.
Trained neural networks are essentially impossible to manually review and validate;
the simulation engineer will therefore not gain any new knowledge from them. We re-
gard artificial neural networks and other learning systems as equivalent to advanced
curve fitting, but without revealing which equation is used to model the data used for
training. With domain knowledge it is legitimate to use curve fitting for extrapolations
(e.g. if one knows that the underlying process produces a logarithmic relationship, then
it is legitimate to carefully extrapolate beyond the data used for logarithmic curve fit-
ting). However, without domain knowledge any attempt at extrapolation it is dangerous
(e.g. if a cubic polynomial has been fitted to data from a process with an inherent loga-
rithmic relationship, then any larger extrapolation will result in misleading predictions).
By analogy, any prediction of the Neural Network’s behaviour in new, not previously
trained situations must be critically questioned, since we do not know of a way to a pri-
ori determine a Neural Network’s extended region of fit (i.e. how much outside of the
area covered by the training the model can be used).
Novak et al. report success with regard to having been able to train a Neural Network
to use the steering function of a virtual wheel loader in the same way a human operator
has done in the set of measurements used to train the network. They conclude their pa-
per with the hope that they can also model the other operator inputs with similar or more
complex networks, and that they will be able to develop an accurate operator model. It
is probable that this work has been continued at the same institution, but no more publi-
cations could be found regarding this matter.
Related to the same application, in 2002 Norris et al. report in [53] on their work
with optimising automatic control applications of electro-hydraulic steering systems in
frame-articulated machines, such as wheel loaders. They present a model employing
fuzzy logic that emulates human operator perception and experience in real-time. Again,
only the steering function is modelled.
Within the same research cluster, Zhang et al. validate control strategies by conduct-
ing human-in-the-loop experiments in their Earthmoving Vehicle Powertrain Simulator.
The corresponding paper [10] gives some insight into their reasoning regarding human-
machine interaction. They acknowledge the difference between task-oriented jobs (such
as wheel loader operation in a short loading cycle, which they want to simulate) and
reference-oriented jobs. They conclude “The total productivity depends on how well the
task is fulfilled, what the fuel economy is, and how long it takes to finish each cycle.”
They realise and write that the effect of human-machine interaction cannot be evaluated
in the basic tests employing reference tracking and the disturbance rejection.
A traditional path/trajectory-following driver model is utilised in [27] by Pióro et al.,
simulating different hybrid powertrain variants for a backhoe loader. Despite the main
author’s reluctance to disclose any details, from the supplied screenshots we can deduct
anyway that they use the DRVDRVA00A component, which is the simplest driver
model available in AMEsim’s IFP Drive library. It only requires a velocity reference
and its outputs are the position of the brake and accelerator pedals, both PID-controlled
with user-adjustable gains. There are no advanced features in this driver model, apart
from a look-ahead time and a time delay between end of braking and start of accelera-
tion, both adjustable. The driver model is not capable of actuating the hydraulic system,
which is absent in all scenarios and only modelled as a power (torque) sink. Due to the
simplistic setup, the cycle time in all simulated variants differs, which also means that
the fuel consumption is not really comparable. For our work this study serves as a good
example of why proper operator models are needed, even though operability might not
be the focus of the analysis – and why all important sub-systems need to be modelled if
relevant simulation results for the complete machine are desired.
In [28] Thiebes and Vollmer reference [IX] (to which [I]-[III] were appended) and
acknowledge the problems of tracking a reference velocity with simple PI-controllers.
However, they do not consider this to be a principle modelling error, as Zhang et al. do
in [10] and as we do in [IX], rather their approach is still to give the machine model
reference values to track, but connect a controller that also tracks position. For Thiebe
and Vollmer the problem seems only that traditional PI-controllers will lead to different
travelling distances if machines of different performance characteristics are compared.
From our perspective the problems only start there. If a machine takes longer to ac-
celerate to a certain speed reference, then it will in total have travelled a shorter distance
if the speed reference values are given at specific times. Let us consider the example of
simulating bucket filling with a wheel loader model where the hydraulic system is only
modelled as a simple load on the engine, to be applied after a certain time. At first, the
machine has to travel to the place of bucket filling. If the machine has less tractive per-
formance than anticipated, then at the time when the “hydraulic load” is applied to the
engine, the machine has not even arrived at the gravel pile.
Thiebe and Vollmer try to correct this situation by also feeding back the position er-
ror and using it to control the vehicle’s speed. If the speed error is low, then the position
error (and thus the position reference signal) is dominantly controlling the speed de-
mand, while the speed reference signal dominates if speed errors are high.
A further development of their reasoning would be to directly tie reference values
like “hydraulic load” to position, rather than time. Consider, however, the example of
bucket emptying: After bucket filling, the operator drives backwards towards the revers-
ing point and with the lifting function engaged the whole time. The operator chooses the
reversing point such that having arrived at the load receiver and starting to empty the
bucket, the lifting height will be sufficient to do so without delay. If the model tracks
machine speed and hydraulic load given by positions, then it will always reverse at the
same point. Given a comparison of machines with different hydraulic system perform-
ance, the bucket height will be different at the load receiver, thus the simulation results
will therefore not be comparable.
The area of application of Thiebe and Vollmer’s approach will be restricted to the
studies of single components and sub-systems. Controlling a model of a working ma-
chine (hybrid or not) by means of speed and position reference signals alone will not
produce correct results, as argued above.
With the complex system shown in Figure 6 and Figure 16 in mind, we also have to
realise that for our purposes it is not correct to disregard a major sub-system like hy-
draulics. The power balance between engine, hydraulics and drive train, together with
the fact that engine power is limited, can lead to reduced drive train performance, de-
pending on the state of the engine and the hydraulics. This can also be the case for hy-
draulics if the machine is featuring a hydrostatic drive train. Or the engine might stall
altogether. In our work it is thus impossible to make such simplifications and still be
able to correctly compare between simulated machine variants.
Inspired by our research reported in [I]-[III], Elezaby et al. essentially copy and re-
validate our state machine/discrete-event approach in [54]. In his dissertation [55]
(originally published version no longer available online), Elezaby relies to a large extent
on [I]-[III] for its background, literature review and description of state of the art, using
much of our material, both text and figures, in a surprisingly liberal way.
At its core, the operator model presented by Elezaby employs a finite state machine
to select strategy and to decide upon the sequence of sub-tasks needed to accomplish the
required mission. The seven sub-tasks in this operator model are
- shut down / emergency brake or stop
- back from the truck / site
- go to truck / site
- lift / lower to a point
- dump load / rack bucket
- select gear
- filling the bucket
The description of the sub-tasks are not detailed enough to deduct anything about the
precise control strategies employed, but at least for bucket filling it seems that Elezaby
follows the ideas of [II]. The difference between our operator and machine models and
Elezaby’s seems to be that he did not have access to the detailed specifications of the
wheel loader’s hydraulic and control system. He reports having tried using PID control-
lers to actuate the hydraulic valves according to the operator model’s commands, but
experienced difficulties with this approach, as in order to avoid stability problems it
required manual fine-tuning for every machine model to be simulated. Elezaby therefore
uses Neural Network controllers in a model reference control scheme for controlling the
hydraulic valves. He reports that this led to his operator model being able to control
wheel loaders of various sizes.
3.1.9 Autonomous excavation

Finding suitable strategies and control schemes for robotic excavation has been an ac-
tive research area for several years. Just like in our work, researchers in the area of
autonomous excavation need to reflect upon working machines, loading cycles, the in-
fluence of the environment and the human operator etc. Our research does not involve
control strategies or any attempt at automation, but there are certain similarities and
possibly also synergies.
Hemami specifically examines bucket filling in [17] and also Wu [56], starting off
with analysing wheel loader operation in general and later focusing on the bucket filling
phase. Both handle the problem as one that can be solved by following predefined tra-
jectories.
Using material gained from personal interviews with professional operators, a litera-
ture study, and statistical analysis of full-scale experiments, Wu reflects in depth upon
the loading cycle and a human operator’s techniques and strategies.
The developed controller creates an ideal trajectory in advance, based on bucket vol-
ume, workspace, and rock pile parameters. During digging, this trajectory is roughly
followed, but adjustments are made to machine speed and bucket position. Wu recog-
nises that different material properties require variations in the scooping strategy, and
identifies three usable strategies, labelled “easy-scoop”, “normal-scoop”, and “hard-
scoop”. “Hard-scoop” implements small oscillations in bucket tilt angle, while in “easy-
scoop” the bucket is tilted backwards faster than in “normal-scoop” and without any
oscillation. Wu makes similar considerations regarding the crowding phase and reflects
upon repeatability in order to ensure that the same principal bucket filling methods are
also applicable to the next loading cycle.
Machine speed control is an aspect for which Wu uses statistical analysis instead of
arguing for a specific strategy. He finds an approximate relationship between engine
speed and bucket tip height, which is then implemented into the developed digging con-
troller. This controller is adaptive, which ensures continuous improvement of the accu-
racy with which machine speed and bucket penetration depth are predicted. Wu also
investigates the occurrence of wheel slip and develops an algorithm for detecting it
through sensor fusion. A strategy for wheel slip reduction is then implemented into the
digging controller.
The models developed in Wu’s thesis are statistical, non-physical models that are
constructed with Neural Nets. Human decision-making is approximated with the help of
fuzzy logic, which also ensures insensitivity to sensor noise. Wu verifies the developed
controller in a series of simulation experiments (not implemented in a real machine in
the work reported in [56]).
Wu’s thesis offers interesting insights into working techniques and strategies of hu-
man operators, which are comparable to the results of our own experiments. Wu does
not analyse the technical system to any great extent (except for the loader’s linkage),
and thus makes no reasoning regarding power transfer etc., but his principle work on the
adaptive digging controller may be applicable to a refined operator model in our future
research.
In his Master thesis [57], Marshall conducts a wide range of full-scale experiments
on bucket filling under excavation conditions comparable to those found in an under-
ground mining situation. As part of the work, excavation trials were conducted by pro-
fessional operators in material of varying particle size and composition. In these ex-
periments, some machine parameters such as translation and hydraulic cylinder forces
were recorded, while other parameters such as the operator’s command input signals
and tractive effort were unfortunately not considered. Furthermore, due to significant
wheel slippage, the vehicle's translational motion could not be measured correctly.
Based on these measurements, Marshall then models the loader’s linkage motions
during excavation using parallel cascade identification (PCI), a nonlinear system identi-
fication technique. However, dynamic problems, especially when identifying dump cyl-
inder motions in free-space, led to outcomes that were “not exceptionally good”. Mar-
shall can show that changes in dump cylinder force correspond well to changes in the
cylinder velocity, and reasons: “The question therefore remains as to whether the model
input velocity changes [...] are due to bucket-rock interaction, or due to signals com-
manded by the operator. Strictly speaking, the answer is likely that both are responsible
[...].”
Automated bucket filling is also in the focus of [58], presenting a state-machine ap-
proach combined with fuzzy logic.
Another example of a path/trajectory-tracking approach can be found in [59], in
which Sarata et al. present their progress towards a wheel loader that performs bucket
loading fully autonomously in a short loading cycle.
3.2 Models developed in our research

In the following we will look at the results of the simulation-related part of our research.
The aim of this work was to provide answers to research questions RQ1-RQ3, with the
goal of being able to assess productivity, energy efficiency and similar operator-
influenced properties of working machines at an early stage in the product development
process. For this, we realised, we needed to both model the operator in a suitable way
and include the working environment.
3.2.1 Machine and environment model

The machine model used in our research was developed in parallel in a collaborative
effort, in which the author participated. All major components and sub-systems are
modelled physically, i.e. no statistical models are used. The simulation of the mechani-
cal domain is done using rigid multi-body simulation in ADAMS. The hydraulic system
is simulated one-dimensionally (ADAMS/Hydraulics). The drive train components de-
veloped by the author are a mixture of one-dimensional models (state equations in
ADAMS) and rigid three-dimensional models (like all other mechanical components).
The machine’s interaction with its environment is limited to ground contact (via its
tyres, using the tyre model supplied with ADAMS) and interaction with an already ex-
isting model of a gravel [19] pile via the loader's bucket.
3.2.2 Methodology for developing operator models

In our work two operator models were realised. Both focus on the tactical (manoeu-
vring) and operational (control) aspects, as defined in [46], and are, in contrast to for
example algorithms for autonomous excavation, not concerned with the strategical level
(planning). Both models are separate entities and the information exchange with ma-
chine and environment is limited to operator inputs and outputs similar to a human be-
ing, i.e. the operator has neither insight nor influence at a deeper level. The machine
model is controlled through variables representing the functions
- accelerator pedal
- lift lever
- tilt lever
- steering
- brake pedal
- gear selector (gear shifting is automatic, except kick-down into 1st gear).
The operator model gets feedback on
- bucket height and angle, also height of the cutting edge over ground
- machine position and orientation in the plane
- machine velocity and distance travelled since start of simulation
- steering angle (between front and rear frame)
- warning flag if any hydraulic cylinder is near min/max extension
- engine speed
- time.
In general the operator model and machine model act as black box models in relation
to each other. No internal states are sent, because the intention is to model human-
machine interaction, which naturally occurs through a limited number of channels.
The methodology employed for both operator models is described in [IV]. In our in-
terviews with professional operators we realised that the way they described the opera-
tion of wheel loaders can be expressed as rules of a state machine, governed by discrete
events. In the orienting phase, interviews are unstructured and aim to establish basic
knowledge of the type of cycle to model, its features, and its characteristics.
In this next step, semi-structured research interviews are conducted with professional
operators who are able to verbally express how they use the machine in the working
task at hand. It is important to go through each cycle phase in detail, noting what event
triggered a reaction, which controls are applied and how long, alternate scenarios, what
defines success, what defines failure etc. In our work we were able to interview one
product specialist who was also a professional test operator and, more importantly, an
experienced machine instructor and thus used to teaching people how to operate wheel
loaders in an efficient manner. Additionally, many brief unstructured interviews were
performed with colleagues of a similar background.
Together with video recordings of the machine in operation and possibly also from
inside the cab, measurement data can serve as an additional source of information. In
our work we equipped test machines with a data acquisitioning system and recorded
control inputs and major machine variables, such as engine speed, traveling distance,
bucket height and angle (etc). We also found video recordings of the machine in opera-
tion to be a valuable complement.
The results from interview studies and possibly additional measurements are then
transformed into general, non-machine specific rules and constraints. This is the last
step before coding the operator model, which makes it necessary to construct the rules
and constraints in a way that is possible to implement as execution paths of a finite state
machine.
The operator model can then be validated by running simulations with varying pa-
rameter setups and comparing to results from test series.
3.2.3 Operator model #1

The first operator model focused on bucket filling and has been realised as state equa-
tions in ADAMS with an algorithm reminiscent of fuzzy logic (using parallel and nested
STEP functions to resemble membership functions, without the need for defuzzifica-
tion). This is explained in more detail in [II]. Figure 24 shows the results of one specific
simulation.
Figure 24. Results of simulated bucket filling: bucket motion and operator input
This approach proved to be successful in the bucket filling phase of a short loading
cycle. Maybe a bit too successful, as the graph for the tilt lever only shows smooth ac-
tuation, in contrast to how human operators use the tilt function during bucket filling.
Using ADAMS functionality also proved quite cumbersome and resulted in less satis-
factory modelling of the remaining phases of a short loading cycle.
In spite of this, we were able to show that this operator model was useful due to its
adaptability. We demonstrated this by simulating the same wheel loader model with two
different torque converters. This component is known to have a vital impact on machine
performance and energy efficiency. In order to obtain the same traction force, a
“weaker” converter requires higher slip between pump and turbine – which for similar
traction force requirements leads to higher engine speeds over a complete loading cycle.
A human operator compensates for this with higher accelerator pedal values, which we
have proven earlier in tests on a physical prototype machine. But this also affects the
hydraulic system, which in turn requires compensation. Figure 25 shows that the opera-
tor model managed to adapt to the new machine characteristics by running the engine at
higher speeds – just as a human operator would have done.
Figure 25. Simulated engine duty for machines equipped with different torque converters
3.2.4 Operator model #2

The second operator model [III], with a focus on the remaining phases of the short load-
ing cycle, was realised in Stateflow and co-simulated with ADAMS (Figure 26).
Simulink was used as the integrating environment. The block “SignalConditioning”
in Figure 26 contains rate limiting functions that prevent the operator model from too
rapid changes in the actuation of a control. The idea with these was to resemble human
limitations. One idea, not realised, was to not directly apply the control commands gen-
erated by the operator model, but to use PID controllers in between operator model and
machine model. The gains of these controllers could then be used as an additional way
to reflect human limitations.
Figure 26. Co-simulation of Stateflow and ADAMS, Simulink as integrating environment
The overall view of the operator model is shown in Figure 27, where each phase has
been modelled as a state containing several sub-states.
Figure 27. Event-based operator model (top level view)
Figure 28 shows the state machine for phase 2 of the operator model. We can see two
parallel threads as a simple representation of a human being’s ability to multitask. The
left thread represents the operator’s occupation with steering the machine to achieve the
typical V-pattern, while the right thread models the operator’s occupation with estimat-
ing when reversal can be initiated. Both are processed in parallel. See Table 3 for a de-
scription of the steps taken.
Figure 28. Phase 2 of the operator model in Stateflow
Table 3. Applied steps in phase 2 of the operator model
# Description
A1 Apply full lift and full gas pedal.
A2 Steer machine so that reversing will take place with machine pointing at work-
place origin (angle at ca 45°). Calculate required steering angle.
A3 Apply full steering, reduce and stop when required steering angle is reached.
A4 Begin steering back when machine is about to point at workplace origin.
A5 Keep driving backwards until path B terminates this phase.
B1 Wait for machine to pass load receiver (fist geometric possibility to reverse).
B2 Wait for extrapolated remaining lifting time to be lower than remaining ex-
trapolated driving time to load receiver.
B3 Calculate required steering angle for perpendicular arrival at load receiver.
Terminate this phase if within achievable limits.
B4 Go to phase 2a (extension, continued straight driving until steering possible).
This operator model’s capability to adapt to basic variations in the working place has
also been validated by running simulations with varying parameter setups. One example
is that of changed speed in the bucket’s lift hydraulics, emulating insufficient pump
capacity. A human operator adapts to this by reversing farther with the wheel loader
until finally driving forward to the load receiver to dump the bucket’s load. This could
also be demonstrated in our simulations without any modification of the operator model.
See Figure 29 for a Machine harmony diagram of both variants; the black curve with
circular markings belonging to the variant with lower lifting speed.
Figure 29. Adaptation to lifting speed
The trace in Figure 30 shows how the operator model is able to adapt to a different
position of the load receiver. In the black curve the load receiver was moved to a more
remote position, which the operator model adapts to and still approaches in a V-shape.
Figure 30. Adaptation to workplace layout
3.2.5 Summary and discussion

With both presented operator models, a “human element” has been introduced in dy-
namic simulation of working machines. Both operator models have been derived from
interview studies in a fairly straightforward manner.
We are of course aware that actual human beings are more than just information
processing systems. Our intention with this work was not to develop a cognitive model
of the joint man-machine system, but rather to improve the machine system simulations
performed during concept development and other phases in the product development
process. As always, it is important to note the scope and limitations of a model.
The operator models developed prescribe the machine’s working cycle in a more ge-
neric way, independently of the machine’s technical parameters. The characteristics of
whole components or sub-systems can therefore be changed without compromising the
validity of the simulation. This gives more relevant answers with respect to machine
performance indicators like productivity and energy efficiency. The results are satisfac-
tory and the methodology is easily usable in other, similar situations.
When presenting these results in [II]-[IV] and [IX] we always concluded that a way
to use simulation to also assess operability still requires more work. It was then specu-
lated that one approach to quantification might be through the definition of an operator
input dose similar to the vibration dose value in the assessment of whole-body vibration
exposure. It was not as easy as we first thought, but in the study presented in [VI] we
managed to derive a similar measure, which we dubbed operator control effort. This
measure, presented in more detail in the next chapter, correlated reasonably well with
the operator’s self-reported workload and can be calculated from the operator’s control
commands – irrespective of whether they are measured signals from a physical test or
generated by the operator models presented above. We have thus closed the loop and
achieved the vision presented in [I]-[IV] and [IX]: operability can be assessed by simu-
lation!
At least this is the case in bucket filling, but it is reasonable to assume that this can be
transferred to other working cycles and other working machines of similar complexity.
Also, we have lately realised the unusual problem that the first operator model was a bit
too good at filling the bucket in a smooth manner. Figure 24 shows none of the other-
wise typical fluctuations in the control command signals, indicative of reduced operabil-
ity, which are clearly present in Figure 15. It seems that our first operator model would
be a good basis for an automatic bucket filling function. However, it reflects an unusu-
ally skilled operator – which is not our intention in simulation, where we instead want to
reflect operators with typical skill levels. In the future we will therefore focus our work
on the second operator model and implement different bucket filling styles.
Operator workload 51
4
Operator workload
A CCORDING TO the definition in [2], operability is “the ease with which a system op-
erator can perform the assigned mission with a system when that system is functioning
as designed”, while operator workload according to [3] is “the portion of the operator’s
limited capacity actually required to perform a particular task”. In this research we focus
on comparative assessment of operability and assume for this a negative monotone rela-
tionship between machine operability and operator workload, i.e. the higher the degree
of operability of a specific machine, the lower the operator workload (and vice versa).
We have thus transformed the problem of quantifying operability into the problem of
comparing operator workload. The operation of a working machine requires both mental
and physical workload; we are therefore interested in ways to measure both.
4.1 Literature research

There are several ways to evaluate workload of a human being. Due to our focus on
operators of working machines, we will exclude methods of loading the operator with
secondary tasks (i.e. small artificial problems the operator is asked to solve in parallel)
from our research. This decision was taken not only because these secondary tasks are
difficult to supervise in the limited space of a mobile working machine’s cabin, but also
because, due to their very nature of applying additional cognitive load, they might inter-
fere with the primary task of operating the machine and potentially even lead to a haz-
ardous situation for both machine operator and bystanders.
For a thorough presentation of (mental) workload measures and their characteristics
such as validity, sensitivity, diagnosticity, selectivity, bandwidth and reliability, pri-
mary-task intrusion etc. the reader is referred to [60] and [61], both excellent starting
points, as well as comprehensive look-up references.
As noted for example in [62], it is as misleading to label self-report measures “sub-
jective” as it is to label psychophysiological measurements “objective”. The former, if
performed honestly, directly involve the highest authority in this regard, which is the
operator him- or herself, while the latter involve third parties trying to deduce workload
from such complex measures as heart rate, which is affected by more than just the work-
load of the person performing the task. It is therefore more meaningful to divide the
quantitative measures into self-reported and observational, with psychophysiological

measures such as heart rate falling into the latter category.
4.1.1 Self-report measures

In [36], Harper and Cooper list experimental methods for assessing pilot-vehicle system
quality, divided into performance measurement and pilot evaluation. With respect to the
former the authors state that since the pilot is adaptive, any performance measurement
should not only include how well the pilot is doing (i.e. task performance), but also how
much effort the pilot is supplying (i.e. workload) in order to achieve that performance.
This matches our long-held hypothesis that a correct performance analysis of a working
machine must include not only productivity and energy efficiency, but also quantified
operator workload.
With respect to pilot evaluation, i.e. the subjective assessment of aircraft handling
qualities by pilots, paper [36] first pinpoints the importance of comment data for provid-
ing specific feedback to engineers and analysts. It then focuses on pilot rating and in
particular on the Cooper-Harper scale, originally introduced in [35], and today still used
in a modified form. It is interesting to note that the authors recommend using only a few
pilots over a long period to evaluate an aircraft’s handling qualities, rather than a larger
group conducting short evaluations. They refer to an earlier study [63] in which the re-
sults of the larger group with shorter evaluation time produced the same central ten-
dency in the rating excursions, but with reduced quality, consistency and meaningful-
ness of the pilot comment data.
Other self-reported measures, like NASA Task Load Index (NASA-TLX) and SWAT
(Subjective Workload Assessment Technique), are described in [60] and [61].
The multi-dimensional NASA-TLX requires the test subjects to self-report on six
sub-scales, assessing mental demands, physical demands, temporal demands, own per-
formance, effort, and frustration. These scales are weighted and compared in pairs with
regard to which contributed most to the operator’s feeling of workload (in total fifteen
comparisons). A workload score is calculated by multiplying each scale’s weight by the
score, adding all six sub-results, and dividing by the total weight.
Also SWAT uses multiple dimensions, time load, mental effort load, and psychologi-
cal stress load, to be assessed on three levels, low, medium and high. Calculation in-
volves a process of two steps, first scale development and then event scoring. In [64]
Rubio et al. compare SWAT and NASA-TLX with Workload Profile (WP), and find the
three to be more or less on par, with WP being somewhat better in terms of sensitivity
and diagnosticity.
4.1.2 Psychophysiological measures

Psychophysioloy not being our area of expertise, we have relied upon the literature, e.g.
[60][61][65][66], and our co-authors in [VII] for knowledge. The following brief over-
view of some of these measures was compiled from these sources (who in turn refer to
many other publications).
In general, the absolute changes in almost any psychophysiological measure are spe-
cific to the individual person and cannot be compared between subjects. They also re-
quire a baseline to be interpreted correctly, as each individual person reacts differently
to changes in workload.
Heart rate (HR) and heart rate variability (HRV)

Together with the respiratory system, the heart supplies oxygen to the human body.
Both react to mental and physical workload, which also means that it might be hard to
separate these effects in later analysis. Even though working in a seated position, pilots,
drivers and operators of working machines have to physically engage various controls
and have also to endure accelerations and jerks as a result of their actions. Any work-
load derived from HR and HRV should therefore be seen as a combined effect of the
operator’s mental and physical efforts.
One has to be aware of that HR is affected by two systems within the central nervous
system; the sympathetic and parasympathetic system. This means that heart rate is af-
fected by both internal attention (e.g. solving mathematical equations) and external at-
tention (reaction to external stimuli). Furthermore, heart rate first slows down when a
person is paying attention to an external stimulus. If the situation turns out to be stress-
ful or dangerous, the heart rate will rapidly rise above the former normal level. But if
the situation is harmless, the heart rate will return to its former normal value.
HR and HRV values can change within seconds, but reliable measurements require a
minimum length of 30-40 seconds and a maximum of five minutes. Sensitivity is re-
ported to decrease for shorter or longer periods. It is also stated that only large changes
in workload can be found – and one has to account for the effects of respiration on heart
rate.
Heart rate is always measured as the inverse of inter-beat interval (IBI), i.e. the time
between two consecutive heartbeats. The problem is how to determine a heart beat. This
can be done by analysing the pulsation of the blood volume in the tip of a finger or toe
by means of photoplethysmography (PPG). In this method the absorption of infrared
light by the blood vessels in the finger/toe is measured (which also gives a measure for
the oxygen concentration in the blood). However, this method does not give very pre-
cise data for the inter-beat interval, since the peak of each pulse is not sharply defined
(which any engineer with knowledge in hydraulics can comprehend by thinking about
the various “orifices” and “flexible hoses” that the cardiovascular system of the human
body consists of). The measuring equipment itself is also not optimal, consisting of a
clamp that, when attached to a finger, will partly obstruct the operator from work.
The less intrusive and potentially more accurate way to determine IBI is to measure
the electrical impulses produced when the heart contracts. This is done in the form of
ECG (electrocardiogram). Within the continuous time signal the characteristic R-wave
(the predominant spike we all know from various films and TV-series) is taken as a
good fix point, and the time between two R-waves is recorded as the IBI. The ECG
method requires three electrodes, to be attached on both sides of the body (either both
arms, wrists or feet, or on the left and right side of the torso). The caveat with ECG is
that it is important that the electrodes have good and stable contact with the body, oth-
erwise measuring artefacts will be produced. As we have found out, this can be a chal-
lenge in the cab of a working machine.
The heart rate of a person varies in a certain pattern even when that person is resting.
The heart rate variability (HRV) measure reflects this variation in heart rate over a pe-
riod of time. During high workload HRV is reported to decrease.
In any statistical analysis of data involving HR and HRV it must be remembered that
HR is the inverse of IBI, and that therefore only measurement data of the latter can be
assumed to follow a normal distribution.
Respiratory rate (RR) and end-tidal CO2 concentration in exhaled air (EtCO2)
Respiration, supplying the blood with oxygen (O2) and expelling carbon dioxide (CO2),
is a very complex function that involves several processes within the human body. As
mentioned before, respiration is connected to heart rate and should therefore be evalu-
ated in context. We have also mentioned that both heart rate and respiration react to
mental and physical workload, which also means that it might be hard to separate these
effects in later analyses.
The venting function of respiration affects the concentration of CO2 in the blood
plasma. This can be assessed via the exhaled air as end tidal CO2 concentration
(EtCO2). With this measure we can determine if excessive breathing occurs, i.e. hyper-
ventilation.
Respiratory activity as a stand-alone measure is not a good indicator of workload
(also because it is interrupted and modified by speech), but it is useful in combination
with HR/HRV and self-report measures.
Respiration rate can be measured by using an abdominal belt to establish an EMG
(electromyogram), i.e. measure the electrical impulses generated by muscular activity in
the area. However, this is must be regarded as more intrusive than the alternative of
measuring RR by measuring the flow of the inhaled and/or exhaled air using a pneu-
mograph. The air flow can be sampled via a nasal cannula or a sample line, inserted
some millimeters into one nostril (however, this requires the operator to exclusively
breath through the nose). When EtCO2 is measured that way, the recording equipment is
also able to calculate RR from the data acquired.
Other psychophysiological measures

Several other physiological measures have been reported to be useful in their respective
area of application:
- peripheral temperature (typically measured on fingers, using a thermistor)
- skin conductance level (also known as galvanic skin response)
- neural activity in the brain, via EEG (electroencephalography)
- eye tracking, via EOG (electrooculography)
- eye blinks
- pupil diameter
- facial muscle activity, via EMG (electromyography)
With regard to all these measures, Summala quotes in [67] a warning by Moray:
“Psychological theories have often striven heroically to account for phenomena in terms
of only one or two variables. There is little doubt that any such effort in the case of
mental workload is doomed to failure because of the complex interactions between task
demands, the subjective attitudes of the operators, and their physiological state. What-
ever model emerges will be complex and multi-variate.”
4.1.3 Simulation approaches

As always, there is also the possibility to measure a phenomenon to simulate it, rather
than acquiring data in a physical test. This can in principle also be applied to human
workload. In [68] Laughery divides human performance models into reductionist mod-
els and first principle models.
Reductionist models are basically a decomposition of a task from the top down to the
level where estimates for performance and workload per task element can be made. The
classic task analysis and time studies fall into this category. Nowadays computer tools
exist for performing more complex task analysis, called task network analysis, one of
them being Micro Saint [68][69].
First principle models represent the opposite approach, bottom-up, where human be-
haviour i modelled in a framework such as the already described cognitive architectures
(see previous chapter).
None of these approaches matched our needs. We realised that we needed to model
the human operator, not only to determine the simulated workload, but also to control
the machine model in the simulation similar to how an actual human operator would
operate an actual working machine. After all, our initial goal was to use simulation to
assess productivity, energy efficiency and similar operator-influenced properties of
working machines.
Our approach reported in [VI] is simply based on using the operator’s control com-
mands and from those signals calculate a measure that we have shown to correlate well
with self-reported operator workload. After the development of our own solution (de-
scribed in the following chapters), we discovered that Markkula and Engström had
some years earlier reported a different approach leading to similar results [70].
In an interesting study [71], Hughes and Jiang used Micro Saint to analyse human
workload during operation of a hydraulic excavator. In this paper they measure operator
effort as the sum of the time for the operator’s physical movements and the operator’s
cognitive task time foregoing the working cycle. Operator workload is simply calculated
as the ratio of operator effort to cycle time of the machine, which seems to be a less
founded choice. Hughes and Jiang compare the operation of expert and novice operators
using excavators with both hydro-mechanically and electronically controlled hydraulic
systems.
Their findings confirmed the common knowledge that

1. experts in contrast to novices use several hydraulic functions at the same
time; and hence experts work faster (also confirmed in [72] by Bernold)
2. that both experts and novices work faster when using the machine with elec-
tronically controlled hydraulics, in comparison to using the machine with
hydro-mechanically controlled hydraulics
Due to a lack of tables with results for working cycle time etc, we cannot perform
calculations of our own to better understand the results presented. Unfortunately, due to
the way operator effort and operator workload were defined, we have to regard the de-
tailed results, presented in diagrams showing workload as a function of operator skill
and control type of the hydraulic system, as being of little use.
4.2 Empirical study #1

In order to gain input for our work with assessing operability by means of simulation, as
well as measuring operability of physical prototypes, we performed a smaller study in
the form of a Master thesis project [6], theoretically and administratively prepared by us
and conducted by two students, for whom we were industrial supervisors.
Three professional wheel loader operators were asked to operate a wheel loader
which we had re-configured by means of software settings. The adjustments included
changing governed speed of the engine (thus reducing overall maximum engine speed),
re-mapping the hydraulic levers (thus changing sensitivity and overall characteristics),
and setting the engine’s smoke limiter function to a more restrictive setting (thus mak-
ing the engine less responsive to torque and speed requests).
Operator workload was mainly assessed using NASA-TLX in combination with a
semi-structured questionnaire asking for comments. We were also curious as to whether
Kansei Engineering [73], a tool for quantifying emotional customer needs and connect-
ing them to selected product properties in a mathematical way, could be applied to our
problem.
For several reasons, including lack of proper preparation and performance, the
study’s results were less relevant than hoped. For instance, with respect to the applica-
tion of NASA-TLX we experienced the same difficulties due to confusion of the six
sub-scales as reported in [60]. Furthermore, at that time our focus was on finding if pri-
mary-task measures (such as cycle time, bucket load, and fuel consumption) were in-
dicative of operability, as well as making simple statistical analyses of the operator’s
control commands in order to find some correlation with self-reported workload. The
results, or rather the lack of such, made us choose to not report the study publicly in
other ways than making the Master thesis report available [6].
In light of the results from our second study, see next section, we plan to re-analyse
the data set and may report the findings in a future paper.
4.3 Empirical study #2

Learning from the first attempt, we set up a second study, again with the goal to gain
input for our work with assessing operability by means of simulation, as well as measur-
ing operability of physical prototypes (and later perhaps even doing so online for adap-
tive operator assistance). This time, we increased the size of the study and prepared it
more thoroughly, in detail reported in [VI] and [VII].
4.3.1 Setup
Eighteen operators participated, each sorted into one of three skill groups according to
his self-reported skill level (Figure 31).
Figure 31. Sorting of the participating operators into skill groups
Each operator had to test-operate a wheel loader in three different software settings,
in randomised order and under similar conditions.
In order to study the effect of the previously mentioned interdependencies during
bucket filling, the wheel loader software was modified to limit engine speed, and thus
traction force, whenever the transmission’s 1st gear was automatically engaged, which
occurs only during bucket filling (all machines normally start in 2nd gear).
The first limitation of engine speed was chosen so that maximum traction force was
reduced to 62% in 1st gear (see the top left diagram in Figure 32), which resulted in a
maximum engine speed that was comparable to what expert operators on average used
the wheel loader at. This setting also avoided wheel spin.
In the other condition, engine speed was limited such that maximum traction force
was reduced to 47% in 1st gear (see the bottom left diagram in Figure 32). We deliber-
ately used an engine speed limit below what was needed for the working task planned,
which also affected the speed of the hydraulic functions more noticeably. Furthermore,
the maximum obtainable traction force in 1st gear was even lower than what was avail-
able in 2nd gear, all of which an operator would experience as clearly negative.
The 3D diagram to the right in Figure 32 shows that the counter-acting effect of the
traction force on the lifting force is clearly less pronounced for the 62% and 47% set-
tings, and that even at maximum use of the traction force available the lifting force is
never cancelled out completely.
Figure 32. Effect of engine speed limitation on traction force and on resulting lifting force
Each machine version was first tested for ten minutes to reduce potential practice ef-
fects, followed by five minutes of live test-driving.
With the intriguing facts visualised in Figure 11 to Figure 18 in mind, we specifically
wanted to study the wheel loader operator during the bucket filling phase. We therefore
chose to modify the short loading cycle such that the operator was instructed to still go
through most of the motions, but not to use steering and thus empty the bucket at the
same spot where it was filled.
During all sessions various data were recorded off the wheel loader’s CAN bus, en-
hanced by additional data, either calculated or acquired from externally mounted sen-
sors. All tests were also recorded on video using an externally placed digital video cam-
era and later synchronised with the acquired data from the CAN bus.
The operators’ subjective evaluations of the machines’ ease of bucket filling, per-
ceived power and overall ranking, together with self-reports on satisfaction with their
task performance and their personal stress level, were captured on visual analogue
scales with free comments, performed directly after each live test-driving session.
Psychophysiological measurements were also made: finger temperature, heart rate
(via Electrocardiogram, ECG), galvanic skin response (skin conductance) and CO2 con-
centration in the operators’ exhaled air were acquired. See Figure 33 and Figure 34 for
placement of sensors. For a more detailed description see [VII].
Figure 33. Right hand controls and sensor placement on right wrist
Figure 34. Left hand controls and sensor placement on left wrist and fingers
We have mentioned previously that each human being has an individual response to
workload, which requires some sort of reference or calibration. We chose to establish a
Psychophysiological Stress Profile (PSP) according to Table 4, essentially taken from
[74]. The reasoning for this particular design and its application can be found in [VII].
Table 4. Psychophysiological Stress Profile (PSP)
# Designation Observation Description

time
1 Base line 3 min Relaxed silent reading of a neutral text
2 Deep breathing 2 min Deep breathing under guidance, approx. 6 bpm
3 Nonverbal stress 2+2 min Two periods of thinking about a stressful situation,
feedback and guidance in-between
4 Relaxing 2 min Relaxing with closed eyes, normal breathing
5 Math stress 2 min Counting aloud backwards from 2500 in steps of 7
Data from psychophysiological measurements have to be evaluated in context and in

relation to the PSP and each other, which today requires expert knowledge. Further-
more, a reliable algorithm to automatically remove measurement artefacts in data for
ECG and respiration has yet to be developed. In [VII] we simulated the output of an
automatic system by having three of the co-authors (Ohlsson, Ohlsson and von Schéele)
use their expertise in the field to manually review and analyse the measurement data.
They derived an absolute workload index for each operator in each of the conditions
(PSP and three machine variants), without knowledge of the operator’s skill level, his
subjective evaluation and self-reports, and which machine variant (setting of traction
force limitation) corresponded to which condition for this particular operator. After each
of them had analysed the data individually, they then compared their results and agreed
on a consensus workload index for each operator in each condition.
4.3.2 General results

It is interesting to look at the operators’ subjective ranking of the tested machine vari-
ants, especially in light of their operating skill.
1 Machine variant 2 Machine variant 3 Machine variant

100% 62% 47% 100% 62% 47% 100% 62% 47%
Best 6 2 2 Best 5 2 Best 3
Ranking
Ranking
Ranking
Medium 2 5 1 Medium 1 4 Medium 3

Worst 1 2 6 Worst 6 Worst 3
2+3 Machine variant

Skill group Participants
100% 62% 47% Matrices show operators’ added votes for
Best 8 2 1 9 ranking of the machine variants.
Ranking
2 6
Medium 1 7 3 3 Machines could be voted for the same place.
Worst 9 2+3 9
Figure 35. Results of the operators’ machine ranking, skill groups 1, 2, 3, and combined 2+3
Figure 35 shows that the operators in group 3, i.e. the experts, share the clear opinion
that the original machine (100% traction force) is the best, the wheel loader limited to
maximum traction force of 62% next best, and the extremely limited machine worst.
This opinion is predominant even in skill group 2, the experienced operators, but two
had already ranked the 62% variant as same or as better than the 100% machine. The
operators in skill group 1, the inexperienced operators, also follow the same voting pat-
tern to a certain degree, but deviate more often. Two operators in this group actually
ranked the 47% machine as best, and one ranked the 100% variant as worst.
The more detailed evaluation in [VI] shows that a relationship also exists between the
ranking a machine is voted for and the score given by means of a visual analogue scale
for ease of bucket filling and machine power, however somewhat less distinct for opera-
tors in skill group 1.
From the spontaneous comments noted during the study, we know that several opera-
tors were unsure how to rank a machine variant. Several, especially inexperienced op-
erators seem to correlate machine ranking with their impression of the machine’s power,
even though they gave a lower score for ease of bucket filling, presumably blaming
themselves for not being able to utilise the power available. One operator commented
on this and reasoned that bucket filling is an important phase in a loading cycle and that
he will therefore give one particular machine variant a lower rank due to him finding it
more difficult to fill the bucket, even though the machine received the highest score for
power.
4.3.3 Discussion of possible measures for operability

In [VI] we discuss in detail which measurable signals that can also be simulated might
correlate to the operators’ subjective assessments of the tested machine variants’ oper-
ability.
Primary-task measures of the joint man-machine system are seemingly obvious
choices, one being time on task, i.e. how long the operator needs to fill the bucket. But
two simple thought experiments disprove this hypothesis. To begin with, merely by
reducing the speed of the lifting function we could accomplish longer bucket filling
phases. However, this would not mean that the wheel loader would have been harder to
operate; it would just have taken longer to lift the bucket out of the gravel pile until the
transmission can be set to reverse, thus concluding the bucket filling phase. Secondly, it
would mean a difference in operability to a human operator if a bucket can be filled in
for example ten seconds just by activating and holding the levers or if the positions of
the levers require frequent readjustment during the same period of time.
Another seemingly obvious choice would have been task completion, i.e. how much
material the operator manages to load into the bucket. However, we reasoned that up to
a certain point the operator always manages to fill the bucket completely; it is just a
matter of effort. This also invalidates bucket load as a measure of operability.
Another easily acquired measure would have been fuel consumption, which indeed
should correlate fairly well with operability, since it is indicative of how much energy
was required to fill the bucket. However, in the future we also want to compare oper-
ability between different machines. Fuel consumption is impacted to a very large degree
by the component and system losses, which are affected by size and the state of tech-
nology. For instance, fuel consumption for the same size of an engine differs when
comparing versions that comply with different exhaust emission legislations (Tier II vs.
Tier III vs. Tier IV). Also, the drive train can feature similarly strong transmission but
of vastly different technologies (e.g. hydrodynamic or hydrostatic, planetary or counter-
shaft) and therefore different internal power losses. Finally, different hydraulic systems
can be employed that deliver similar power but with different internal losses (e.g. open
centre or load sensing systems, employing fixed or variable displacement pumps). Fuel
consumption during bucket filling is therefore also not a usable measure of a wheel
loader’s operability.
All of the measures mentioned above are merely secondary effects of the operator’s
way of using the wheel loader – and they are also profoundly impacted by the ma-
chine’s system design and the choice of components. After a discussion we therefore
reason in [VI] that a much better way of deriving a measure indicative of operability is
to measure close to the operator.
Focusing on the operator’s three most important control commands during bucket
filling, accelerator pedal, lift lever and tilt lever, we discuss in length both measures in
the time domain and the frequency domain. The former include the operator input dose
suggested in previous papers [II]-[IV], extended from simple variance to skewness and
kurtosis. We also discuss measures of complexity (including using ZIP-compression)
and if they could be applied to the time signals of the operator control commands.
The usefulness of all of the measures mentioned above is argued against in [VI] by
theoretical reasoning first. However, we then still examined most of them more closely,
using the test data to prove the reasoning correct, i.e. that the correlation was indeed
limited or non-existent.
4.3.4 Control effort

The theoretical discussion in [VI] concludes that the most promising approach would be
to generate a measure of the operator control effort by approximating the control com-
mands for accelerator pedal, lift and tilt function with straight lines (Figure 36), count-
ing the number of breakpoints required (n) and combining them according to Equa-
tion 1.
2 2 2
n sum = nacc + nlift + ntilt (1)
Because it is more difficult for humans to control movements with variable speed,
using polynomials of a higher order than one (i.e. lines) is not proposed.
The question arises of what algorithm to use for such an approximation. Since the
work of Markkula and Engström [70] was discovered first later in a discussion with one
of the authors, we decided to develop an algorithm of our own. A colleague informed us
that a similar task is typical within the field of mobile robotics: line extraction from a
cloud of points resulting from a 2D laser scan (see paper [VI] for references and a more
detailed review).
Figure 36. Accelerator pedal signal, measured and approximated
We used our own version of the tried and tested Split and Merge algorithm, but sub-
sequently developed a new algorithm anyway, which we dubbed Stretch and Relax
(Algorithm 1). Both algorithms gave similar results, but Stretch and Relax - Absolute
was slightly faster in our tests than our version of Split and Merge.
Algorithm 1. Stretch and Relax – Absolute (SR-A)
# Description
1 Start with first data point (S1)
2 Use next point as break point candidate BP1
3 Fit line L1 between S1 and BP1 with least squares method
4 Use BP1 as preliminary start point S2 for next line
5 Use next point as future break point candidate BP2
7 Repeat steps 5-6 until residual sum of squares for L2 exceeds pre-
defined threshold or until end of data set is reached
8 If index of BP2 ≥ BP2max then BP1max = BP1 and BP2max = BP2
10 Use BP1max as break point for line L1 and as new start point S1 for
next line segment L1
11 Repeat steps 2-10 until end of data set is reached
Using the algorithm and equation above, we calculated control effort values for all
operators in all conditions and compared to the operators’ subjective evaluations. Figure
37 shows that a strong monotone relationship between control effort and the operators’
subjective evaluation of the ease of bucket filling could be found in five cases, all of
them being operators with higher skills. In five more cases, most of them operators with
medium skills, we were able to find at least a partial monotone relationship, meaning
that the control effort data for the machine ranked in-between highest and lowest ease of
bucket filling were not significantly different from one of their neighbours, but other-
wise the same relationship could be identified, i.e. higher ranking of a machine and its
ease of bucket filling co-occur with lower calculated values for control effort.
We can thus conclude that of all nine experienced and expert operators, seven (78%)
show a monotone correlation between control effort and ease of bucket filling, i.e. the
easier the operator reported the bucket filling had been, the less control effort we calcu-
lated. Of the nine inexperienced operators, only three (33%) show such a correlation. As
speculated earlier in paper [VI], there seems to be a clear relation to the individual skill
level. Not combining skill group 2 (experienced) and skill group 3 (experts), the figures
from groups 1, 2 and 3 are 33%, 67% and 100%, respectively.
Strong monotone relationship between 27 22

control commands and self-assessed 28
ease of bucket fill or machine ranking 14 25
Partial monotone relationship 26 20

(e.g. only between best and 29
worst machine or only for machine rank 08 06
but not for ease of bucket filling)
No monotone relationship
11
18
04 13 21
12 02
Operator skill 05
(self-assessed)
0 25 50 75 100
Skill group 0 Skill group 1 Skill group 2 Skill group 3

0 participants 9 participants 6 participants 3 participants
Figure 37. Existence of a relationship between operator control effort and ease of bucket filling
We chose to look for a relationship between control effort and primarily ease of
bucket filling, secondarily to machine rank, because these are most closely connected to
operability. In some cases control effort shows a clearer monotone relationship with
machine power, but not so in other. We were also curious as to whether we could see a
clearer relationship between control effort and traction force, which would show that
several operators were making erroneous subjective assessments, yet their bodies re-
sponded “correctly”. Paper [VI] shows that this does not seem to be the case to any
great extent.
In summary, this provides us with a tool to evaluate test data not only from physical
testing, but also from virtual testing by using the control commands of an operator
model similar to the ones presented in [II] and [III]. It needs to be pointed out that the
findings of this study are not limited to wheel loaders, but should be generally applica-
ble, in particular for working machines of similar complexity.
4.3.5 Psychophysiological measurements

While the results with regard to control effort are presented in [VI], the psychophysi-
ological part of our second empirical study is presented in [VII].
Table 5 shows the experts’ consensus on the operators’ workload during the PSP and
the three test conditions. To the right this table, sorted by operator skill level, also shows
in which individual order the machine variants with their respective traction force set-
ting were tested by the operators.
Table 5. Results: workload index and test order of variants
With the exception of one, all operators show a higher workload during the Psycho-
physiological Stress Profile than during the first machine test. This can assumed to be
due to the novelty of having to go through the PSP tests with various sensors attached.
Comparing the different conditions, i.e. machines with different traction force settings,
it also shows that the workload does not vary much within each subject, despite the op-
erators clearly commenting on experiencing the machine variants differently.
No clear overall trend emerges when comparing the operators’ responses in the
questionnaire on how well they felt they had managed (“personal satisfaction”) and how
stress-free they felt during the operation with the experts’ consensus on the operators’
workload. See the tables in paper [VII] for these and also the operators’ responses re-
lated to the tested machines, evaluated for their ease of bucket filling, overall impres-
sion of power and comparative rank. Again, no clear trend is immediately obvious.
In conclusion, the workload index, derived from the psychophysiological measure-
ments, mostly shows no variation with any of the independent or other dependent vari-
ables.
With all recorded data available, it was tempting to look for similarities in perform-
ance of the human operator (i.e. the workload) and the performance of the machine he
controls (even though we have reasoned earlier that no general causality can be identi-
fied). In [VII] we looked at cycle productivity (e.g. expressed in ton/hour), cycle effi-
ciency (e.g. expressed in ton/unit of energy), as well as average bucket load, bucket fill-
ing time and energy required during bucket filling.
Again, no clear trend could be identified. For several operators the workload index
is approximately constant even though the machine performance data vary. For others it
is the opposite.
During the study, the test leader instructed the operators to keep their pace and try to
fill the bucket completely each time and in each condition. Here we have conflicting
requirements, since the “47%” machine variant with severely limited traction force sim-
ply does not permit a bucket to be filled as fast and as much as with the unmodified
“100%” machine. Some operators chose to keep the bucket fill factor constant, which
led them to spend more time in the gravel pile. Others chose to keep pace with the con-
sequence of less material in the bucket. This trade-off is visualised and elaborated on in
[VII].
It was peculiar to see that the four operators with the highest skill levels showed
constant high workload for all machines, not affected by traction force limitation. This
is presumably a ceiling effect. It was also interesting to see that while three of these
operators accepted that the machines with limited traction force settings do not easily
permit as high bucket loads, one operator worked hard to achieve this – and in turn sac-
rificed productivity by spending too much time on filling the bucket, especially when
operating the severely limited “47%” machine. We also saw a slightly higher workload
in this condition.
Apart from these observations on an individual level it was hard to see any general
trend. In addition to productivity we also examined more closely the relationship of the
operators’ workload to fuel consumption and energy efficiency of both bucket filling
and the complete cycles – again, no clear trend emerged.
As a final analysis we compared the workload index of all operators together with
their performance and their individual skill level. Figure 38 shows all values for opera-
tor workload and all mean values for bucket load, bucket fill time, bucket fill productiv-
ity, cycle time and cycle productivity for all three measured conditions (traction force
settings) over operator skill level.
In the diagrams to the left in Figure 38, the lowest values per operator are marked
(the other values marked on the vertical line belong to the two conditions that did not
lead to the overall lowest mean value). Note that the vertical lines do not show standard
deviation, but simply connect the three values per operator. In the diagrams to the right,
the highest values per operator are marked. Note that there is no correspondence be-
tween diagrams, i.e. the marked lowest values for bucket load for example do not corre-
spond to the marked lowest values for bucket fill time or operator workload (for such
visualisation see paper [VII] section 6.2).
Figure 38. Lowest and highest values for workload and performance (mean values) over skill
(no clear trend is immediately obvious, see text for observations made).
Analysing the diagrams in Figure 38 it appears that, except for a few individuals, the
lowest workload of an operator generally seems to increase with skill (top left diagram),
which was unexpected, because it means that the inexperienced operators in general
have a lower workload. These individuals also show lower performance in comparison,
but one hypothesis was that inexperienced operators would be stressed by too many
other factors (one example being noticeably reduced stability of the machine with
loaded bucket in top position) and would thus have a very high workload in all condi-
tions. Clearly, the data we gathered do not support this hypothesis.
Looking at highest workload per operator, the aforementioned ceiling effect around
a workload index of 4.0 is apparent. The expert operators in skill group 3 showed high
workload already in all conditions, while the experienced individuals belonging to skill
group 2 seem more able to relax, while being able to perform almost as well as the ex-
perts.
Another, but expected general trend can be seen for the mean value of the bucket
load, both lowest and highest (second diagram from top, left and right side), which
seems to increase with skill. Again, note that there is no correspondence between the
diagrams. From the analysis in [VII] we know that there is no general trend between
bucket load and workload on an individual level (within subjects), even though both
seem to increase with skill (between subjects).
As had been expected, the greatest variations in performance can be seen for the in-
experienced operators in skill group 1. It was, however, unexpected to find out that the
expert operators in group 3 focussed so intensely on filling the bucket even with the
traction force being severely limited, to the extent that they spent comparatively more
time doing so and thus sacrificed productivity.
It is interesting to note that none of the operators were aware of their high workload.
In their opinion, the work performed was little demanding, especially in comparison to
other possible tasks such as loading of shot rock or digging in clay.
4.3.6 Summary
In summary, our successful work with deriving a measure for operator control effort
provides us with a tool to evaluate test data not only from physical testing, but also from
virtual testing by using the control commands of an operator model similar to the ones
presented in [II] and [III]. These findings of our study are not limited to wheel loaders,
but should be generally applicable, in particular for working machines of similar com-
plexity.
With regard to the psychophysiological measurements the picture is less clear. Even
though we were able to make some interesting observations at both individual and skill
group level, the results of this study indicate that measures other than ones based on
heart rate, finger temperature, skin conductance and respiratory air flow should be con-
sidered. Traditionally, such measurements are performed in the stationary, stable and
clean environment of a clinic, rather than in a mobile environment, exposed to large and
frequent accelerations. This has also been discussed in for example [65], but we believe
that all of the above mentioned parameters are even harder to measure reliably in the
tough environment of a mobile working machine.
However, these measurements will be less prone to artefacts in the controlled envi-
ronment of a human-in-the-loop simulator. This of course requires the use of a simulator
that has both the fidelity required to produce correct results on the machine side and the
immersiveness required to produce correct results on the operator side. The research
summarised in [62] shows that this is possible in principle.
In a simulator as well as in reality it is important to provide relevant test scenarios.
We suspect that the operators in our study did not experience enough pressure, because
there were no immediate consequences for not keeping the set cycle pace or not filling
the bucket completely [VII]. In future we will try to add this pressure by either more
closely representing a real working place or by adding productivity-related feedback
artificially (e.g. by visual or acoustic means).
We also found that the accuracy of the psychophysiological measurements con-
ducted appears to be too low for our requirements. On the other hand, promising re-
search is being done into the use of Artificial Intelligence techniques (such as Case-
Based Reasoning) to evaluate data from psychophysiological measurements
[74][75][76]. With proper input, perhaps such systems can provide the accuracy we
require and might then also be able to recognise measurement artefacts. The need to
automatically analyse all acquired data in context is the major driver for such AI-based
analysis, since a manual review of multiple channels with data from interacting and
compensating systems can today only be performed by experts.
Work is also being done concerning development of new wireless sensors, e.g. [77],
that can be deployed without distracting the operator from the working task, while at the
same time being able to acquire data without the amount of measuring artefacts we have
to cope with today.
5
Adaptive aid
T HIS TOPIC was not explicitly part of our research, other than being considered the
logical continuation of our work with the concept of machine harmony and its evolution
(Figure 8). Providing workload-adaptive assistance to the operator is certainly an excit-
ing prospect to pursue.
Due to a lack of published research on our ideas for enhancing operability, some of
the more simple ones are described in [78][79][80][81], we would at least like to pro-
vide and discuss some literature references and speculate on future work in our field of
application.
5.1 General
Considering operator workload as a sort of cost value, we see an analogy to the Eco-
nomic Principle, which describes rational decisions in situations of scarcity. The ex-
treme choices are represented by the Minimum Principle (minimise cost with regard to a
given goal) and the Maximum Principle (maximise utility for a given level of cost). In
reality, the desired situation will be found somewhere along the Pareto front, with the
Minimum Principle solution and the Maximum Principle solution at opposite ends.
By analogy, our hypothesis is that the operator minimises workload for a required
level of productivity (i.e. when the working machine is part of a production chain). This
optimisation is presumed to be done by trial and error. When no level of productivity
needs to be matched, the operator uses the machine at a pace that is neither too fast nor
too slow. However, this level might not be optimal with respect to the machine’s energy
efficiency. With the help of adaptive operator assistance functions, we might be able to
guide the operator towards a better balance between workload, energy efficiency and
productivity.
However, this thinking might be naïve in light of the self-reinforcing complexity
circle Hollnagel and Woods rather bleakly describe in [1]: “The growing technology
potential is invariably seized upon and exploited to meet performance goals or effi-
ciency pressures. This is referred to as the Law of Stretched Systems, originally sug-
gested by Lawrence Hirschhorn: Under resource pressure, the benefits of change are
taken in increased productivity, pushing the system back to the edge of the performance
envelope.” This was stated in a context of discussing system complexity leading to op-
portunities for malfunctions and accidents, but operator workload is certainly one com-
ponent.
Also Summala notes in [67] with respect to car driving that “motivational factors
inherent in driving result in a built-in tendency to go to the limit – or a little below (...).”
He continues with “subjective (and even objective) time benefits and fun can still be
drawn from faster and more aggressive driving, tendency to complete a trip may lead to
driving while too drowsy, and increased time-sharing easily follows from automacity of
driver behavior and automation of the vehicle and highway system.”
There is thus a risk that increased automation in working machines might not bring
all the benefits we imagine for operator workload. In our studies we have seen tenden-
cies towards this behaviour, visualised for example in Figure 21. On the other hand,
with the operators of mobile working machines being subjected to as high workloads as
today, any sufficiently validated automation must be seen as an improvement, e.g. as
presented in [82] for agricultural tractors, in [83][84] for tree harvesters, in [84] for in-
dustrial cranes, and in [85] for excavators and similar machines.
5.2 Implementation of adaptive aid

Making any automatic function workload-adaptive has to be valued against confusion
due to reduced predictability. This requires a consistent behaviour, which seems to con-
tradict the very essence of adaptive aid: supporting the human operator depending on
the current workload. A solution might be to offer aid at a few, distinct levels and in-
form the operator clearly on which level is enabled and what type of support will be
given at that level. Once a certain support level has been activated, it should not be de-
activated for a foreseeable period of time. It also seems appropriate to let the operator
explicitly accept or deny this assistance, rather than just activating or deactivating it
automatically. On the other hand, this forced operator input increases workload and
therefore reduces the benefits of the automation itself.
A good example of how not to offer automatic help is Microsoft’s infamous Office
assistant Clippy. This function frequently interrupted the computer user with tips and
offers of interactive support, often not only unwelcome but also not related to the actual
task in progress. Many users reacted by always automatically declining the Office assis-
tant or deactivating it altogether – and thus missed the opportunities where this auto-
matic function could have led to an actual increase in productivity.
Before deciding upon any automation, it has to be carefully considered which func-
tions shall be performed by the machine, and which shall be performed by the human
operator. This is called function allocation. Giving adaptive assistance to the operator,
rather than always automate a certain function in a certain way, means that the classic
field of function allocation has to be expanded. Several publications can be found on the
topic of adaptive function allocation, e.g. [86] and [87].
In order to create trust in an automatic function, i.e. establishing the belief that by
using the automation e.g. productivity or safety will be increased or e.g. workload or
Adaptive aid 73
fuel consumption will be decreased, it needs to be transparent to the operator. Ideally,

the operator understands exactly the algorithm, scope and limitations of any automatic
routine. However, due to the complexity of such functions, an understanding this deep
will hardly ever be the case. According to [89], trust in automation guides people to rely
on it in uncertain situations, even when the complexity of the automation makes a com-
plete understanding impractical. But there is the risk that the operator either overtrusts
the automation and therefore misuses it, or the operator undertrusts the automation and
therefore disuses it, as discussed in [88]-[91]. Both misuse and disuse are examples of
inappropriate reliance on automation. Misuse refers to failures that occur when people
inadvertently violate critical assumptions made by the developer of the automation used,
while disuse refers to failures that occur when people reject the capabilities of automa-
tion [88]-[90]. An appropriate use of a system is facilitated by calibrated trust, i.e. trust
that matches the system capabilities. As [89] argues, this could actually require an
automatic function to be made more simple, rather than more complex.
Visser and Parasuraman summarise in [92] that the challenge in the design of adap-
tive automation is to determine what, when and how much to automate. With respect to
what to automate, they refer to four types of automation based on a human information-
processing model [93]:
- information acquisition
- information analysis
- decision making
- decision / action implementation.
The research reported in [94] indicates that human operators seem to cope better
with adaptive automation provided in information acquisition and decision / action im-
plementation, rather than in information analysis and decision making.
With respect to when to automate, Visser and Parasureman refer to three major cate-
gories in [92]:
- critical events
- measurement-based invocations
- model-based invocations
Many examples can be found in the literature. In the following section we will
briefly look at some specific ideas in our field of application.
5.3 Applications for adaptive aid in working machines

Adaptive aid in general has to cope with the problem that operator workload will natu-
rally be higher in certain tasks. It is thus necessary to combine workload measurements
with means to detect the current task and sub-task. Adaptive aid should only be offered
when the measured workload is significantly higher than what is normal for the work in
progress (or higher than what has been declared an absolute threshold), or when the
measured performance of the operator is lower than expected.
Navigation and positioning of both machine and tool are already subject to automa-
tion, either already available as commercial products [82] or demonstrated in applied
research [83][84][85]. As discussed before, making any automatic function workload-
adaptive has to be valued against confusion due to reduced predictability. In this regard,
navigation and positioning cannot be considered candidates for adaptive aiding.
Considering the operability challenges described in chapter 2, we find some automa-
tion opportunities related to the control of multiple functions during bucket filling, re-
versing and bucket emptying. The question again is whether workload-adaptivity is re-
quired at all. This might be the case in bucket filling, where both a balance of forces and
a balance of power must be maintained by the operator. An experienced operator might
want to have unlimited control over drive train and hydraulic functions, which easily
proves too much for a novice. In such cases, limiting individual functions in order to
improve balance within the system might help. The idea of remapping the characteris-
tics of operator controls like accelerator pedal and hydraulic levers, described in
[80][81], as well as the traction force limitation employed in [VI][VII] can all be made
adaptive to the operator’s workload.
Considering Figure 6 and Figure 16, the idea of adding external power (torque) to
the system’s summation point seems obvious [78]. This could be provided by either an
electric or a hydraulic motor, but also by purely mechanical means. Such add-on sys-
tems do not need to be full-fledged hybrid systems like the ones discussed in [V], but
this would certainly make sense. The extra power (torque) could be administered in
relation to the operator’s need for help with balancing the system, and not necessarily
always be available as a power boost option.
More ideas can be found in [95].
Finally, we can also think of other ways to provide the operator with the information
necessary for the task. In [31], MacAdam compiles a list of the primary sensory chan-
nels used in driving (importance according to the order of listing):
- vision
- motion cures from the vestibular (inner ear) and kinesthetic (body distrib-
uted) channels
- tactile
- auditory (which he considers a supplementary and reinforcing cue for the
visual channel, especially during high workload conditions, but acknowl-
edges that for tracking control tasks, some studies show human time delays
that were as short for auditory as for visual channels)
Considering for example the studies reported in [96] and [97], we suggest the use of
auditory feedback for guiding the operator towards more productive or energy-efficient
use of the working machine [79]. From our own experience with using acoustic feed-
back in sailplanes we know that the auditory channel can be employed for rather precise
tracking tasks, reducing the visual workload. However, the warning of drowning other
auditory alerts, raised in [97], needs to be considered.
Adaptive aid 75
5.4 Real-time assessment of operator workload

Wilson and Russell reported in [99] on the successful use of artificial neural networks
(ANN) to determine the workload of operators of uninhabited aerial vehicles (UAV) in
real-time. The ANN used data acquired by EEG, ECG and EOG (an earlier version [98]
also included analysis of respiration) to determine each participating operator’s mental
workload. Above a certain threshold adaptive aid was given (or rather simulated by re-
ducing the difficulty of the simulated mission).
The question is if this is a feasible way outside of the military world. Certainly, the
equipment and the individual pre-application training required for both operator and the
ANN are prohibitively costly for use in an ordinary commercial vehicle or working ma-
chine.
We are aware of research projects in which Case-based Reasoning is used for auto-
matic analysis of psychophysiological measurements [75][76], but these projects are at
the moment focused on finger temperature measurements, which we have found diffi-
cult to make properly in a working machine.
In a recent publication [100] Haufe et al. reported having used EMG and EEG to
predict upcoming emergency braking during car driving in a simulator. By means of
EMG, muscle tension in the lower leg has been detected before any movement towards
the brake pedal has even started (and thus long before the brake pedal sensor can read
any emergency activation). The EEG readings predicted emergency braking even prior
to EMG, but these measurements have in discussions been found to be more sensitive to
artefacts. In contrast to the work of Wilson and Russell, no specifically trained ANN
was employed in the study by Haufe et al. This is a significant improvement, even
though estimating human workload from EEG data is still in its infancy and far away
from being generally usable [101]. The authors of [100] discuss whether an EEG elec-
trode cap is practical to wear in everyday life, but acknowledge the progress made and
ongoing development in that area. From our perspective, it seems to be easier to intro-
duce such methods for professionals first. For instance, given further development of
hardware, software and our general understanding of neurophysiology, an electrode cap
could be integrated into the safety helmet worn by a working machine operator.
Another interesting prospect is to determine workload based on the operator’s eye
movement, as reported in [102].
6
Review of
appended papers
T HE APPENDED PAPERS contribute with answers to the previously formulated re-

search questions as visualised in Figure 39 below:
Figure 39. Contribution of the appended papers to the formulated research questions
(bold black lines = explicit focus, thin grey lines = implicitly covered or side discussion)
As we can see, each research question has been covered by at least one paper that ex-
plicitly focuses on that question (bold black connections). Often a paper also discusses
another research question or implicitly covers one without going into detail (thin grey
connections in Figure 39).
The research questions and corresponding hypotheses, as formulated at the beginning
of this thesis, will be reviewed in the next chapter with regard to what answers have
been provided and what scientific contribution we have made to the field. In the follow-
ing we will briefly summarise and comment each paper.
Paper I
Using Dynamic Simulation in the Development
of Construction Machinery
The paper essentially describes the technical problem and the academic need at the be-
ginning of our research, which was then very much focussed on simulation in the con-
cept design phase of product development. The paper gives an overview of the research
project, and explores the challenges of the wheel loader application. The results of a
literature review are listed, and future research possibilities are examined together with
possible solutions. Reflections are made upon the role of dynamic simulation in the
product development process, and upon simulation techniques and methods in general.
A revised product development process with regard to the research topic is proposed.
The main contribution of this first paper is to actually define our research. It also
contributes with answers to research questions RQ1 and RQ5.
All research leading to this paper and the writing of the paper itself are the work of
the main author. The paper’s co-author provided valuable feedback and encouragement.
Paper II
Dynamic Simulation of Construction Machinery:
Towards an Operator Model
A first version of a rule-based operator model is developed. With the focus on the
bucket filling, one specific technique of human operators is described in detail. The op-
erator model consists of min/max relationships reminiscent of fuzzy-sets and is shown
to work well. Its ability to reproduce human adaptation to basic variations in machine
capability is demonstrated. Besides this, reflections are made upon modular simulation
and level of detail, power transfer in a wheel loader is examined, and the question of
quantification of operability is raised.
The main contribution of this paper lies in the description and successful modelling
of one bucket filling technique used by human operators. Research question RQ3 is
clearly in focus, but also RQ1 and RQ2 are reflected upon.
All research leading to this paper and the writing of the paper itself are the work of
the main author. The framework and the machine model used in the simulations are a
collaborative work by the main author and two fellow engineers, one of them being the
paper’s first co-author, Allan Ericsson, who also did most of the actual development of
the operator model. Both co-authors of this paper provided valuable feedback during
research and writing.
Review of appended papers 79
Paper III
An Event-driven Operator Model for
Dynamic Simulation of Construction Machinery
The paper gives a thorough description of a wheel loader’s short loading cycle. A sec-
ond operator model is developed, which is realised as a finite state machine (a pure dis-
crete-event approach) and focuses on the remaining elements of a short loading cycle.
Neither bucket filling nor the phases after bucket emptying are included. It is demon-
strated that this operator model is able to adapt to basic variations in workplace setup
and machine capability. Originally presented in [VIII], a publication in Swedish, the
Machine harmony diagram is re-introduced as a way to indirectly visualise important
aspects of operability.
The main contribution of this paper is in the description and successful modelling of
the short loading cycle. Research question RQ3 is clearly in focus, with some reflec-
tions on RQ1.
Paper IV
A Methodology for Modeling the Influence of Construction Machinery
Operators on Productivity and Fuel Consumption
This paper essentially wraps up our work with operator models by presenting a unified
methodology for the development of operator models. It explains some phases of the
short loading cycle in more detail, both how the interviewed operators describe them
and how they are modelled in [II] and [III].
The main contribution of this paper is in the more detailed explanation of how the
previously presented operator models were developed. It thus focuses on research ques-
tion RQ3, but also gives some thought to RQ1 and RQ2.
Paper V
Hybrid Power Systems for Construction Machinery:
Aspects of System Design and Operability of Wheel Loaders
This paper marks our departure towards new research questions. Some thoughts on clas-
sification of hybrid systems for working machines, originally published in [X], are re-
introduced. However, for the main part we focus on operability aspects in various
phases of the short loading cycle and how the situation could be improved by non-
conventional systems. The catchy title probably helped the paper to get accepted, but
the topic was not as much a nod to the Zeitgeist as one would think, but was more a
reflection of our practical work at that time.
The main contribution of this paper for the broader audience is presumed to be in the
systematic approach to classifying hybrid working machines. For the more narrowly
defined audience, the description of the operability aspects of wheel loader is probably
of greater interest. For us, the main contribution of this paper is in breaking our previous
focus on simulation. The paper focuses strongly on research question RQ5, but also
reflects on RQ4 and RQ6.
Paper VI
Study of a Method for Assessing Operability of Working Machines
in Physical and Virtual Testing
This paper marks our return to an earlier vision and research question: to quantify oper-
ability and to be able to do so also in simulation. In a quite substantial study we collect
evidence for the correctness of hypothesis H2, i.e. that we can use the operator’s control
commands to calculate a measure for control effort, which in turn correlates well with
the operator’s subjective assessment of the operability of the machine in use. We pre-
sent how such a measure of control effort can be derived, and perhaps just as important,
we also disprove some old ideas on measures of operability. We restricted the study to
the bucket filling phase, due to its prominent role in the short loading cycle of wheel
loaders, but it is reasonable to assume that our findings are applicable also to other
working cycles and other working machines of similar complexity.
The setup of the study and recruitment of the operators is described in detail, which
will help in future undertakings of similar proportions. We also report interesting find-
ings regarding limiting traction force by limiting engine speed in 1st gear, which in the
future might even become the seed of an adaptive operator assistance function.
The main contribution of this paper is in closing the loop and proving that the opera-
tor’s control commands can be used to derive a comparative measure of a machine’s
operability. It thus contributes mostly in answering research questions RQ2 and RQ4,
however with an implicit contribution also to RQ5. Whether the presented Stretch and
Relax algorithm really is novel or optimal is not important in this context and therefore
left for future work.
Paper VII
A Case Study on Quantifying the Workload of Working Machine
Operators by Means of Psychophysiological Measurements
This paper presents the second half of the study, first reported on in [VI]. The focus here
is on trying to find a correlation between the operators’ workload, assessed through self-
reports and psychophysiological measurements, evaluated by experts in the field and
visualised in the form of a workload index. We also examine whether recorded machine
Review of appended papers 81
performance data might correlate with the derived workload index. While interesting
observations could be made on both individual and skill group level, we were not able
to show that the chosen psychophysiological measurements can be used as intended, i.e.
to assess operability of machine in later phases of the product development process or
even online in real time.
The main contribution of this paper is in the study’s detailed description and critical
discussion, which will be of help in successive studies in the future. Traditionally, such
measurements are performed in the stationary, stable and clean environment of a clinic,
rather than in a mobile off-road environment, exposed to large and frequent accelera-
tions. It should be seen as a starting point for future investigations, rather than as a con-
clusion of our previous research.
This paper is mostly concerned with research questions RQ4 and RQ6, but with a
certain contribution also to RQ5.
All research and the practical work leading to this paper and most of the writing of
the paper itself are the work of the main author. The co-authors contributed with text on
the scientific background to the psychophysiological measurements and by performing a
blind evaluation of the measurement results, resulting in a quantified workload level for
each operator and condition. The main author performed all other analysis tasks. All co-
authors of this paper provided valuable help, education and feedback during research
and writing.
7
Scientific
contributions
THE DIFFERENT WORKING AREAS of our research

- Wheel loaders as complex systems
(representative of all similar working machines)
- Simulation of the total system (with a focus on operator models)
- Real working cycles and physical testing
- Adaptive operator assistance (giving motivation for further research),
all have operability aspects in focus and consist of sub-areas that we related to each
other in our work as shown in Figure 40:
Figure 40. Mindmap of the four major working areas of our research
The main scientific contributions of this thesis, as found in the appended papers and the
previous chapters, can be summarised as follows:
SC1. Examination of the wheel loader as a complex system with operability chal-
lenges (in part a descriptive research approach).
SC2. Critical review of how simulation is used in the product development proc-
ess of working machines and description of why the simulation of operator-
influenced properties like productivity and energy efficiency must include
operator models.
SC3. Described methodology for developing discrete event-based operator mod-
els that can adapt to basic variations in machine characteristics and work-
place setup, and demonstrated that by implementing such operator models,
dynamic simulation can be used to evaluate operator-influenced machine
properties such as productivity and energy efficiency.
SC4. Development of the operator control effort measure for assessing operabil-
ity of a working machine. This measure is either calculated from the re-
corded control commands of a human operator (for evaluation in later
stages of the product development process) or from the control commands
generated by an operator model such as developed by us (which would be
performed in the concept design phase, early in the product development
process).
SC5. A first approach to the use of psychophysiological measurements to assess
a working machine’s operability, both in later stages of the product devel-
opment process, but also online for adaptive operator assistance functions.
In all cases the wheel loader was used as an example of a complex working machine,
but the results obtained are not restricted to these machines. Rather, they are applicable
to all working machines of similar complexity, to be found not only in construction but
also agriculture, forestry, mining, and other fields.
Some of the minor scientific contributions are:
- Description of short loading cycles and load & carry cycles and dividing
into phases (Figure 4 and Figure 5, frequently quoted in other publications)
- Visualisation of power transfer and control in an easy to understand scheme
(Figure 6 and Figure 16, and earlier variants thereof) and connection to the
Machine harmony pyramid (Figure 7 and Figure 8)
- Presentation of the Machine harmony diagram (Figure 9 and Figure 10)
- Presentation of power distribution in short loading cycle (Figure 11 and
Figure 13 have been quoted several times)
- Detailed description of one theory of bucket filling (see [II])
- Systematic approach to classifying hybrid working machines (see [V])
- Development of the Stretch and Relax algorithm for approximating time-
domain signals using straight line graphs
Scientific contributions 85
- Description of the setup of the study presented in [VI] and [VII], so that it
can be used as a template for future work
We can now revisit the research questions formulated at the beginning of this thesis, and
answer them:
RQ. How to quantify operability of working machines in all phases of the de-
velopment process?
A. Essentially, we have proven our main hypothesis correct, i.e. we can assess
operability comparatively by comparing calculated or measured operator
workload for different machines. Higher workload means lower operability,
and vice versa. The operator control effort measure was developed using
recorded operator control commands during live testing of a physical ma-
chine (such as the prototypes available later in the product development
process), but control effort can just as well be calculated from the control
commands generated by the operator models developed by us (which would
be useful in the concept design phase, early in the product development
process).
The answers to the previously formulated sub-questions provide a more detailed picture:
RQ1. How to simulate productivity, energy efficiency and similar operator-
influenced properties of working machines, where the human operator is
essential for the performance of the total system?
A1. The hypothesis H1 has been proven correct, i.e. by extending the simulation
models to not only cover the necessary machine sub-systems, but also in-
clude models of the working environment and the human operator, qualita-
tively better answers regarding the above mentioned complex product prop-
erties can be found. Initially there is no need to model the human operator
in detail; sufficiently good results can be obtained by implementing princi-
pal operator decision models and strategies.
RQ2. How can operator models be used to also evaluate a machine’s operability
in dynamic simulations?
A2. The operator control effort measure proved hypothesis H2 correct, i.e. the
way the human operator uses the controls at his/her disposal is indicative of
the operator’s workload and thus of the machine’s operability. Using re-
corded data from real working cycles, a measure of the operator’s control
effort can be found that correlates well with the operator’s subjective as-
sessment of the operability of the machine in use. This measure can also be
calculated from the command signals generated by an operator model in a
dynamic simulation.
RQ3. What methodology to use to derive operator models for dynamic simulation
that adaptively model human-machine interaction in working machines in a
transparent and flexible way?
A3. We have proven the correctness of hypothesis H3 by both investigating

path tracking and developing rule-based operator models, using rules ex-
tracted from interview studies with professional operators.
RQ4. How can operability and operator workload be assessed in later phases of
the product development process, which today relies on subjective evalua-
tions of physical prototypes by professional test operators?
A4. The first half of hypothesis H4 was correct, i.e. we can use the developed
measure for operator control effort. However, we were not able to success-
fully demonstrate the use of psychophysiological measurements. In H4 we
speculated on the future use of these measurements in testing on human-in-
the-loop simulators, which indeed seems easier to realise, due to a more
stable environment.
RQ5. How is operability affected by system architecture and are there ways to
minimise operator workload and thus enhance operability?
A5. In both this thesis and some of the appended papers we describe in essence
what has already been formulated in H5, i.e. the machine harmony pyramid
with the operator on top and the machine’s hardware as the base as a good
analogy of how operability is affected by system architecture. We also rea-
soned on the opportunities in new human-machine interfaces, workload-
adaptive operator assistance functions, and automation of repeatedly used
sequences of actions and movements that today are manually controlled by
the operator.
RQ6. How can operator workload be quantified such that workload-adaptive op-
erator assistance functions can be realised?
A6. Hypothesis H6 read “Provided techniques exist for measuring with little or
no artefacts and methods for evaluating the measurements on-line and in
real time, the measurement of psychophysiological data proposed in H5,
e.g. heart rate, heart rate variability, finger temperature and respiratory rate
(as well as several others) will meet the requirements.” We have not been
able to demonstrate this, to a certain extent due to a lack of techniques for
measuring artefact-free and methods for evaluating the measurements
online and in real time.
8
Conclusions and
outlook
IN OUR RESEARCH we have examined wheel loaders as complex systems with cer-
tain challenges with respect to operability. We have shown that the human operator
plays an important role in the control of the combined system, and that we therefore,
too, need operator models in order to perform simulation on the combined system’s
behaviour, even when interest is limited to engineering issues with regard to the ma-
chine itself and the joint cognitive system perspective is disregarded.
It has been shown that rule-based operator models are capable of introducing “a hu-
man element” into dynamic simulation of working machines, thus giving more relevant
answers with regard to operator-influenced total system properties like productivity and
energy efficiency. A method for developing such operator models has also been pre-
sented.
We have demonstrated that the operator’s control commands can be used to derive a
measure of operator control effort, which in the presented study correlates well with the
operator’s self-reported workload with respect to ease of bucket filling and machine
operability. This measure of control effort can be calculated both from control com-
mands recorded during testing of a physical machine by a human operator, or from con-
trol commands generated in a simulation by an operator model such as the two pre-
sented by us.
A next step can be to try to repeat the study presented in [VI] by means of simula-
tion. This would require extending our operator models with different skill levels. The
question is how skill manifests itself, what separates an experienced operator from an
inexperienced? There may be several aspects to this, for example bucket filling style,
ability to control several functions in parallel, repeatability, tactical questions and so on.
It is confirmed in [71] and [72] that expert operators use several functions at the same
time and generally carry out bucket motions more smoothly than novices.
Modelling the operator of a working machine in a cognitive architecture such as
ACT-R is another tempting idea.
Given machine and environment models of sufficient accuracy and computational ef-
ficiency, it would also be of interest to run optimisations. For instance, what is the most
energy-efficient style for bucket filling, when the complete cycle is considered? Can
today’s accepted layout of a short loading cycle be tweaked to make it more efficient? It
would also be of interest to apply methods from Probabilistic Design, e.g. Monte Carlo
simulations to assess the effect of parameter variations due to tolerances in the total
system, i.e. machine, environment and operator. Ultimately, it would be of great value
to investigate the trade-offs between machine productivity, its energy-efficiency and the
induced operator workload by means of simulation already in the concept design phase
of a new product – but also online in the machine, guiding an operator towards the de-
sired balance by using adaptive aid.
With respect to the latter, we have presented the concept of machine harmony and its
evolution. Providing automatic assistance to the operator is certainly an exciting pros-
pect to continue with, for instance with the rather simple ideas sketched in [78]-[81] or
the method for limiting traction force employed in [VI] and [VII].
At the moment we have no robust and validated method for assessing operator work-
load online and in real time, which prevents us from making such assistance functions
workload-adaptive. However, some of our co-authors in [VII] together with others are
actively pursuing Artificial Intelligence-based ways to automatically evaluate psycho-
physiological measurements, which might also increase the accuracy of the measures
and are robust with respect to measurement artefacts.
Furthermore, development work is being done on new wireless sensors that can be
deployed without distracting the operator from the work task, while at the same time
being able to acquire data without the amount of measuring artefacts we have to cope
with today.
Using human-in-the-loop simulators instead of machines in a real working environ-
ment will also greatly reduce our problems with equipment for psychophysiological
measurements. It would be exciting to repeat the study presented in [VI] and [VII] on
such a simulator and compare the outcomes.
As mentioned before, despite the focus on a specific type of working machine, our
methodology and general conclusions should be applicable to all working machines of
similar complexity. A logical next step would be to prove just that.
In light of these many exciting possibilities we want to end this part of our thesis
with Captain Jack Sparrow’s words at the end of the film "Pirates of the Caribbean":
Now, bring me that horizon!
References
[1] Hollnagel, E. and Woods, D. D. (2005) “Joint Cognitive Systems. Foundations

of Cognitive Systems Engineering”. CRC Press, Taylor & Francis Group, Boca
Raton, FL, USA.
[2] Uwohali Inc. (1996) “Operability in Systems Concept and Design: Survey, As-
sessment, and Implementation”. Final Report, Website of Kennedy Space Cen-
ter, NASA, USA.
http://science.ksc.nasa.gov/shuttle/nexgen/Nexgen_Downloads/
Ops_Survey_of_Tools_Report_by_MSFC_1996.pdf
[3] O’Donnell, R. D. and Eggemeier, F. T. (1986) “Workload assessment methodol-
ogy”. Handbook of Perception and Performance, Vol. II: Cognitive Processes
and Performance, pp 42.1-42.49
[4] Volvo CE Image Gallery. Online resource.
http://vce.ams.llr.se
[5] Vännström, J. and Lindholm, S. (2007) “Educational Interface for Reducing Fuel
Consumption”. Master thesis, Luleå University of Technology, Luleå, Sweden.
http://epubl.luth.se/1402-1617/2007/114/
[6] Stener, P. and Snabb, R. (2008) ”Körbarhetskvantifiering av hjullastare”. Master
thesis, Linköping University, Linköping, Sweden.
[7] Boman, M. (2006) “On predicting fuel consumption and productivity of wheel
loaders”. Master thesis, Luleå University of Technology, Luleå, Sweden.
http://epubl.ltu.se/1402-1617/2006/009/
[8] Karlsson, J. (2010) “Analyzes of a wheel loader usage”. Master thesis, Mälarda-
len University, Eskilstuna, Sweden.
http://urn.kb.se/resolve?urn=urn:nbn:se:mdh:diva-11240
[9] Gellersted, S. (2002) ”Manövrering av hjullastare (Operation of Wheel
Loaders)”. Technical report, JTI – Institutet för jordbruks- och miljöteknik,
Uppsala, Sweden.
http://www.jti.se/index.php?page=publikationsinfo&publicationid=195&returnto=109
[10] Zhang, R., Alleyne, A. G. and Carter, D. E. (2003) “Multivariable Control of an

Earthmoving Vehicle Powertrain Experimentally Validated in an Emulated
Working Cycle”. Conference paper, ASME 2003 International Mechanical Engi-
neering Congress and Exposition.
http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.133.7304
[11] Heybroek, K. (2008) “Saving energy in construction machinery using displace-
ment control hydraulics”. Licentiate thesis, Department of Management and En-
gineering, Linköping University, Sweden.
[12] Lugnberg, M. and Ousbäck, M. (2008) “System Integration of a Pump Con-
trolled Open Circuit Solution in a Wheel Loader Application”. Master thesis,
Department of Management and Engineering, Linköping University, Sweden.
[13] Larsson, J. (2003) “Interoperability in Modeling and Simulation”. Doctoral the-
sis, Department of Mechanical Engineering, Linköping University, Linköping,
Sweden.
[14] Makkonen, P. (1999) “On Multi Body Systems Simulation in Product Design”.
Doctoral thesis, Department of Machine Design, Royal Institute of Technology,
Stockholm, Sweden.
http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-2800
[15] Andersson, H. (2009) “Aircraft Systems Modeling: Model Based Systems Engi-
neering in Avionics Design and Aircraft Simulation”. Licentiate thesis, Depart-
ment of Mechanical Engineering, Linköping University, Linköping, Sweden.
[16] Steinkellner, S. (2011) “Aircraft Vehicle Systems Modeling and Simulation under
Uncertainty”. Licentiate thesis, Department of Mechanical Engineering,
Linköping University, Linköping, Sweden.
[17] Hemami, A. (1994) “Motion trajectory study in the scooping operation of an
LHD-loader”. IEEE Transactions on Industry Applications, vol. 30, no. 5, Sept-
Oct 1994, pp 1333-1338.
http://dx.doi.org/10.1109/28.315248
[18] Blouin, S., Hemami, A. and Lipsett, M. (2001) “Review of Resistive Force Mod-
els for Earthmoving Processes”. ASCE Journal of Aerospace Engineering, vol.
14, no. 3, July 2001, pp 102-111.
http://dx.doi.org/10.1061/(ASCE)0893-1321(2001)14:3(102)
References 91
[19] Ericsson, A. and Slättengren, J. (2000) “A model for predicting digging forces
when working in gravel or other granulated material”. 15th European ADAMS
Users' Conference, Rome, Italy.
http://www.mscsoftware.com/support/library/conf/adams/euro/2000/Volvo_
Predicting_Digging.pdf
[20] Worley, M. D. and La Saponara, V. (2008) “A simplified dynamic model for
front-end loader design”. Proceedings of the Institution of Mechanical Engi-
neers, Part C: Journal of Mechanical Engineering Science, vol. 222, no. 11, No-
vember 2008, pp 2231-2249.
http://dx.doi.org/10.1243/09544062JMES688
[21] Cannon, H. N. (1999) “Extended Earthmoving with an Autonomous Excavator”.
Master thesis, Robotics Institute, Carnegie Mellon University, Pittsburgh (PA),
USA.
http://www.ri.cmu.edu/pubs/pub_2103.html
[22] Tan, C. P., Zweiri, Y. H., Althoefer, K. and Seneviratne, L. D. (2005) “Online
Soil Parameter Estimation Scheme Based on Newton-Raphson Method for
Autonomous Excavation”. IEEE/ASME Transactions on Mechatronics, vol. 10,
no. 2, April 2005, pp 221-229.
http://dx.doi.org/10.1109/TMECH.2005.844706
[23] Cambou, B. (1998) “Behaviour of granular materials”. Courses and lectures of
International Centre for Mechanical Sciences, Birkhäuser Verlag.
http://books.google.com/books?id=ga6jWoM_K2IC
[24] Pöschel, T. and Schwager, T. (2005) “Computational granular dynamics: models
and algorithms”. Springer Verlag.
http://books.google.com/books?id=4jHUX-7Qw-UC
[25] Tijskens, B., Lethé, G. and De Cuyper, J. (2010) “Towards realistic prediction of
soil-bucket interaction: Cosimulation between LMS Virtual.Lab Motion and
DEMeter”. Proceedings of the 1st Commercial Vehicle Technology Symposium,
CVT 2010. University of Kaiserslautern, Kaiserslautern, Germany.
[26] Fröberg, A. and Nielsen, L. (2004) “Dynamic Vehicle Simulation – Forward,
Inverse and New Mixed Possibilities for Optimized Design and Control”. SAE
2004 World Congress & Exhibition, Detroit (MI), USA, March 8-11, 2004.
http://dx.doi.org/10.4271/2004-01-1619
[27] Pióro, D. J., Shepherd, J. and Maciejowski, J. M. (2010) “Simulation and analy-
sis of powertrain hybridisation for construction equipment”. International Jour-
nal of Electric and Hybrid Vehicles, vol. 2, no. 3, 2010, pp 240-258.
http://dx.doi.org/10.1504/IJEHV.2010.033700
[28] Thiebes, P. and Vollmer, T. (2011) “Modellierung des Fahrers zur Untersuchung
von Antriebssträngen in der 1D-Simulation am Beispiel eines Radladers mit
Hybridantrieb”. Proceedings of 3. Fachtagung Hybridantriebe für mobile
Arbeitsmaschinen, pp 47-59.
[29] Jonasson, K. (2002) “Analysing Hybrid Drive System Topologies”. Licentiate
thesis, Department of Industrial Electrical Engineering, Lund University, Lund,
Sweden.
http://www.lu.se/o.o.i.s?id=12588&postid=587894
[30] Kjellin, G. (2009) “Interactive driver model for dynamic simulations of a
hauler”. Master thesis, Department of Aeronautical and Vehicle Engineering,
Royal Institute of Technology, Stockholm, Sweden.
[31] Macadam, C.C. (2003) “Understanding and Modeling the Human Driver”. Vehi-
cle System Dynamics, vol. 40, pp 101-134.
http://www.tandfonline.com/doi/abs/10.1076/vesd.40.1.101.15875
[32] Bergeron, H. P. and Adams, J. J. (1964) “Measured Transfer Functions of Pilots
during Two-axis Tasks with Motion”. NASA Technical Note D-2177, Washing-
ton, D.C., USA.
http://www.ncbi.nlm.nih.gov/pubmed/14131280
[33] McRuer, D. T. and Jex, H. R. (1967) “A Review of Quasi-Linear Pilot Models”.
IEEE Transactions on Human Factors in Electronics, vol. HFE-8, no. 3, pp 231-
249.
http://dx.doi.org/10.1109/THFE.1967.234304
[34] Gaines, B. R. (1969) “Linear and Nonlinear Models of the Human Controller”.
International Journal of Man-Machine Studies, vol. 1, pp 333-360.
http://pages.cpsc.ucalgary.ca/~gaines/reports/PSYCH/LinNonlin69/
[35] Cooper, G. E. and Harper, R. P., Jr. (1969) “The Use of Pilot Rating in the
Evaluation of Aircraft Handling Qualities”. NASA Technical Note D-5153,
Washington, D.C., USA.
http://www.dtic.mil/cgi-
bin/GetTRDoc?Location=U2&doc=GetTRDoc.pdf&AD=AD0689722
[36] Harper, R. P. and Jr., Cooper, G. E. (1986) “Handling Qualities and Pilot Evalua-
tion”. Journal of Guidance Control and Dynamics, vol. 9, pp 515-529.
http://www.mae.wmich.edu/faculty/Ro/ME540/FQ.pdf
[37] Cowley, W. L. and Skan, S. W. (1930) “A Simplified Analysis of the Stability of
Airplanes”. R&M No. l333, March 1930, Aeronautical Research Committee,
London, England.
[38] Lüdtke, A., Werber, L., Osterloh, J.-P. and Wortelen, B. (2009) “Modeling Pilot
and Driver Behavior for Human Error Simulation”. Proceedings of HCII 2009:
Digital Human Modeling, LNCS 5620, pp 403-412.
http://dx.doi.org/10.1007/978-3-642-02809-0_43
References 93
[39] Stengel, R. F. and Broussard, J. R. (1978) “Prediction of Pilot-Aircraft Stability

Boundaries and Performance Contours”. IEEE Transactions on Systems, Man,
and Cybernetics, vol. 8, no. 5, pp 349-356.
http://dx.doi.org/10.1109/TSMC.1978.4309975
[40] Hess, R. A. (1995) “A model-based analysis of handling qualities and adverse
aircraft-pilot coupling in high angle of attack flight”. Conference paper, IEEE In-
ternational Conference on Intelligent Systems for the 21st Century, vol. 3, pp
2663-2669.
http://dx.doi.org/10.1109/ICSMC.1995.538185
[41] Anderson, M. R., Clark and C., Dungan, G. (1995) “Flight test maneuver design
using a skill- and rule-based pilot model”. Conference paper, IEEE International
Conference on Intelligent Systems for the 21st Century, vol. 3, pp 2682-2687.
[42] Rashevsky, N. (1966) “Neglected Factors in Highway Safety”. University of
Michigan Mental Health Research Institute, Grant GM-12032-01.
[43] Erséus, A. (2010) “Driver-Vehicle Interaction. Identification Characterization
and Modelling of Path Tracking Skill”. Department of Aeronautical and Vehicle
Engineering, Royal Institute of Technology, Stockholm, Sweden.
[44] Plöchl, M. and Edelmann, J. (2007) “Driver models in automobile dynamics ap-
plication”. Vehicle System Dynamics, vol. 45, nos. 7-8, pp 699-741.
http://dx.doi.org/10.1080/00423110701432482
[45] Rasmussen, J. (1983) "Skills, Rules, and Knowledge; Signals, Signs, and Sym-
bols, and Other Distinctions in Human Performance Models". IEEE Transactions
on Systems, Man, and Cybernetics, vol. SMC-13, no. 3, pp 257-266.
http://www.carlosrighi.com.br/177/Ergonomia/Skills%20rules%20and%20knowledge%2
0-%20Rasmussen%20seg.pdf
[46] Michon, J. A. (1985) "A critical view of driver behaviour models: What do we
know, what should we know?". In: Human Behaviour and Traffic Safety, pp 485-
520, Plenum Press, New York, NY, USA.
www.jamichon.nl/jam_writings/1985_criticial_view.pdf
[47] Salvucci, D. D., Boer, E. R. and Liu, A. (2001) “Towards an Integrated Model of
Driver Behavior in a Cognitive Architecture”. Transportation Research Record,
no. 1779, pp 9-16.
[48] Gluck, A. and Pew, R., 2005, “Modeling Human Behavior with Integrated Cog-
nitive Architectures: Comparison, Evaluation, and Validation”. Lawrence Erl-
baum Associates, USA.
[49] Salvucci, D. D. (2006) “Modeling Driver Behavior in a Cognitive Architecture”.

Human Factors, vol. 48, no. 2, pp 362-380.
http://dx.doi.org/10.1518/001872006777724417
[50] Salvucci, D. D. (2007) “Integrated Models of Driver Behavior”. In: Integrated
models of cognitive systems, pp 356 - 367. Oxford University Press, New York,
NY, USA.
[51] Robbins, B., Carruth, D. and Morais, A. (2009) “Bridging the Gap between HCI
and DHM: The Modeling of Spatial Awareness within a Cognitive Architec-
ture”. Proceedings of HCII 2009: Digital Human Modeling, LNCS 5620, pp
295-304.
http://dx.doi.org/10.1007/978-3-642-02809-0_32
[52] Novak, M. J., Larson, M. J. and Larson, C. S. (1993) “Evaluation of Advanced
Steering Control with Computer Simulation”. SAE-Paper 932383.
http://dx.doi.org/10.4271/932383
[53] Norris, W. R., Screenivas, R. S., Zhang, Q. (2002) “A Novel Real-Time Human
Operator Performance Model for Performing Adaptive System Design”. Pro-
ceedings of the 2002 Conference of American Society of Agricultural Engineers,
Automation Technology for Off-Road Equipment, pp 287-306.
http://asae.frymulti.com/abstract.asp?aid=10019&t=1
[54] Elezaby, A.A., Abdelaziz, M., Cetinkunt, S. (2008) “Operator Model for Con-
struction Equipment”. Proceedings of IEEE/ASME 2008 International Con-
ference on Mechatronic and Embedded Systems and Applications, MESA 2008,
pp 582-585.
http://dx.doi.org/10.1109/MESA.2008.4735708
[55] Elezaby, A. A. (2010) “Virtual Autonomous Operator Model for Construction

Equipment Applications”. Doctoral thesis, University of Illinois at Chicago, Chi-
cago, IL, USA.
http://gradworks.umi.com/34/17/3417331.html
[56] Wu, L. (2003) “A Study on Automatic Control of Wheel Loaders in Rock/Soil
Loading”. Doctoral thesis, University of Arizona, Tucson, Arizona, USA.
http://wwwlib.umi.com/dissertations/fullcit/3090033
[57] Marshall, J. A. (2001) “Towards Autonomous Excavation of Fragmented Rock:
Experiments, Modelling, Identification and Control”. Master Thesis, Queen's
University, Kingston (Ontario), Canada.
[58] Lever, P. J. A. (2001) “An automated digging control for a wheel loader”. Ro-
botica, vol. 19, no. 5, pp 497-511.
http://dx.doi.org/10.1017/S0263574701003435
References 95
[59] Sarata, S., Koyachi, N., Tubouchi, T., Osumi, H., Kurisu, M. and Sugawara, K.
(2006) “Development of Autonomous System for Loading Operation by Wheel
Loader”. Proceedings of the 23rd International Symposium on Automation and
Robotics in Construction, pp 466-471.
http://www.iaarc.org/publications/fulltext/isarc2006-00150_200608211032.pdf
[60] De Waard, D. (1996) “The Measurement of Driver’s Mental Workload”. Doc-
toral thesis, University of Groningen, The Netherlands.
http://irs.ub.rug.nl/ppn/297329391
[61] Castor, M., Hanson, E., Svensson, E., Nählinder, S., Le Blaye, P., MacLeod, I.,
Wright, N., Alfredson, J., Ågren, L., Berggren, P., Juppet, V., Hilburn, B., Ohls-
son, K. (2003) “GARTEUR Handbook of Mental Workload Measurement”, Fi-
nal Report for Flight Mechanics Group of Responsibles (FM GoR) of the Group
for Aeronautical Research and Technology in Europe (GARTEUR) by
GARTEUR Action Group FM AG13.
http://wwwserv2.go.t-systems-sfr.com/garteur/documentos/GARTEUR_TP-145.htm
[62] Nählinder, S. (2009) “Flight Simulator Training: Assessing the Potential”. Doc-
toral thesis, Department of Management and Engineering, Linköping University,
Sweden.
[63] Kidd E. A. and Bull, G. (1962) “Handling Qualities Requirements as Influenced
by Pilot Evaluation Time and Sample Size”. Cornell Aeronautical Laboratory
Report TB-1444.
http://www.stormingmedia.us/56/5620/0562004.html
[64] Rubio, S., Díaz, E., Martín, J. and Puente, J. M. (2004) “Evaluation of Subjective
Mental Workload: A Comparison of SWAT, NASA-TLX, and Workload Profile
Methods”. Applied Psychology: An International Review, vol. 53, no. 1, pp 61-
86.
http://dx.doi.org/10.1111/j.1464-0597.2004.00161.x
[65] Roscoe, A. H. (1992) “Assessing pilot workload. Why measure heart rate, HRV
and respiration?”. Biological Psychology, vol. 34, nos. 2-3, pp 259-287.
http://dx.doi.org/10.1016/0301-0511(92)90018-P
[66] Park, B. (2009) “Psychophysiology as a Tool for HCI Research: Promises and
Pitfalls”. Proceedings of HCII 2009: New Trends, LNCS 5610, pp 141-148.
http://dx.doi.org/10.1007/978-3-642-02574-7_16
[67] Summala, H. (2000) “Automatization, automation, and modeling of driver’s be-
havior”. Recherche – Transports – Sécurité, vol. 66, pp 35-46.
http://dx.doi.org/10.1016/S0761-8980(00)80004-7
[68] Laughery, R. (1998) “Computer Simulation as a Tool for Studying Human-

centered Systems”. Proceedings of the 1998 Winter Simulation Conference, vol.
1, pp 61-65.
http://dx.doi.org/10.1109/WSC.1998.744900
[69] Keller, J. (2002) “Human Performance Modeling for Discrete-event Simulation:
Workload”. Proceedings of the 2002 Winter Simulation Conference, vol. 1, pp
157-162.
http://dx.doi.org/10.1109/WSC.2002.1172879
[70] Markkula, G. and Engström, J. (2006) “A Steering Wheel Reversal Rate Metric
for Assessing Effects of Visual and Cognitive Secondary Task Load”. Proceed-
ings of the 13th ITS World Congress.
http://trid.trb.org/view.aspx?id=847195
[71] Hughes, K. K. and Jiang, X. (2008) “Modeling Human Performance of Hydrau-
lic Excavator Operators”. Human Factors and Ergonomics in Manufacturing &
Service Industries, vol. 20, no. 5, pp 408-423.
http://dx.doi.org/10.1002/hfm.20191
[72] Bernold, L. E. (2007) “Quantitative Assessment of Backhoe Operator Skill”.
Journal of Construction Engineering and Management, vol. 133, no. 11, pp 889–
899.
[73] Schütte, S. (2005) “Engineering Emotional Values in Product Design – Kansei
Engineering in Development”. Doctoral thesis, Department of Mechanical Engi-
neering, Linköping University, Linköping, Sweden.
[74] Begum, S., Ahmed M. U., Funk, P., Xiong, N. and von Schéele, B. (2006) “Us-
ing Calibration and Fuzzification of Cases for Improved Diagnosis and Treat-
ment of Stress”. Proceedings of the 8th European Conference on Case-Based
Reasoning, pp 113-122.
[75] Begum, S. (2011) “A Personalised Case-Based Stress Diagnosis System Using
Physiological Sensor Signals”. Doctoral thesis, Mälardalen University, Västerås,
Sweden.
[76] Ahmed, M. U. (2010) “A Case-Based Multi-Modal Clinical System for Stress
Management”. Licentiate thesis, Mälardalen University, Västerås, Sweden.
References 97
[77] Voskamp, J. and Urban, B. (2009) “Measuring Cognitive Workload in Non-

military Scenarios. Criteria for Sensor Technologies”. Proceedings of HCII
2009: Foundations of Augmented Cognition - Neuroergonomics and Operational
Neuroscience, LNCS 5638, pp 304-310.
http://dx.doi.org/10.1007/978-3-642-02812-0_36
[78] Filla, R. (2006) “A method for operating a working machine and a working ma-
chine with an improved ability to meet transient loads”. International patent ap-
plication, WO 2008/041890 A1.
http://www.wipo.int/patentscope/search/en/WO2008041890
[79] Filla, R. (2007) “A method and a system for providing feedback to a vehicle op-
erator”. International patent application, WO 2008/123808 A1.
[80] Filla, R. (2008) “A method for controlling a hydraulic system”. International
patent application, WO 2009/145682 A1.
[81] Filla, R. (2008) “A method for controlling a power source”. International patent
application, WO 2009/148364 A1.
[82] Kormann, G. (2010) “Operator Assist Systems for Agricultural Vehicles. Guid-
ance and Automation Products”. Proceedings of the 1st Commercial Vehicle
Technology Symposium, CVT 2010. University of Kaiserslautern, Kaiserslautern,
Germany.
[83] Löfgren, B. (2009) “Kinematic Control of Redundant Knuckle Booms with
Automatic Path-Following Functions”. Doctoral thesis, Department of Machine
Design, Royal Institute of Technology, Stockholm, Sweden.
[84] Tervo, K. (2010) “Human Adaptive Mechatronics Methods for Mobile Working
Machines”. Doctoral thesis, Department of Automation and Systems Technol-
ogy, Aalto University, Espoo, Finland.
http://lib.tkk.fi/Diss/2010/isbn9789526035307
[85] Enes, A. R. (2010) “Shared Control Of Hydraulic Manipulators To Decrease
Cycle Time”. Doctoral thesis, School of Mechanical Engineering, Georgia Insti-
tute of Technology, Atlanta, GA, USA.
http://hdl.handle.net/1853/37251
[86] Scallen, S. F. and Hancock, P. A. (2001) “Implementing Adaptive Function Al-
location”. The International Journal of Aviation Psychology, vol. 11, no. 2, pp
197-221.
http://dx.doi.org/ 10.1207/S15327108IJAP1102_05
[87] Inagaki, T. (2003) “Adaptive Automation: Sharing and Trading of Control”. In:
Handbook of Cognitive Task Design, pp 147-169.
http://www.css.risk.tsukuba.ac.jp/pdf/commentary07.pdf
[88] Parasureman, R. and Riley, V. (1997) “Humans and Automation: Use, Misuse,
Disuse, Abuse”. Human Factors, vol. 39, no. 2, pp 230-253.
http://archlab.gmu.edu/people/rparasur/Documents/ParasRileyHF97.pdf
[89] Lee, J. D. and See, K. A. (2004) “Trust in Automation. Designing for Appropri-
ate Reliance”. Human Factors, vol. 46, no. 1, pp 50-80.
http://dx.doi.org/10.1518/hfes.46.1.50_30392
[90] Lee, J. D. (2008) “Review of a Pivotal Human Factors Article: ‘Humans and
Automation: Use, Misuse, Disuse, Abuse’”. Human Factors, vol. 50, no. 3, pp
404-410.
http://dx.doi.org/10.1518/001872008X288547
[91] Parasuraman, R. and Manzey, D. H. (2010) “Complacency and Bias in Human
Use of Automation: An Attentional Integration”. Human Factors, vol. 52, no. 3,
pp 381-410.
http://dx.doi.org/10.1177/0018720810376055
[92] De Visser, E. and Parasuraman, R. (2011) “Adaptive Aiding of Human-Robot
Teaming – Effects of Imperfect Automation on Performance, Trust, and Work-
load”. Journal of Cognitive Engineering and Decision Making, vol. 5, no. 2, pp
209-231.
http://dx.doi.org/10.1177/1555343411410160
[93] Parasuraman, R., Sheridan, T. B. and Wickens, C. D. (2000) “A Model for Types
and Levels of Human Interaction with Automation”. IEEE Transactions on Sys-
tems, Man, and Cybernetics, vol. 30, no. 3, pp 286-297.
http://dx.doi.org/10.1109/3468.844354
[94] Kaber, D. B., Wright, M. C., Prinzel, L. J. and Clamann, M. P. (2005) “Adaptive
Automation of Human-Machine System Information-Processing Functions”.
Human Factors, vol. 47, no. 4, pp 730-741.
http://dx.doi.org/10.1518/001872005775570989
[95] Marzani, S., Tesauri, F., Minin, L., Montanari, R. and Calefato, C. (2009) “De-
signing a Control and Visualization System for Off-Highway Machinery Accord-
ing to the Adaptive Automation Paradigm”, Proceedings of HCII 2009: Aug-
mented Cognition, LNAI 5638, pp 42-50.
http://dx.doi.org/10.1007/978-3-642-02812-0_6
[96] Vinge, E. and Pitkin, E. (1972) “Human Operator Dynamics for Aural Compen-
sator Tracking”. Proceedings of the IEEE Transactions on Systems, Man, and
Cybernetics, SMC-2.
http://dx.doi.org/10.1109/TSMC.1972.4309160
References 99
[97] Donmez, B., Cummings, M. L. and Hudson, D. G. (2009) “Auditory Decision

Aiding in Supervisory Control of Multiple Unmanned Aerial Vehicles”. Human
Factors, vol. 51, no. 5, pp 718-729.
http://dx.doi.org/10.1177/0018720809347106
[98] Wilson, G. F. and Russell, C. A. (2003) “Real-Time Assessment of Mental
Workload Using Psychophysiological Measures and Artificial Neural Net-
works”. Human Factors, vol. 45, no. 4, pp 635-643.
http://dx.doi.org/10.1518/hfes.45.4.635.27088
[99] Wilson, G. F., Russell, C. A. (2007) “Performance Enhancement in an Uninhab-
ited Air Vehicle Task Using Psychophysiologically Determined Adaptive Aid-
ing”. Human Factors, vol. 49, no. 6, pp 1005-1018.
http://dx.doi.org/10.1518/001872007X249875
[100] Haufe, S., Treder, M. S., Gugler, M. F., Sagebaum, M., Curio, G., Blankertz, B.
(2011) “EEG potentials predict upcoming emergency brakings during simulated
driving”. Journal of Neural Engineering, vol. 8, no. 5, pp 1-11.
http://dx.doi.org/10.1088/1741-2560/8/5/056001
[101] Lei, S., Roetting, M. (2011) “Influence of Task Combination on EEG spectrum
Modulation for Driver Workload Estimation”. Human Factors, vol. 53, no. 2, pp
168-179.
http://dx.doi.org/10.1177/0018720811400601
[102] De Greef, T., Lafeber, H., van Oostendorp, H. and Lindenberg, J. (2009) “Eye
Movement as Indicators of Mental Workload to Trigger Adaptive Automation”,
Proceedings of HCII 2009: Augmented Cognition, LNAI 5638, pp 219-228.
http://dx.doi.org/10.1007/978-3-642-02812-0_26
(Internet links verified on August 18, 2011)

Paper I
Using Dynamic Simulation
I
in the Development of Construction Machinery
Reno Filla (1) and Jan-Ove Palmberg (2)
(1)VOLVO WHEEL LOADERS AB, ESKILSTUNA, SWEDEN
(2) DEPT. OF MECHANICAL ENGINEERING, LINKÖPING UNIVERSITY, SWEDEN
Abstract
As in the car industry for quite some time, dynamic simulation of complete vehicles is
being practiced more and more in the development of off-road machinery. However,
specific questions arise due not only to company structure and size, but especially to the
type of product. Tightly coupled, non-linear subsystems of different domains make pre-
diction and optimisation of the complete system’s dynamic behaviour a challenge. Fur-
thermore, the demand for versatile machines leads to sometimes contradictory target
requirements and can turn the design process into a hunt for the least painful compro-
mise. This can be avoided by profound system knowledge, assisted by simulation-
driven product development. This paper gives an overview of joint research into this
issue by Volvo Wheel Loaders and Linköping University on that matter, and lists the
results of a related literature review. Rather than giving detailed answers, the problem
space for ongoing and future research is examined and possible solutions are sketched.
Keywords: simulation, complex systems, integrated product development

A witty quote proves nothing.
(Voltaire)
This paper has been published as:

Filla, R. and Palmberg, J.-O. (2003) “Using Dynamic Simulation in the Development of
Construction Machinery”. The Eighth Scandinavian International Conference on Fluid
Power, Tampere, Finland, Vol. 1, pp 651-667, May 7-9, 2003.
(Internet link refers to a Technical Report of the original paper)

Using Dynamic Simulation… 103
1 Introduction
The general motives for “Virtual Prototyping” are probably familiar to all engineers:
Stricter legal requirements (e.g. with regard to exhaust emissions and sound) and
tougher customer demands (e.g. with regard to performance and handling) lead to more
advanced, complex systems, which are harder to optimise. With traditional methods,
development will cost more and need more time. In contrast to this, increased competi-
tion demands lower development cost and shorter project times.
“Virtual Prototyping” has been generally adopted in the vehicle industry as a major
step towards solving this conflict both on the consumer side (cars) and on the commer-
cial side (trucks and buses). Having started with simulation of sub-systems, the state-of-
the-art is simulation of complete vehicles, mostly for evaluation of handling, comfort,
and durability but also for crash-tests.
One reason for the off-road equipment industry lagging behind can be found in the
size of these companies: being significantly smaller, broad investments in the latest
CAE tools (together with the necessary training) take longer until amortisation. The
other, and probably more important reason is that the products are very different to
those of the on-road vehicle industry – not only geometrically (size), but topologically
(sub-systems of various domains and their interconnections).
Cases have recently been published where complete machines were simulated for
evaluation of the simulation technique itself [1], sub-systems [2], comfort-related as-
pects [3], or durability [4]. This paper too, will deal with dynamic simulation of com-
plete machines, but for analysis and optimisation of overall performance and related
aspects. The focus will be on wheel loaders with hydrodynamic transmissions, but most
findings (and questions) will be also applicable to other off-road machinery.
Figure 1. Multi-body model of a Volvo L220E wheel loader, loading gravel

104 Paper I
2 Background
Due to the versatility of these machines, wheel loaders need to fulfil a great many re-
quirements, which are often interconnected and sometimes contradict each other. This is
true of essentially all industrial products and is widely recognised. Since this paper fo-
cuses on performance (and related issues), aspects such as total cost of ownership, mar-
ket availability, reliability etc. will not be discussed explicitly.
To give just a few examples, the following performance-related aspects are important
(varying with the working task):
- Geometric parameters (lift height, digging depth, dump reach, parallel
alignment)
- Loads, torques and forces (tipping load, break-out torque, lift force, traction
force)
- Speeds and cycle times (complete machine and sub-systems)
- Consumption and emissions (fuel consumption, exhaust emissions, sound
& vibration)
- Controllability (precision, feedback, response)
While some of the items are clearly determined by more than just one sub-system
(e.g. lift force, which is determined by hydraulics and load unit), others seem to be pos-
sible to attribute to one single sub-system (e.g. traction force). One might thus, wrongly,
be tempted to leave such aspects out of the optimisation loop when it comes to trading-
off product targets against each other when choosing technical solutions.
A modern wheel loader of hydro-dynamic design, however, consists of tightly cou-
pled, non-linear sub-systems of different domains. Since all sub-systems interact even
in seemingly simple cases, prediction and optimisation of the complete system’s dy-
namic behaviour is a challenge.
Figure 2 shows how the sub-systems of main interest are interacting when loading
gravel.
Figure 2. Simplified transfer scheme of a wheel loader, loading gravel

(TC = Torque Converter, ECU = Electronic Control Unit, V = Vehicle, I = Instrument)
For loading granular material like gravel, the bucket first has to penetrate the pile.
This requires traction force, which is achieved by transferring torque from the engine
via a torque converter, transmission, axles, and wheels to the ground. A typical se-
quence for actually filling the bucket is then to break material by tilting backwards a
little, lifting a little, and penetrating even further. The lift and tilt functions require en-
gine torque to be transferred via hydraulic pumps (converting torque to hydraulic pres-
sure), cylinders (converting hydraulic pressure to longitudinal force), and loading unit.
As shown, the two different transfer paths compete for the limited engine torque.
Furthermore, in the act of loading, these two paths are brought together in the gravel
pile: the penetration establishes a reaction force at the bucket, which counteracts the
break-out force, as well as the lift force. Like a short circuit, loads can be transferred
back to the origin, in this case the engine, and lead to overload (greater interaction be-
tween the systems or the engine stops completely).
This has to be avoided by a design that carefully balances traction and lift/tilt. But
these functions can not be optimised without influencing others, e.g. machine speed and
lifting time. In addition to this, Figure 2 neglects the fact that loading unit geometry and
bucket design are of great importance for a smooth bucket filling. In addition, a ma-
chine operator who handles a loader in an unintelligent way can have difficulty in
achieving a favourable balance. In such cases (but not exclusively), implementing so-
phisticated electronic control strategies can be of great help.
Above, one single phase in one single (yet frequent) handling case has been de-
scribed. A typical loading cycle consists of several phases, where balance has to be es-
tablished in each. Additionally, there are a great many different handling cases, each
with its own requirements that need to be satisfied.
With the ambition of developing a product that is significantly better than its prede-
cessor every time, this becomes harder and harder to achieve. The trend towards more
electronic control is as striking in the off-road equipment industry as it is in the on-road
branch (passenger cars, trucks, and busses). In Figure 2, this is symbolised by five elec-
tronic control units, each serving a specific purpose (component or sub-system). The
ring represents the common data bus (CAN-bus) the ECUs are connected to.
3 Design Process, Current Position

Since this research project is being carried out at Volvo Wheel Loaders, parts of their
product development process will be used as examples throughout this paper. However,
there are several indications that the procedure is similar for other companies in the off-
road industry.
The intention is not to describe the entire development process, but rather to focus on
the parts that handle overall machine performance, as described in the previous section.
After setting up preliminary product targets, an initial calculation loop is started, with
the objective of finding a good balance between loading unit, drive train (including en-
106 Paper I
gine) and hydraulics. In this process, product targets are gradually refined, until the
overall system’s specifications are judged acceptable and feasible (Figure 3).
Figure 3. Initial calculation loop for balancing the main systems
After this step, which involved several assumptions, derivative analysis is performed
for all major sub-systems, which will either confirm those assumptions or require a re-
turn to the first loop with modified input parameters.
As sketched in section 2, a wheel loader (a complex system of tightly coupled, non-
linear sub-systems) shows complex (non-linear) behaviour in complex (real-life) situa-
tions. To optimise the first one, the latter two need to be understood as fully as possible.
In this case, there are multiple handling situations, each involving humans with complex
behaviour of their own. Since the initial loop only relies on static calculations (though
guided by valuable experience), there is the possibility that the supposedly found opti-
mum will not work flawlessly in a dynamic reality.
It is therefore praxis to build early, so-called “functional prototypes” as the next step.
Using them for extensive real-life tests, the intention is to uncover any static and dy-
namic problem at an early stage in the design phase, where reiteration is still inexpen-
sive. However, these functional prototypes are often not built with exactly the same
components as planned for the production machines, since those are still in develop-
ment and therefore only available as early prototypes (or not at all). This poses the risk
that dynamic shortcomings will not be detected until later.
One example is that engines have been found to be critical. Electronically controlled
engines can have higher response times than older, mechanically controlled ones. There
are several reasons for this, one being that the ECU itself needs some time to collect all
the information from the sensors, process it, and calculate the appropriate action. Addi-
tionally, modern engines have to be certified according to the latest exhaust emission
legislation. Virtually no diesel engine today shows any sign of much visible smoke in
transient load situations. This is achieved by limiting the injected fuel quantity at low
turbo pressures, thus making the engine a little less responsive. A second example is
hydraulic pumps: modern wheel loaders consist of load-sensing hydraulic systems,
which in most cases are superior to the old open-centre design. In such LS-systems the
response time of the pump is a critical factor. If a functional prototype is built with one
or more pumps with incorrect dynamic behaviour, together with a diesel engine as de-
scribed above, the complete loader’s dynamics can differ significantly from those of a
series-manufactured one.
A requirement for the next stage, building “real” prototypes, is that all major compo-
nents are at least mature prototypes, close to series production. In this stage, all major
testing is done and detailed design enters its final phase. Due to the above mentioned
procedure for functional prototypes additional dynamic complications are sometimes
detected. This is unfortunate, but can be handled by evaluation and preparation of pos-
sible adjustments in advance.
In the next stage, building a so-called “production prototype”, all components have to
be series manufactured. Originally, this prototype was intended to allow the production
facility to test and refine assembly methods. But since even mature prototypes of major
components might show a different dynamic behaviour to those from series production,
this last prototype can also be used for one more validation. Any detected, performance-
related issue in this stage is very costly and can jeopardise the complete project time
schedule – but building pre-series machines with shortcomings in the next step is unac-
ceptable. Testing of this prototype (and its now updated predecessors) is therefore very
thorough, i.e. resource-consuming.
In general, evaluation of machine performance consists of several stages, such as
component validation (e.g. engine), sub-system testing (e.g. lift time and traction force),
and finally testing of the complete loader. The latter is first performed by experienced
test personnel at the company’s proving ground. Having passed those tests, the proto-
types are sent to selected customers, who use them in their everyday work. This ensures
that a greater variety of operators have the chance to judge the design.
Such a procedure is often referred to as a “V-approach”, where global targets are
broken-down into targets for sub-systems and components (top-down), with the valida-
tion being done the other way around (bottom-up). It is interesting that the more the
testing resembles real life, the less measurable the outcome becomes. Specifications of
isolated components are easily validated by measurements in test benches. Sub-systems
are a little harder to test, mainly because there are few test benches that are appropriate,
but also because not every aspect of the dynamic behaviour has been specified. Testing
complete machines by professional test personnel sometimes reveals that these people
are far more experienced than the average wheel loader operator. The risk of being bi-
ased, even involuntarily, exists. Despite this, with a limited proving ground and limited
personnel, it is impossible to take all variations of handling case and operating habit
108 Paper I
into account. Therefore, the machines are field-tested by selected customers. However,
in the last two scenarios the validation of complete machine performance relies strongly
on the operator and his/her subjective judgement. While the employed test personnel are
more used to expressing their impression in measurable terms, it is the voice of the cus-
tomer that counts in the market. The challenge is how to quantify the criteria for a well-
balanced machine. And how to quantify all boundary conditions during a measurement,
i.e. how to ensure repeatability.
4 Design Process, Vision

The aim of the present project is evolution of the current product development process,
rather than revolution by means of Design Science (see [5] for a critical review by
Frost). The research question is therefore how to augment the existing design process
with dynamic simulation. As mentioned before, the focus is on analysis and optimi-
sation of overall performance and related aspects.
The revised design process has to fulfil the following non-optional targets: Lead to
development of
… products of at least equal high performance, efficiency, and operability
… but with increased robustness
… in a shorter time
… at a lower development cost
compared with today. Saved resources (time, money, and people) can then be spent
on optimising one or all of the aspects mentioned in the first item in the list.
In an earlier project, a valuable lesson has been learned: speed matters when it comes
to iterations, especially in the concept phase. When Volvo’s old loading unit calculation
programme was to be replaced by a more modern version, this was done with a proprie-
tary simulation system, which was based on a multi-body system (MBS) and a modern
database. The development was done in-house. This new simulation system has proved
to be more flexible, more accurate, and especially more efficient for the user, except for
some pre-study engineers, who used the superior speed of the old calculation pro-
gramme to brute-force optimise loading unit geometries. Since the new system obtains
results by multi-body simulation, rather than calculation of hard-coded explicit equa-
tions, one run takes a couple of seconds longer than with the old programme. Brute-
force optimisations of the old type are no longer time efficient. If this had been known
before, i.e. if it had been included in the project targets, one could have developed a
special downscaled version, that was less accurate but faster. Introduction of the new
system forced those pre-study engineers to abandon a time-efficient technique that
worked well.
A similar risk can be seen with this research project: the current initial (static) calcu-
lation loop (Figure 3) is fast and reasonably accurate. The shortcoming today is rather
that the dynamic behaviour of the complete machine is first evaluated by testing a func-
tional prototype, followed by testing a “real” prototype. Therefore a moderately revised

process (as shown in Figure 4) is proposed.
Product
START
targets
Initial static calculation loop
Static loading unit

simulations
(Proprietary)
Revise
product targets Product
targets
Static drivetrain Static hydraulic

calculations calculations
(Proprietary) (Proprietary)
N Find alternative
Targets
fulfilled?
fulfilling existing
product targets
Dynamically augmented,
static calculations
Targets N Deviation N
fulfilled? approvable?
Y Y
Dynamic simulation
of complete machine
Targets N Deviation N
fulfilled? approvable?
Y Y
System Layout
Figure 4. Revised design process

110 Paper I
As practised today, the process starts with feeding the product targets into the initial
static calculation loop. If no satisfactory solution can be found, the targets need to be
revised. The next step is described as “Dynamically augmented, static calculations” (see
below for an explanation). Here, as well as in the next step “Dynamic simulation of
complete machine”, non-fulfilled product targets do not necessarily lead to a reiteration,
as long as the deviation can be approved. Since product targets will probably never
cover all product properties (including dynamic behaviour), this checkpoint will give
the whole process some flexibility. Only if the deviation is too high will a new static
calculation loop need to be started (with the original product targets as input). If no al-
ternative layout candidate can be calculated, the product targets need to be revised.
Since simulating complete machines can take a long time, a quick check on whether
the solution from the initial loop is dynamically feasible, would be of great help. This is
supposed to be done in the second step “Dynamically augmented, static calculations”.
Concrete methods still need to be developed. However, one example can be given (ap-
plies to machines with hydrodynamic transmissions and LS-hydraulics): a critical phase
in a so-called “short loader cycle” (or V-cycle) is when the machine, coming backwards
with a full bucket, changes direction towards the load receiver (e.g. an articulated hauler
or a dump truck). During that time, there is a close interaction between the main subsys-
tems:
1. In order to reverse the machine, the operator lowers the engine speed (other-
wise the gear shifting will be jerky and the transmission couplings might wear
out prematurely). Less torque is available at lower speeds. Additionally, the
engine response is worse at lower speeds, mainly due to inertia of the turbo
charger (and smoke limiter settings, as explained earlier).
2. When switching gears from reverse to forward, the loader is still rolling back-
wards. This forces an abrupt change of rotational direction of the torque con-
verter’s turbine wheel, thus greatly increasing the slip, which leads to a sudden
increase in torque demand from the engine.
3. Some operators do not stop lifting the loading unit while reversing, thus requir-
ing high oil flow throughout the whole process. The oil flow is proportional to
the hydraulic pump’s displacement and shaft speed, and the load-sensing
pumps can be assumed to be directly connected to the engine crankshaft. Thus,
at lower engine speeds and high demand of oil flow, the displacement goes to-
wards maximum (simply put). The amount of torque that a pump demands of
the engine is proportional to its displacement and hydraulic pressure. Since the
loader’s bucket is full and due to the loading unit’s geometry, hydraulic pres-
sure is high. With displacement at maximum, this too leads to an increased
torque demand.
Both drive train and hydraulics thus suddenly apply higher load to an already weak-
ened engine. All depends on the time scale of these three concurrent phenomena, which
is why a satisfactory answer can only be given by a detailed dynamic simulation (or
testing of a real machine, i.e. a functional prototype). However, an approximate, less
time-consuming first approach is possible: given the loader speed when switching gears
from reverse to forward, and given the engine speed at that time, the maximum slip be-
tween the pump wheel and the turbine wheel of the torque converter can be calculated,
and thus the maximum demanded engine torque (by using the torque converter’s speci-
fications). In the worst case scenario, hydraulic pressure and pump displacement can be
assumed to be maximal. Together with the engine speed, this gives the second engine
torque demand. If the sum of both torque demands is larger than the available steady-
state engine torque at that speed, the proposed system layout will almost certainly lead
to dynamic problems. If the available steady-state engine torque is considerably larger
than the sum of both torque demands, the system will most probably function as in-
tended. To check the case in between, it is important to consider that due to factors such
as turbo charger inertia and smoke limiter mentioned before, an accelerating engine
seldom has full steady-state torque at lower engine speeds. Therefore, checking against
the static torque curve might give a false sense of security.
Figure 5. Dynamic torque curve of a modern, turbo-charged diesel engine.

Less gap between static and dynamic torque at lower acceleration (grey, dashed curve)
due to faster boost pressure build-up relative to engine speed
112 Paper I
What is needed is a dynamic engine torque curve, measured at a typical acceleration

rate. Assuming the engine is released to low idle during reversing (worst case); the ac-
celeration rate is simply obtainable as the speed difference from the engines max torque
speed to low idle, divided by a typical time for the reversal phase. Using a test bench
with an electro-magnetic brake, an engine run-up with a forced acceleration as de-
scribed above can be performed. The amount of available torque during that phase
equals the necessary braking torque (Figure 5).
In Figure 5, the grey, dashed curves represent a test, where the engine was allowed to
accelerate less. In theory, this should not only generate more available torque because
the rotating parts consume less torque for acceleration, but also because there is more
time for turbo boost pressure to be built up. As can be seen in the first diagram, the en-
gine torque itself does not increase faster in time for the case with less forced accelera-
tion. Instead, confirming the theory, more torque is available relative to engine speed,
expressed as a smaller gap between static and dynamic torque curve for the slower ac-
celeration. If the engine were allowed to accelerate sufficiently slowly, the dynamic
torque curve (bold line style) would follow the static one (thin line style).
In order to quickly sort out system layout candidates that will lead to dynamic prob-
lems, more such checking methods need to be developed, together with refinement of
the existing rules of thumb.
Taking a critical look at the proposed revised design process, even if it has been la-
belled as “evolutionary rather than revolutionary” in the beginning of this section, it
implies a bigger change than might be obvious at first glance. With its introduction,
functional prototypes are only built to verify system layouts that have passed both steps
that check the dynamic behaviour. These steps are given the great responsibility of de-
ciding whether a certain system layout can be further pursued or is to be sent back for
reiteration. The worst scenario would be to reject cases that actually would work in real-
ity, but nobody will ever know since only approved cases enter the functional prototype
phase. To avoid this and to build confidence, there has to be a smooth transition be-
tween current practice and the proposed process.
In the literature, this is referred to as the change from simulation-verified design to
simulation-driven design. Campbell shows the usual charts and gives some practical tips
on this in [6]. T Larsson examines this topic from a research perspective in [7]. Even
though Murphy and Perera are writing about their findings on Discrete Event Simula-
tion, their paper [8] provides some valuable insight, such as classifying “Team build-
ing” as one major contribute factor for success, while “Software selection” is only men-
tioned as a minor factor. They also list the activity of making effective internal
promotion for simulation as a major success factor, advice that is similar to the findings
of [9], where Beskow and Ritzén present the findings of their research on change man-
agement in general.
5 Puzzle Pieces
In order to fulfil the vision drafted in the previous section, progress needs to be made in
various areas. Some pieces of this puzzle have already fallen into place, though.
5.1 Simulation and Modelling Techniques

As shown earlier, off-road equipment consists of systems from various domains, and
most of them need to be taken into account when simulating performance of complete
machines – which is usually a collaborative activity. As noted by many researchers,
engineers have often already chosen one domain-specific simulation program that they
are familiar with. Instead of forcing migration to one monolithic simulation system that
can be used in several domains (but offers only limited functionality in the individual
domain), a better approach is to couple the specialised, single-domain tools. This has
the advantage that both pre- and post-processing are done decentralised, in the engi-
neer’s domain-specific tools. In [10] J Larsson develops a technique for such co-
simulation and applies it to a model of a complete wheel loader. The present research
project will use this approach and further develop it.
This paper focuses more on the early stage of product development, but simulation is
equally beneficial in all stages. As T Larsson notes in [7], the more knowledge is
gained, the more the level of detail increases on the way from concept development
over detail design to product support. Being able to reuse most of the model in all stages
demands a modular approach. With that, individual modules can be replaced with up-
dated versions (e.g. for increasing the level of detail for one specific sub-system) with-
out having to make changes to the surrounding modules. Once the model of the com-
plete machine has been divided into modules, communication between them is then
only allowed through their public interfaces. Thus, an update of a module’s internal
structure does not affect the surrounding modules, since the interface is not changed. In
multi-body systems, such an interface usually consists of some markers and transfer
variables. But in some special cases, other descriptions of module interaction such as
surface contact or field interaction need to be modelled (if the simulation tool does not
provide a feature for this). Sinha et al. write in [11] about the need for modelling inter-
action dynamics between modules as modules of their own. In the present research pro-
ject though, this should be avoidable by carefully selecting the module interfaces so that
they are still physically motivated and intuitive, yet allow rigid coupling of markers and
variables.
A hierarchical modular approach combined with co-simulation will give the engineer
great flexibility regarding model creation. An engine module for instance could be
modelled as a true multi-body module with pistons, piston rods, crankshaft, rotational
damper, flywheel etc. (combustion pressure to be modelled as a variable force acting on
the piston). This would give the possibility to study torsional vibrations throughout the
drive train. Or the engine could just be approximated as a set of explicit MBS-equations
acting on a flywheel (i.e. no additional 3D-parts are modelled). Combining this with co-
simulation, the MBS-equations could be replaced by a Matlab/Simulink model of the
engine, which delivers the torque value to the multi-body system, where it is applied to
114 Paper I
the flywheel. In either case, other drive train modules will always connect to the fly-
wheel and thus not be structurally affected by switching between different engine mod-
ules. However, their behaviour and that of the complete system are both affected by the
switch, since the different engine modules in this example have different levels of de-
tail.
Looking at the relation of level of detail to modelling, Finn and Cunningham observe
in [12] that “the process of constructing a mathematical model is an inexact activity,
which is not supported by any underlying theory or governing laws”. Others go so far as
to equate mathematical modelling with art, which is quite an uncomfortable thought for
an engineer. Brooks and Tobias published a comprehensive overview in [13] and pro-
vide valuable guiding considerations on level of detail and complexity: “The purpose of
a complexity measure is to characterise the model so that this information can aid the
choice of model by predicting model performance. Ideally we would like to have a sin-
gle, system independent definition and measure of complexity covering all the aspects
of the level of detail of a model and applicable to all conceptual models. However, no
such definition or measure exists and as a result the term complexity itself is a source of
confusion due to its usage in many different contexts.” Such as using this term for any-
thing that someone finds difficult to understand. The authors further note that “a single
absolute measure of model performance cannot be obtained, but a meaningful compari-
son of alternative models in similar circumstances should be possible”. However a
search of the available literature reveals that little progress has apparently been made
since these observations (1996).
Another aspect of modular modelling is that flexibility regarding input data procure-
ment is gained. Planning to reuse an existing model by parameter update implies that
exactly the same type of input data is available for the update, as it was for the original
model. If a hydraulic pump has been modelled in great detail with hundreds of parame-
ters, then an update is difficult if all the engineer has got is a characteristics curve. “The
more detailed a model is, the more time must be spent on defining the parameters, based
on measurements or theoretical engineering reasoning”, notes Makkonen in [14]. Test
data might not be compatible due to different measurement set-ups, and theoretical rea-
soning requires the engineer to reason in a similar (if not identical) way to the creator of
the original model. In practice, it might prove more time and cost efficient to just de-
velop a new module from scratch, but with an identical interface.
Later on, assembly and replacement of modules could be done semi-automatically
using macros. However, it is dangerous to create too rigid procedures that let the engi-
neer out of the loop. In order to replace just one module in a system, the new module
has to have a similar level of detail, otherwise the entire system's behaviour is affected.
Which level of detail will be chosen for the various modules is dependent on the phe-
nomena to be studied; an experienced engineer's knowledge is vital in this decision
process and should not be overruled by some automatic routine.
5.2 Models
Naturally, since the goal is simulation of complete machinery for the purpose of analy-
sis and optimisation of overall performance (and related aspects), most of the machines
sub-systems and components need to be modelled in a modular way and in various lev-
els of detail. Despite some issues with input data procurement (see above), this is rela-
tively easy to achieve. The challenge lies in those parts of the system that are beyond
the company’s control: operator and environment.
The research on car driver models is significant and involves topics such as lane
change manoeuvres and path tracking. Recently, the research on driver models for ar-
ticulated heavy vehicles has increased. However, wheel loaders are a very different kind
of machinery and we are unlikely to find a suitable off-the-shelf operator model. Loader
operators need to maintain the delicate balance between hydraulics and drive train. In-
teraction with other road-users is non-existent (the only exception being the need to
match a certain time frame in order to co-operate effectively with operators of other
construction equipment, such as haulers or excavators). The driver model does not need
to include proper reaction to unforeseen events (parking car, red light etc.); and visual
orientation can be modelled roughly, because it can be assumed that a professional
driver knows the workplace very well. Nechyba presents his research in [15], demon-
strating how to model human control strategies from real-time human control data. His
application of these for a continuous model uses a flexible cascade neural-network
combined with extended Kalman filtering. For the discontinuous case, control actions
are modelled by individual statistical models. Kiencke et al. introduce a hybrid driver
model in [16], which describes the human perception process by discrete event tech-
niques and can be adapted to different driver types. The driver model, made in
SES/Workbench, was then evaluated in a co-simulation with a Matlab/Simulink vehicle
model.
The development of in-depth operator models is in itself a substantial task and will
be left out of the discussion in this paper. However, using the company’s extensive
knowledge of wheel loader operation in general, it is not too difficult to extract some
basic rules, which can, for example, be represented either by a model in Mat-
lab/Simulink or by means of IF/STEP-statements and timers directly in an MBS model.
Some promising results have already been obtained in this way.
Another important part of the overall system is a proper environment model. For
most work scenarios, this might be trivial to achieve. However, this is not the case for
bucket loading of gravel or other granular material. In [17] Singh presents the state-of-
the-art in automation of earth moving. Related to this are the studies made by Hemami
in [18] on the motion trajectory during bucket loading of an LHD-loader (Load-Haul-
Dump). A review of resistive force models for earthmoving processes is made by
Blouin et al. in [19]. In [20] Ericsson and Slättengren present a model for predicting
digging forces. This model, used at Volvo Wheel Loaders, is realised as a general force
subroutine in the multi-body system ADAMS and has been verified with measurements
of cylinder pressures from excavation of coarse gravel.
116 Paper I
5.3 Simulation Goals

Simulation-driven design implies using analysis to provide insight for synthesis. This
requires one to specify the goal of the conducted simulation. Besides the already men-
tioned expectations to reduce the cost and duration of product development, important
goals for Volvo Wheel Loader with this project are
- Optimisation of machine performance
- Optimisation of efficiency (with fuel efficiency as one aspect)
- Optimisation of operability (which driveability is just a subset of)
- Optimisation of a design’s robustness
These goals are at the same time overlapping and partly contradictory. In order to
solve this conflict, techniques for multi-objective optimisation will be needed – a re-
search area that Andersson gives an account of in [21]. This also includes a study on
achieving robust design by means of a genetic algorithm. A wealth of papers exists on
tolerance analysis and design of experiment techniques. Simpson et al. present in [22] a
review of such meta-modelling techniques as RSM (response surface methodology) and
Taguchi; but at the same time they warn that many applications using these methods are
“statistically questionable, because many analysis codes are deterministic in which error
of approximation is not due to random effects”.
Optimisation applied to a simulation of a complete machine like a wheel loader could
either be optimisation of the technical system or optimisation of the driver model, i.e. to
answer the question of how a machine of a given design should be operated to achieve
the highest fuel efficiency or highest performance. Optimisation of robustness is here
not limited to geometric tolerances, but includes variances of all possible parameters,
even such as workplace altitude (the barometric pressure heavily affects the response of
engines) or experience of the machine operator.
The term operability includes driveability as a subset, valid for the drive train only.
Operability, a term that later on in this project needs to be defined, considers the inter-
play of all sub-systems of the complete machine and includes such aspects as how well-
balanced an operator perceives a certain machine to be, or how efficiently a work cycle
can be performed. As touched upon in section 3, the operator’s subjective judgement
needs to be quantified in order to find an appropriate objective function for optimisa-
tion. Wicke presents his research on driveability and control aspects of vehicles with
Continuously Variable Transmissions in [23]. Insight gained there should be applicable
at least to the driveability-part of off-road machinery’s operability. Quantifying the lat-
ter also includes developing real-world test procedures that are compatible with those in
the virtual world, and vice versa.
6 Conclusion and Outlook

This paper presented the joint research by Volvo Wheel Loaders and Linköping Univer-
sity on simulation of complete machines for analysis and optimisation of overall per-
formance. The motivation on the side of the industrial partner is to develop products of
equally high performance, efficiency, and operability, but with more robustness regard-
ing these aspects, in a shorter time and at a lower total development cost. A revised
product development process (with regard to the research topic) has been proposed.
Examples of areas for future research have also been presented.
Research in the immediate future will focus on a definition of operability (including
quantification), as well as practical simulation problems (such as defining modules and
physically motivated and intuitive interfaces between them).
In a longer perspective, a measure of complexity needs to be developed. Research
will be done on how to optimise operability, which will also include a look at control
strategies.
References
[1] Larsson, J., Krus, P. and Palmberg, J.-O. (2001) “Modelling, Simulation and Vali-
dation of Complex Fluid and Mechanical Systems”. The Fifth International Con-
ference on Fluid Power Transmission and Control, Hangzhou, P.R. China, April
3-5, 2001.
[2] Illerhag, H. and Sjögren, F. (2002) “Study of Driveline Functionality during Off-
road Driving of an Articulated Hauler”. 15th European ADAMS Users' Confer-
ence, Rome, Italy, Nov 15-17, 2000.
http://www.mscsoftware.com/support/library/conf/adams/euro/2000/
Volvo_Driveline_Offroad.pdf
[3] Thomas, B. (2001) “Konzeption und Simulation eines passiven Kabinen-
federungssystems für Traktoren”. Doctoral thesis, Institut für Landmaschinen und
Fluidtechnik, Technische Universität Braunschweig, Braunschweig, Germany.
http://www.shaker.eu/en/content/catalogue/index.asp?lang=en&ID=8&ISBN=978-3-8265-
9563-9
[4] Pokras, V. (2002) “Multibody dynamic analysis of TI272 large mining haul
truck”. 2002 LMS Conference US, Troy (OH), USA, Sept 18-19, 2002.
[5] Frost, R. B. (1999) “Why Does Industry Ignore Design Science?” Journal of En-
gineering Design, Vol. 10, No. 4, Dec 1999, pp 301-304.
http://dx.doi.org/10.1080/095448299261218
[6] Campbell, R. M. (1997) “Smarter to Market: Why Automotive Engineering
Teams Use Virtual Prototypes in Design and Testing”. White Paper, Mechanical
Dynamics Inc.
118 Paper I
[7] Larsson, T. (2001) ”Multibody Dynamic Simulation in Product Development.

Performing Changes in Product Development”. Doctoral thesis, Division of Com-
puter Aided Design, Department of Mechanical Engineering, Luleå University of
Technology, Luleå, Sweden.
http://pure.ltu.se/portal/en/publications/multibody-dynamic-simulation-in-product-
development(f400f3f0-74d9-11db-962b-000ea68e967b).html
[8] Murphy, S. P. and Perera, T. (2002): “Successes and failures in UK/US develop-
ment of simulation”. Simulation Practice and Theory, Vol. 9, No. 6-8, May 2002,
pp 333-348.
http://dx.doi.org/10.1016/S0928-4869(01)00048-9
[9] Beskow, C. and Ritzén, S. (2000) ”Performing Changes in Product Development:
A Framework with Keys for Industrial Application”. Research in Engineering De-
sign, Vol. 12, No. 3, Dec 2000, pp 172-190.
http://dx.doi.org/10.1007/s001630050032
[10] Larsson, J. (2001) “Concepts for Multi-Domain Simulation with Application to
Construction Machinery”. Licentiate thesis, Department of Mechanical Engineer-
ing, Linköping University, Linköping, Sweden.
[11] Sinha, R., Paredis, C. J. J. and Khosla, P. K. (2002) “Behavioral Model Composi-
tion in Simulation-Based Design”. The 35th Annual Simulation Symposium, San
Diego (CA), USA, April 14-18, 2002.
[12] Finn, D. P. and Cunningham, P. (1998) “Solution reuse for model generation in
numerical simulation”. Artificial Intelligence in Engineering, Vol. 12, No. 3, July
1998, pp 297-314.
http://dx.doi.org/10.1016/S0954-1810(97)00028-9
[13] Brooks, R. J. and Tobias, A. M. (1996) “Choosing the Best Model: Level of De-
tail, Complexity, and Model Performance”. Mathematical and Computer Model-
ling, Vol. 24, No. 4, Aug 1996, pp 1-14.
http://dx.doi.org/10.1016/0895-7177(96)00103-3
[14] Makkonen, P. (1999) “On Multi Body Systems Simulation in Product Design”.
Doctoral thesis, Department of Machine Design, Royal Institute of Technology,
Stockholm, Sweden.
[15] Nechyba, M. C. (1998) “Learning and Validation of Human Control Strategies”.
Doctoral thesis, The Robotics Institute, Carnegie Mellon University, Pittsburgh
(PA), USA.
[16] Kiencke, U., Majjad, R. and Kramer, S. (1999) “Modeling and performance analy-
sis of a hybrid driver model”. Control Engineering Practice, Vol. 7, No. 8, Aug
1999, pp 985-991.
http://dx.doi.org/10.1016/S0967-0661(99)00053-2
[17] Singh, S. (1997) “The State of the Art in Automation of Earthmoving”. ASCE
Journal of Aerospace Engineering, Vol. 10, No. 4, Oct 1997, pp 179-188.
LHD-loader”. IEEE Transactions on Industry Applications, Vol. 30, No. 5, Sept-
Oct 1994, pp 1333-1338.
http://dx.doi.org/10.1109/28.315248
els for Earthmoving Processes”. ASCE Journal of Aerospace Engineering, Vol.
14, No. 3, July 2001, pp 102-111.
when working in gravel or other granulated material”. 15th European ADAMS Us-
ers' Conference, Rome, Italy.
http://www.mscsoftware.com/support/library/conf/adams/euro/2000/Volvo_
Predicting_Digging.pdf
[21] Andersson, J. (2001) “Multiobjective Optimization in Engineering Design - Ap-
plication to fluid Power Systems”. Doctoral thesis, Department of Mechanical En-
gineering, Linköping University, Linköping, Sweden.
http://www.lania.mx/~ccoello/andersson_thesis.pdf.gz
[22] Simpson, T. W., Poplinski, J. D., Koch, P. N. and Allen, J. K. (2001) “Metamod-
els for Computer-based Engineering Design: Survey and recommendations”. En-
gineering with Computers, Vol. 17, No. 2, July 2001, pp 129-150.
http://dx.doi.org/10.1007/PL00007198
[23] Wicke, V. (2001) “Driveability and Control Aspects of Vehicles with Continu-
ously Variable Transmissions”. Doctoral thesis, Department of Mechanical Engi-
neering, University of Bath, United Kingdom.
http://www.bath.ac.uk/mech-eng/auto/publications/
(Internet links updated and verified on August 17, 2011)

Paper II
Dynamic Simulation of Construction
Machinery: Towards an Operator Model
II
Reno Filla (1), Allan Ericsson (1), and Jan-Ove Palmberg (2)
VOLVO WHEEL LOADERS AB, ESKILSTUNA, SWEDEN
(1)
(2) DEPT. OF MECHANICAL ENGINEERING, LINKÖPING UNIVERSITY, SWEDEN
Abstract
In dynamic simulation of complete wheel loaders, one interesting aspect, specific for
the working task, is the momentary power distribution between drive train and hydrau-
lics, which is balanced by the operator.
This paper presents the initial results to a simulation model of a human operator.
Rather than letting the operator model follow a pre-defined path with control inputs at
given points, it follows a collection of general rules that together describe the machine’s
working cycle in a generic way. The advantage of this is that the working task descrip-
tion and the operator model itself are independent of the machine’s technical parame-
ters. Complete sub-system characteristics can thus be changed without compromising
the relevance and validity of the simulation. Ultimately, this can be used to assess a
machine’s total performance, fuel efficiency, and operability already in the concept
phase of the product development process.
Keywords: dynamic simulation, complex systems, operator model, driver model

Neo, sooner or later you're going to realize just as I did,
that there's a difference between knowing the path and walking the path.
(Morpheus in the film "The Matrix")

Filla, R., Ericsson, A. and Palmberg, J.-O. (2005) “Dynamic Simulation of Construction
Machinery: Towards an Operator Model”. IFPE 2005 Technical Conference, Las Vegas
(NV), USA, pp 429-438, March 16-18, 2005.

…Towards an Operator Model 123
1 Introduction
Originally spear-headed by large corporations of the automotive industry, dynamic
simulation of complete vehicles is increasingly practiced in the development of off-road
machinery. Handling and ride comfort are common simulation fields, as are perform-
ance and efficiency. As noted in [1], the challenge in simulating the complete system’s
dynamic behaviour is that in off-road machinery there are non-linear subsystems of
various technical domains (drive train, hydraulics, electronics, mechanics, etc.), which
are all tightly coupled. In the case of a wheel loader, the drive train and the hydraulics
are parallel systems; both are competing for engine torque, which is in limited supply.
Figure 1 shows how power is transferred through all relevant wheel loader subsystems,
with the machine being used in the typical working task of loading gravel.
Figure 1. Simplified power transfer scheme of a wheel loader loading gravel
As described in [1], the momentary distribution of engine power to both parallel

transfer paths is specific for the working task at hand, influenced by the environment,
and controlled by the operator – who ultimately balances the complete system.
Therefore, in order to evaluate the complete system’s performance, efficiency, and
operability, the simulation must not be limited to the machine itself, but has to include
operator, environment, and working task.
The results of a project aimed at simulating a wheel loader’s environment, i.e. dig-
ging forces when working in gravel, have already been reported in [2]. Similar projects
have also been documented in the literature, e.g. [3] and [4]. However, very little can be
found on adding models of human operators to complete machine models, and using
these to evaluate virtual prototypes in simulated working cycles. Mostly, current re-
search is aimed at automating tasks. This paper will instead focus on the development
of an operator model and a description of the working task, both sufficiently detailed to
draw conclusions about a machine's total performance, efficiency, and operability.
124 Paper II
2 Level of Detail
In this paper, “working task” will be defined as the summary of all descriptions of how
the simulated machine shall be operated in its environment. The “operator model” de-
scribes how the machine shall be controlled to accomplish the working task.
In the simulation of complete machines interacting with their environment, the ap-
proach to modelling of operator and working task will very much depend on the ques-
tions that the simulation is to provide the answers to.
In Figure 2 (from right to left) it is shown that with a more detailed operator model,
i.e. a more detailed model of a human being and their decision process, the description
of the working task can be simpler, and restrictions on how the machine is to be oper-
ated are transferred from the working task description to the operator model. Instead of
mere data, information is used – or even knowledge instead of mere information.
On the other hand, a less advanced operator model can to a certain degree be com-
pensated for with more information or even significantly more data (Figure 2, from left
to right). This will require less understanding of the total system but might still provide
insightful simulation results, depending on the context (i.e. simulation goals).
Figure 2. Relationship between operator model and working task description
When for example the aim of the simulation is to answer questions about mechanical
loads on certain components of a wheel loader, the working task could consist of data
for propeller shaft speed, steering wheel position, and current lengths of hydraulic lift
and tilt cylinders – all recorded during a real loading cycle. The operator model could
then be a simple PID controller that actuates the machine’s input devices (throttle,
brake, steering wheel, and hydraulic levers) in order to follow the recorded machine
movement. Such a scenario would be located in the right hand area of the solution space
in Figure 2. Performing an inverse simulation from the same set of working task data
would abolish the need for an operator model completely.
At the opposite end (left side), one can find simulations for evaluation of human-
related product properties like ergonomics or operability. Ideally, the operator as a sen-
sitive, decision-making, and strategically planning human being is modelled in great
detail. The working task could then be described in a very simple way, e.g. “load this
gravel on a truck”.
A decision for a specific level of detail can only be taken after careful consideration
of the scope of the assignment. As always, it is important to specify what type of ques-
tions the simulation needs to give answers to – and what can be ignored.
3 Literature Review
In [5] Vogel gives a comprehensive overview of the driver behaviour models used in
traffic simulation. She then combines some existing theories and develops her own,
control theory-based framework. The existence of a "mental model" of the modelled
system enables the driver to anticipate events and act preventively (instead of just me-
chanically reacting to target deviations). The author notes, "It is a very ambitious desire
to provide a complete model of driver behaviour, and any such attempt will certainly
provoke much criticism. On the other hand, not attempting at least to incorporate the
possibility to model all aspects of driver behaviour can be criticized, too."
Another, much simpler, micro-level model for traffic simulation [6] uses a cellular
automaton that incorporates mechanical restrictions (in the form of limited acceleration
and braking capabilities) and some form of human behaviour (modelled as the driver’s
excessive response to local traffic conditions). Other examples are reported in [7] and
[8]. Since simulation of traffic flow, just like road safety is a very active research area,
many more papers, reports, and dissertations can be found.
Finding suitable strategies and control schemes for robotic excavation is another very
active research area. Here, the focus is mostly on finding and following optimal bucket
trajectories. Long Wu’s approach to model bucket filling in [9] is motivated by his in-
tention to develop an autonomous wheel loader. Among other things, he discusses
bucket filling techniques and the shape of an optimal trajectory, in order to allow repeti-
tion in the following cycles. This latter aspect, giving autonomous machines tactical
capabilities, has been previously covered by Singh [10] in great detail. In [11] he gives
an update on the state of the art in 2002.
Another effort to automate the excavation process is reported in [12], which also fea-
tures an extensive literature review, which also includes Hemami’s early work [13]. Shi
et al. recognize in [14] that “modeling the dynamics of tool/soil interactions is very dif-
ficult and computationally intensive, and thus not practical for real-time and online exe-
cution.” Instead, they use fuzzy logic with very encouraging results.
126 Paper II
With today’s powerful computer hardware, Human-in-the-loop simulations are a fea-

sible scenario, avoiding the necessity to model human behaviour (but of course sacrific-
ing repeatability to a certain extent). One early example is reported in [15].
4 Development of the Wheel Loader Operator Model

Earlier projects within Volvo Wheel Loaders aimed at the development of a complete
machine model, covering the simulation domains mechanics, drive train, hydraulics,
and control system [16]. Having gained experience in projects involving multi-domain
simulation of complete vehicles, Volvo Wheel Loaders wanted to extend these simula-
tions to evaluate the potential operability of virtual prototypes, as well as total perform-
ance and fuel efficiency. These measures are heavily dependent on the way the human
operator uses the machine. The aim was to develop an operator model, which should be
as simple as possible with just the level of detail needed to reproduce relevant real-life
phenomena.
Using a pre-recorded loading cycle as working task description and an operator
model with a PID controller just following the cycle data (or even simpler: inverse
simulation) would have worked for one specific machine. However, the slightest change
in the setup (e.g. by altering torque converter characteristics) would either require new
recorded data (which do not exist yet for virtual prototypes), or would inevitably pro-
duce erroneous results, when the working task description of the original machine setup
was used. In [17] Zhang et al. correctly note: “The job is task-oriented, not reference-
oriented. The operator is not explicitly following any speed or position reference when
driving, steering and lifting. (…) The total productivity depends on how well the task is
fulfilled, what the fuel economy is, and how long it takes to finish each cycle. There-
fore, the performance and the efficiency of the human-machine interaction need to be
maximized.”
For such simulations, a human operator’s ability to adapt to a new machine is needed
to be reproduced by a model, preferably as simple as possible. Research in the areas of
cellular automata and self-organization has shown that complexity can and often does
emerge from quite simple, yet repetitive rules. In control theory, fuzzy logic is often
used to derive a controller from the simple, yet vague rules a human operator would
intuitively use for the same task. As stated earlier, Shi et al. have also reported on using
fuzzy logic for robotic excavation [14].
These examples motivated the use of techniques similar to fuzzy sets and discrete
events in the development of the operator model presented here. Since all subsystems of
the machine as well as the gravel pile were modelled in the 3D Multi-Body Simulation
package ADAMS (Figure 3), it was decided to also code the first version of the operator
model in this simulation package.
Other simulation programs are surely better suited to handle this specific task, but us-
ing ADAMS proved to be feasible. Relationships between state of the system and hu-
man action were continuously described using cubic polynomial STEP functions
(somewhat similar to fuzzy sets). Discrete states were saved using DIFF equations.
Figure 3. Models of the wheel loader and its environment, 3D view
All rules employed in the operator model were devised by studying and interviewing
professional wheel loader operators. In [18] Gellersted documents a related study.
In the simulation, the operator model controls the machine model by engine throttle,
lift and tilt lever, steering wheel, and brake only – just as a human operator does. Also,
only signals that a human operator can sense are used in the model (excluding e.g.
torque converter slip or hydraulic pump displacement). In its first version, the operator
model has no tactical capabilities; it does not plan ahead of the current cycle by alternat-
ing the bucket fill location, like a human operator would do. The next sections will fo-
cus on the bucket filling phase itself, which is the most operator-dependent phase in a
wheel loader’s working cycle. All other phases have also been modelled, but will not be
discussed in depth.
5 Bucket Filling
As mentioned, the literature reveals several different approaches to both modelling soil
or granular material, and describing the optimal trajectory for a tool cutting through
these materials. For this first version of the operator model, a less complex strategy was
chosen: it was reasoned that the most efficient way to fill a bucket should be to move it
upwards through the gravel pile on a velocity vector with a bearing δ that matches the
pile’s slope angle ε (Figure 4). Of course, this will only be the case towards the end of
the scooping, since the process starts with the bucket’s cutting edge being parallel to the
ground. Thereby, one specific bucket filling method is imitated, where the operator fol-
lows the slope of the gravel pile, instead of just forcing the bucket into the pile and tilt-
ing backwards.
At the same time, the bucket’s cutting edge should remain at a certain angle of attack
γ relative to the bucket’s velocity vector, and the bucket’s bottom at a certain angle of
clearance α relative to the gravel pile.
128 Paper II
Figure 4. Bucket filling approach
The operator model needs to simultaneously control the machine speed (via the en-
gine throttle) and bucket lift and tilt functions (via hydraulic levers) in order to satisfy
these and a number of other requirements. In the simulation, the bucket filling phase
starts when the bucket’s cutting edge penetrates the gravel pile at a certain, controlled
speed. This causes a down-shift into gear 1 and the repeated execution of the following
rules (not necessarily in order):
• “Traction control 1”: When relative wheel slip exceeds a certain value, the
maximum throttle value will be ramped down. This decreases torque converter
slip and thus reduces traction force.
• “Traction control 2”: Above a certain limit for integrated relative wheel slip,
the lift function is ramped up. This increases the load on the front wheels,
which improves traction.
• “Bucket velocity vector control”: Deviation of δ from ε above a certain thresh-
old will lead to a ramp-up of the maximum throttle value. This ensures that the
bucket follows the gravel pile’s slope.
• “Bucket attitude control”: The tilt function is ramped up in order to maintain α
and γ as the lifting unit is raised and the machine is driven forward.
• “Exit trigger 1”: Above a certain angle of the bucket relative to the gravel
pile’s slope, the tilt function will be fully activated until the bucket is com-
pletely tilted back.
• “Exit trigger 2”: Above a certain angle of the lifting unit, the tilt function will
be fully activated until the bucket is completely tilted back.
The results of rules governing the same operator input will either be multiplied or
added to calculate the total input value. The bucket filling phase ends when the bucket
has left the gravel pile, which will result in a gear shift into R2 (reverse) and full activa-
tion of the lift function.
6 Simulation Results
Figure 5 shows the results of one specific simulation, using the above rules for bucket
filling:
Figure 5. Bucket filling: operator input and simulation result
The upper diagram shows the x and y positions of the cutting edge and its global an-
gle. The latter, visualized as sloped lines with attached values, is drawn each 0.5 sec-
onds, which gives a feeling of the speed of the process. The lower diagram (which uses
the same x axis) illustrates the operator input for engine throttle, lift, and tilt function
(all calculated by merging the results of the previously defined rules).
Figure 6 shows the operator input during the complete loading cycle: engine throttle,
brake, and steering wheel in the upper diagram, lift and tilt function in the lower one:
130 Paper II
Figure 6. Complete loading cycle: operator input
As explained in the introduction, the motivation for the development of a sufficiently

detailed operator model (and a description of the working task) was to be able to draw
conclusions about a machine's total performance, efficiency, and operability by simulat-
ing virtual prototypes, rather than testing physical ones.
Of particular interest is how a wheel loader’s engine power is being split up between
drive train and hydraulics over a complete loading cycle. Figure 7 illustrates this for one
of the conducted simulations:
Figure 7. Complete loading cycle: power distribution to hydraulics

and drive train (total engine power = top of the black area)
The engine’s response and fuel consumption can vary dramatically depending on the
specific combination of torque and speed that accomplished power. It is therefore im-
portant to analyze the engine’s load duty (as shown in Figure 8):
Figure 8. Complete loading cycle: engine load duty (normalized)
This pattern is remarkably similar to the results obtained in tests of physical proto-
types, and indicates that the developed operator model can be useful. However, the goal
is to be able to perform virtual tests of not yet physically realized machine configura-
tions. It must therefore be proven that the operator model can adapt. The wheel loader
model’s torque converter has thus been changed to allow this. This component is known
to have a vital impact on such important complete machine properties as performance
and fuel efficiency. In the simulation, a “weaker” torque converter was chosen. In order
to obtain the same traction force, a “weaker” converter requires higher slip between
pump and turbine, which leads to higher engine speeds over a complete loading cycle.
132 Paper II
A human operator compensates for this with higher throttle values. But this also affects
the hydraulic system, which in turn requires compensation. Figure 9 shows that the op-
erator model managed to adapt to the new machine characteristics by running the en-
gine at higher speeds – just as a human operator would have done:
Figure 9. Complete loading cycle: engine load duty for machines

equipped with different torque converters
The simulations conducted also replicate the fact that, in addition to higher engine
speeds, a “weaker” torque converter also leads to higher fuel consumption and longer
cycle times.
7 Discussion
In all conducted simulations, the operator model shows reasonably correct behaviour:
the results for bucket filling, power distribution, engine load duty, and the ability to
adapt to different torque converter characteristics all indicate that the model can be use-
ful for testing total performance and fuel efficiency of virtual prototypes. However, the
operator model in its first version clearly needs more development and validation.
In general, extracting controller rules from human operators through interviews and
implementing them as min/max relationships or fuzzy sets is not a novelty. But this
paper describes a new field of application with the ambition to connect to on-going
work on quantification of a human operator’s perception of a machine’s operability.
Applications in other areas, like active power distribution in hybrid vehicles, seem to be
an interesting prospect.
Simulating operability is much harder than simulating total machine performance
and fuel efficiency, as a generally agreed-upon definition of operability (especially for
wheel loaders) is still missing. However, it should be possible to utilize the results of
existing research into “mental workload”, possibly connecting an operability measure
with the operator’s efforts to control the power distribution between hydraulics and
drive train. If such a measure could be found, then simulation would be of great assis-
tance in optimizing machine characteristics for example for maximum efficiency or
robust operability (the latter both regarding component tolerances, varying environ-
mental influences, but also different operator skills). One approach to quantification
might be through the definition of “operator input dose” similar to vibration dose value
in the assessment of whole-body vibration exposure.
Realizing the operator model as a set of equations in ADAMS proved to be possible,
but cumbersome. For that type of problem, realization as a finite state machine in a dis-
crete-event simulation package (in co-simulation with ADAMS) would have been bet-
ter. This will be done in future work.
In the second chapter, the diametric relationship between operator model and work-
ing task description is discussed to a certain extent, but the following text seems to deal
only with the operator model. This is because no distinction was made between the two
in the code of the first prototype. The original idea was that with more knowledge, the
working task description (“What to do?”) transitions from demanding to describing,
while the operator model (“How to do it?”) changes from executing to planning. In this
specific case, it has been found to be impractical to develop separate, yet linked models
for these two. This could be so in other cases as well, turning the distinction of operator
model and working task description more into a thought construct than a useful concept.
On the other hand, similar concepts in other fields like computer science (object-
oriented programming) or business management (project vs. line organizations) allow
exceptions without dismissing the whole concept as totally irrelevant.
8 Conclusion
A first version of a rule-based operator model has been developed, that shows good
potential for introducing “a human element” into dynamic simulation of complete wheel
loaders. With this, more relevant answers can be obtained with regard to total machine
performance and fuel efficiency in complete loading cycles. This can be used to signifi-
cantly support the product development process by substituting many tests of physical
prototypes with equivalent tests of virtual prototypes. However, using dynamic simula-
tion to assess operability of complete machines still requires more work.
Acknowledgements
The financial support of Volvo Wheel Loaders AB and PFF, the Swedish Program
Board for Automotive Research, is hereby gratefully acknowledged.
Our sincere thanks are also due to the many people at Volvo and Linköpings Univer-
sitet without whose theoretical and practical support the work presented in this paper
would not have been possible.
134 Paper II
References
[1] Filla, R. and Palmberg, J.-O. (2003) “Using Dynamic Simulation in the Develop-
ment of Construction Machinery”. The Eighth Scandinavian International Confer-
ence on Fluid Power, Tampere, Finland, Vol. 1, pp 651-667.
Volvo_Predicting_Digging.pdf
els for Earthmoving Processes”. ASCE Journal of Aerospace Engineering, vol. 14,
no. 3, July 2001, pp 102-111.
[4] Tan, C. P., Zweiri, Y. H., Althoefer, K. and Seneviratne, L. D. (2005) “Online Soil
Parameter Estimation Scheme Based on Newton-Raphson Method for Autono-
mous Excavation”. IEEE/ASME Transactions on Mechatronics, vol. 10, no. 2,
April 2005, pp 221-229.
http://dx.doi.org/10.1109/TMECH.2005.844706
[5] Vogel, K. (2002) “Modeling Driver Behavior – A Control Theory based Ap-
proach”. Doctoral thesis, Department of Mechanical Engineering, Linköping Uni-
versity, Linköping, Sweden.
[6] Lee, H. K., Barlovic, R., Schreckenberg. M. and Kim, D. (2004) “Mechanical re-
striction versus human overreaction triggering congested traffic states”. eprint
arXiv:cond-mat/0404315
http://arxiv.org/abs/cond-mat/0404315
[7] Bengtsson, J. (2001) “Adaptive Cruise Control and Driver Modeling”. Licentiate
Thesis, Lunds Institute of Technology, Lund, Sweden.
http://www.cs.clemson.edu/~johnmc/courses/cpsc875/resources/acc/6.pdf
[8] Macadam, C.C. (2003) “Understanding and Modeling the Human Driver”. Vehicle
System Dynamics, vol. 40, pp 101-134.
[10] Singh, S. (1995) “Synthesis of Tactical Plans for Robotic Excavations”. Doctoral
thesis, Carnegie Mellon University, Pittsburgh (PA), USA.
[11] Singh, S. (2002) “State of the Art in Automation of Earthmoving, 2002”. Work-
shop on Advanced Geomechatronics, Sendai University, Japan.
Experiments, Modelling, Identification and Control”. Master Thesis, Queen's Uni-
versity, Kingston (Ontario), Canada.
Oct 1994, pp 1333-1338.
http://dx.doi.org/10.1109/28.315248
[14] Shi, X., Wang, F.-Y. and Lever, P. J. A. (1996) “Experimental results of robotic
excavation using fuzzy behavior control”. Control Engineering Practice, Vol. 4,
No. 2, Feb. 1996, pp 145-152.
http://dx.doi.org/10.1016/0967-0661(95)00220-0
[15] Grant, P., Freeman, J. S., Vail, R. and Huck, F. (1998) “Preparation of a Virtual
Proving Ground for Construction Equipment Simulation”. 1998 ASME Design
Engineering Technical Conferences, Atlanta (GA), USA, Sept.13-16, 1998.
http://members.asme.org/catalog/ItemView.cfm?ItemNumber=I416CD
Construction Machinery”. Licentiate thesis, Department of Mechanical Engineer-
ing, Linköping University, Linköping, Sweden.
Earthmoving Vehicle Powertrain Experimentally Validated in an Emulated Work-
ing Cycle”. Conference paper, ASME 2003 International Mechanical Engineering
Congress and Exposition.
[18] Gellersted, S. (2002) ”Manövrering av hjullastare (Operation of Wheel Loaders)”.
Technical report, JTI – Institutet för jordbruks- och miljöteknik, Uppsala, Sweden.
(Internet links updated and verified on August 17, 2011

Paper III
An Event-driven Operator Model for
Dynamic Simulation of Construction Machinery
Reno Filla
VOLVO WHEEL LOADERS AB, ESKILSTUNA, SWEDEN III
Abstract
Prediction and optimisation of a wheel loader’s dynamic behaviour is a challenge due to

tightly coupled, non-linear subsystems of different technical domains. Furthermore, a
simulation regarding performance, efficiency, and operability cannot be limited to the
machine itself, but has to include operator, environment, and work task.
This paper presents some results of our approach to an event-driven simulation
model of a human operator. Describing the task and the operator model independently
of the machine’s technical parameters, gives the possibility to change whole sub-system
characteristics without compromising the relevance and validity of the simulation.
Keywords: discrete simulation, continuous simulation, complex systems, operator

model, driver model
Every passing minute is another chance to turn it
all around.
(from the film “Vanilla Sky”)

Filla, R. (2005) “An Event-driven Operator Model for Dynamic Simulation of
Construction Machinery”. The Ninth Scandinavian International Conference on Fluid
Power, Linköping, Sweden.

An Event-driven Operator Model… 139
1 Introduction
In the development of off-road machinery, dynamic simulation of large and complex
technical systems is being increasingly practised. In the case of a wheel loader, most
subsystems are non-linear and tightly coupled, which makes prediction and optimisation
of the complete system’s dynamic behaviour a challenge. Both drive train and hydrau-
lics are competing for the limited engine torque. As described in [1], the momentary
power distribution is specific for the task at hand and is controlled by the operator, who
ultimately balances the complete system. Figure 1 shows how power is transferred
through all relevant wheel loader subsystems, with the machine being used in the typi-
cal work task of loading gravel.
Figure 1. Simplified power transfer scheme of a wheel loader loading gravel
So-called full vehicle models can be useful in various types of simulation. How the
model needs to be controlled and subjected to virtual loads is very much dependent on
the purpose of the simulation. Paper [2] gives examples and reasons of different ap-
proaches.
Our work aims to simulate a wheel loader’s total performance, efficiency, and oper-
ability, and to investigate the robustness of these three important product properties.
Because a human operator adapts to the machine and the work situation, assessing the
above mentioned properties by means of simulation also requires modelling operator,
environment, and task at an appropriate level of detail.
Traditional backwards simulation forces a technical system to follow a prescribed
cycle, which can result in physically erroneous results. Forward simulation with fixed
operator input would use each machine in the same way, producing physically correct
but irrelevant results. An experienced operator can adapt to a new machine in order to
achieve the highest possible performance. Being able to simulate this adaptation to a
certain degree is a necessary first step towards optimisation of the complete system.
140 Paper III
Paper [2] describes the initial results of our work on an operator model, with a focus
on the bucket filling phase. Since the complete loader was modelled in a program for
multi-body simulation (Figure 2), our first approach was to also realise the operator
model in this simulation package. As noted in the discussion section of that paper, this
proved to be possible but cumbersome. It seemed that for this type of problem, realisa-
tion as a finite state machine in a discrete-event simulation program (in co-simulation
with the multi-body simulation package) would have been better.
Consequently, this was our next step. Choosing a program suite that supports both
continuous and discrete event simulation, the operator can now be modelled with a vari-
ety of control strategies, completely separate from the machine. A simulation can be
performed as either co-simulation (where both sides solve their own equations and
communicate the results) or as a plant import into the multi-body simulation program
(where the operator model has been compiled into an executable file, which is refer-
enced as a General State Equation).
Figure 2. Models of the wheel loader and its environment
With the bucket filling phase already covered in [2], this paper focuses on the re-
maining essential elements of a wheel loader working cycle.
2 Loading Cycle Description

Wheel loaders are highly versatile machines, and with each task and workplace being
unique, it is difficult to define a standard test cycle that covers every possible aspect.
However, tasks do exist that are more common than others. Figure 3 shows a so-called
short loading cycle, sometimes also dubbed V-cycle or Y-cycle for its characteristic
driving pattern. Typical for this cycle is the loading of some kind of granular material
on an adjacent load receiver (e.g. a dump truck).
In paper [3], such a short loading cycle is divided into different phases, and some ex-
amples are given for especially interesting situations that occur in some phases.
Phase 1, Bucket filling, begins as soon as the bucket’s cutting edge is in contact with
the material. The operator then uses the kick-down function to shift into the lowest gear.
Lifting the loading unit somewhat increases force on the front axle, which improves
traction. The operator then simultaneously controls the machine speed (via the engine
throttle) and lift and tilt function (via hydraulic levers) in order to fill the bucket. Again,
this has already been covered in [2] and will not be considered in this paper.
Phase 2, Leaving bank, begins after the bucket is fully tilted back. The lift function is
activated, the transmission is put into reverse, and the engine speed is increased. After
leaving the pile, the machine is steered to the side in the characteristic V-pattern. The
lifting function is activated all the time until reaching the load receiver at the end of
phase 5.
Phase 3, Retardation, begins when the operator judges that the remaining distance to
the load receiver is sufficient for the lift hydraulics to accomplish the necessary bucket
height while driving (operators usually prefer driving a longer distance to waiting in
front of the load receiver for a sufficient bucket clearance). Most operators use engine
braking for retardation.
Reversing in phase 4 is unfortunately usually accomplished by putting the transmis-
sion into forward while the machine is still moving backwards. This is convenient, but
increases fuel consumption, since the torque converter’s turbine wheel is forced to ro-
tate backwards until enough torque has been built up and the machine is moving for-
wards.
142 Paper III
Phase 5, Towards load receiver, begins when the machine speed is positive, i.e. the
machine is moving forward. The operator steers the loader to achieve the V-pattern that
is characteristic for a short loading cycle. At the end of this phase, the loader arrives
perpendicularly at the load receiver. If the operator judged correctly in phase 3, then the
bucket has reached a sufficient height to begin unloading. Usually this is the case, be-
cause an experienced operator is able to adapt to any machine very quickly.
In phase 6, Bucket emptying, the operator often drives forward very slowly while at
the same time raising the loading unit and tilting the bucket forward. This gives the pos-
sibility to not just dump the material on the load receiver, but actually place it so that
after 3 to 4 buckets a dump truck is evenly loaded without any material being spilled.
In paper [3], the short loading cycle continues with phases 7-10: Leaving load re-
ceiver, Retardation and reversing, Towards bank, and Retardation at bank (the last is
often combined with the next bucket filling by using the machine’s momentum to drive
the bucket into the gravel pile, rather than applying traction with the wheels). In this
stage of our work, however, only phases 1-6 are covered, because of the significant in-
teraction between all main subsystems.
The challenge in such a short loading cycle, from a manufacturer’s point of view, is
to find an appropriate, robust, and maintainable balance between productivity (material
loaded per time), efficiency (fuel consumed per load), and operability over the complete
cycle.
Productivity and efficiency are well-defined, but a generally agreed definition of op-
erability has still to be found. In [4], a definition is offered, that also works well even
for construction machinery: “Operability is the ease with which a system operator can
perform the assigned mission with a system when that system is functioning as de-
signed”. The limitation to states where the system is functioning as designed, effectively
excludes properties like robustness and reliability. But this still gives no explanation as
to how to quantify operability; instead, the concept “ease of performance” is introduced.
Paper [3] briefly discusses some ways of visualising operability-related, measurable
state variables in a wheel loader cycle. Amongst others, a diagram displaying bucket
height over integrated machine speed (i.e. the machine’s travelling distance) is intro-
duced (Figure 4).
The ratio of lift speed to machine speed, which is crucial to how long the loader
needs to be driven backwards until reversing, can here be seen as the curve’s slope.
In the diagram in Figure 4 the curve is fairly straight from the beginning of phase 2
(where the loader leaves the bank) to the end of phase 5 (where the loader arrives at the
load receiver). This indicates that the operator judged very well when to reverse. Oth-
erwise, the slope would become steeper at the end of phase 5, indicating that the opera-
tor needed to slow down or even stop the loader in order for the bucket to reach suffi-
cient height for emptying.
The distance between the point of reversing and the bank (or load receiver) is an in-
dication of how well a balance between the two motions of lifting and driving has been
achieved. Because such a diagram points out any imbalance, engineers at Volvo CE
have begun to call it Machine harmony diagram.
Figure 4. Bucket height over integrated machine speed in a short loading cycle
Another interesting aspect is the visualisation of the operator’s technique for bucket
filling and bucket emptying. For this, the machine harmony diagram is preferably ac-
companied by another diagram displaying bucket height over bucket angle (Figure 5).
This additional diagram also indicates the loader linkage’s capability for parallel align-
ment (i.e. how much the bucket angle changes during raising and lowering of the load-
ing unit).
Figure 5. Example of bucket handling in a short loading cycle
The chosen bucket filling technique indirectly influences where the loader needs to
reverse: Leaving the bank at the beginning of phase 2 with a higher bucket means that it
will take less time to raise it from there to a sufficient height for emptying. As explained
before, an experienced operator chooses the point of reversing so that the bucket will
have the necessary height approximately when the loader reaches the load receiver. This
144 Paper III
means that in theory the point of reversing will be nearer to the bank when leaving with
a higher bucket position, because the operator adapts to the machine.
It can be easily shown that this is the case also in practice. Figure 6 shows three se-
lected short loading cycles, originating from a non-stop recording of a longer period,
during which the operator changed his bucket filling technique. As can be seen, the
bucket height when leaving the bank varies, but the operator managed very well to
chose a reversing point in order to reach the load receiver with the bucket at a sufficient
height for emptying.
Figure 6. Recorded short loading cycles with different bucket filling techniques
Again, it is this kind of human adaptation to the machine which is the reason for our
work on operator models: we want to be able to assess a new machine’s productivity,
efficiency, operability, and their robustness by means of simulation.
3 Simulation Setup
In previous work, the author and two colleagues modelled a wheel loader in a simula-
tion suite for continuous multi-body simulation. The model features a detailed hydraulic
system, connected to a fairly detailed mechanical model (rigid bodies) and drive train.
The engine, with engine controller and gas dynamics, was realised as state variables,
enhanced by user-written subroutines that are linked to the numerical solver. The envi-
ronment is represented as a model of a gravel pile [5], and as a standard tyre/road
model.
Since using the loader model for the simulations presented in [2], the interface to the
operator model has been enhanced. Already in the previous version, the operator model
was logically separated from the machine model, which was controlled through vari-
ables representing the functions engine throttle, brake, steering, lift, and tilt. However,
everything was still located within one model space. Now, by using co-simulation, the
separation is complete: the wheel loader is modelled and simulated in the multi-body
simulation program, while the operator is completely modelled and simulated in the
previously mentioned package that supports both continuous and discrete event simula-
tion. The operator model sends commands through the machine controls, which the
loader model receives, uses and answers with e.g. loader position and orientation, and
engine speed (Figure 7). Some additional elements have been added to simplify debug-
ging.
Figure 7. Simulation setup (overall view)
Both sides act as black box models towards each other. No internal states are sent,
because the intention is to model human-machine interaction, which naturally occurs
through a limited number of channels. As long as the interface is unchanged, the models
on either side can be replaced by just copying a different file into the simulation's work
directory.
In the near future, our ambition is to compile operator models into executable files
that can be imported as General State Equations into the multi-body simulation pro-
gram. This will significantly simplify the simulation process. Initial tests in that respect
have shown very promising results.
146 Paper III
4 Event-based Operator Model

Figure 8 shows the overall view of the operator model, where each phase has been mod-
elled as a state containing several substates.
Figure 8. Operator model (top level view)
Paper [2] mainly focussed on modelling one bucket filling technique, with encourag-
ing results. Therefore, the operator model presented in this paper has been designed to
begin where the bucket filling phase in [2] ended, immediately before tilting back the
bucket and leaving the gravel pile. An initialisation phase has been modelled (Figure 9),
during which the bucket is lifted and tilted to the same position that resulted in [2]. The
engine is also set to the same speed. Already during model preparation the loader has
been moved to the same position it had at the end of bucket filling in [2].
Figure 9. Initialisation phase
As described earlier, at the end of phase 1, Bucket filling, the operator tilts the
loader's bucket back as far as possible. In the model, this is done in the intermediate
phase 1a.
In the next phases the operator will begin by calculating and following a theoretical
path, which is only defined by the location of bank and load receiver. Later on, the op-
erator model will diverge from this path to react to certain events. For all calculations of
position and orientation, a global co-ordinate system with its origin in the intersection of
bank and load receiver is used (Figure 10a).
Figure 10a. V-pattern, theoretical Figure 10b. Theoretical path solutions
The theoretical path begins with a straight line that is perpendicular to the bank. It
also ends with a straight line, this one perpendicular to the load receiver. In between, the
path consists of two circular arcs that are tangential to each other and the two lines.
Such a geometrical figure has one degree of freedom, which can be expressed as the
angle of the tangent that is shared by the arcs. Figure 10b shows four possible solutions.
Which solution to take is dependent on how much space the workplace offers. For a
given angle α the circular arcs’ radii ra and rb result in:
1 ⎛ (a + b ) (1 + cos α ) ⎞
ra = ⎜ − (a − b )⎟ (1)
2⎝ sin α ⎠
1 ⎛ (a + b ) cos α ⎞
rb = ⎜⎜ + (a − b )⎟⎟ (2)
2 ⎝ (1 − sin α ) ⎠
Many operators prefer to steer the loader approximately equally much to the right as
to the left. This corresponds to Figure 3 in which the global tangent orientation is stated
to be α ≈ 45°. Equations 1 and 2 can then be simplified to:
a+b
ra = +b (3)
2
148 Paper III
rb =
(
a − b 1− 2 ) (4)
2− 2
However, for the model presented in this paper, the reasoning was that an operator
prefers to aim at a fixed point while reversing. This is easier to accomplish and also
results in a 45 degree angle when the global origin is used as fixed point and when, as in
most cases, load receiver and bank are located equidistantly from the global origin. The
circular arcs’ radii and the global tangent orientation can be calculated as:
(a + b ) (a + b )2 + 4ab − (a 2 − b 2 )
ra = b + (5)
4a
(a + b ) (a + b )2 + 4ab + (a 2 − b 2 )
rb = a + (6)
4b
a + ra
cos α = (7)
ra + rb
The case of “aiming at the origin” can easily be checked by comparing β, the
loader’s bearing to the origin with θ, its orientation in the global co-ordinate system
(Figure 11).
Figure 11. Global position, orientation θ and bearing β
Phase 2, Leaving bank, begins with putting the transmission into reverse. Then, full
lift and the machine’s steering function are activated until an articulation angle is
reached that corresponds to a radius according to Eq. 5 (i.e. plan to reverse with aim at
origin). However, due to the lack of feed-back path control in this version of the opera-
tor model, the machine model will not follow the theoretical path exactly. But this is not
necessary either, because another rule is activated when the machine prematurely aims
at the origin. This rule overrides the original path and steers the machine to drive along
the new-found bearing line.
As soon as the loader passes the load receiver, an estimation routine begins to con-
tinuously extrapolate the current lifting/driving ratio in order to determine whether the
reversing phase can be entered. The remaining distance to the load receiver is calculated
as a combination of a circular arc with radius rc and a straight line (Figure 12).
Figure 12. Path to load receiver
Knowing the machine’s current position (x, z) and global orientation θ, the turning
radius rc and the remaining travelling distance L can be found as:
d
rc = (8)
1− sin θ
⎛π ⎞
Lc = rc ⎜ − θ ⎟ (9)
⎝ 2 ⎠
Ld = z − rc cos θ (10)
L = Lc + Ld (11)
However, most wheel loaders have a maximum articulation angle of 35-40 degrees
and might therefore not be able to achieve the currently required turning radius rc. To
handle this situation in a simulation, an intermediate phase 2a is activated. During this
phase, the operator model continues the V-pattern from phase 2 and conducts calcula-
tions according to Eq. 8-11. As soon as the required turning radius rc is actually achiev-
able, phase 2a is left. It is advantageous to handle this situation in a special phase; oth-
erwise, an extended phase 2 might be interpreted as a limited lifting speed of the loader
150 Paper III
rather than a limited turning capability.

Immediately after activation of phase 3, Retardation, the engine throttle is released
and the articulation steered back to zero. In addition to the engine, the service brakes are
also used to decrease the machine’s speed.
Phase 4, Reversing, begins when the machine’s speed has decreased to a level that
where setting the transmission to forward is safe. The brakes are released, the engine
throttle is activated again, and lifting is stopped until sufficient speed has been built up.
In phase 5, Towards load receiver, the required turning radius rc is again calculated
with Eq. 8. However, a theoretically correct but impractical solution to Eq. 8-11 might
be found that yields a negative value for Ld, the length of the straight line between load
receiver and circular arc (Figure 12). One possible strategy is to just move forward in a
straight line until the current machine position and global orientation result in a positive
value for Ld. When the machine finally reaches the load receiver, the articulation angle
is adjusted so that the bucket is parallel to the load receiver, regardless of the wheel
loaders orientation. However, in a loading cycle that is not too short, the complete
wheel loader should arrive with zero articulation and perpendicular to the load receiver.
During phase 5, the lifting function is activated until the bucket is at a sufficient
height for emptying. If this is not the case when the machine reaches the load receiver,
an intermediate phase 5a is started before the bucket is emptied in phase 6. In this in-
termediate phase, the operator model is using the lift function to reach a bucket height
that is sufficient for emptying into the load receiver.
In phase 6, Bucket emptying, the operator model uses engine throttle, lift, and tilt
function simultaneously in order to simulate a human operator’s strategy of placing ma-
terial so that the load receiver is evenly loaded after three to four short loading cycles.
As explained earlier, a short loading cycle continues with phases 7-10, Leaving load
receiver, Retardation and reversing, Towards bank, and Retardation at bank, which are
not covered yet in the presented operator model.
5 Simulation Results
In order to give answers with respect to a simulated wheel loader’s productivity, effi-
ciency, and operability, the operator model needs to be updated to include all phases of
a short loading cycle, most importantly bucket filling. Until this is done, simulation re-
sults in respect of engine load duty and fuel consumption should be considered with
care. For this reason this paper will only present how the operator model succeeded in
adapting to variations in workplace layout and machine properties.
Figure 13 shows the results of a short loading cycle that has the same workplace lay-
out as the cycle presented in paper [2]. Compared to the model in [2], this one repre-
sents an operator with a very rough operating style; most control inputs are either on or
off, and the engine throttle is very often at maximum value.
Figure 13. Control input signals from the operator model
The diagram on the bottom displays the current cycle phase (#0 is the initialisation
phase).
152 Paper III
Figure 14 displays the machine movement from the cycle above in both a location
plot and a harmony diagram. Since phase 1, Bucket filling, is not included in the current
version of the operator model, the cycle in the harmony diagram starts with a vertical
line that shows how the loader model lifts the bucket in the simulation’s initialisation
phase.
Figure 14. Machine location plot and harmony diagram
Because the operator model has not been hard-coded with fixed locations for bank
and load receiver, but designed parametrically, it is able to adapt. Figure 15 shows this
for two different load receiver positions.
In the initial simulation setup (Figure 13 and Figure 14) the machine’s movement
was more restricted by its limited turning capability, rather by its lifting speed. The
simulated wheel loader’s control signal has therefore been artificially decreased by
50%, which results in a slower lifting speed. Figure 16 shows how the operator model
adapts to this situation by choosing a different reversing point.
Figure 15. Adaptation to workplace layout
Figure 16. Adaptation to lifting speed

154 Paper III
Finally, as already explained earlier and shown in Figure 6, an operator’s bucket fill-
ing technique has an indirect influence on the location of the reversing point. With a
higher bucket position at the beginning of phase 2, the point of reversing can be chosen
nearer to the load receiver. Figure 17 presents results of two simulations that start with
different bucket heights (in both cases the loader model with artificially decreased lift-
ing speed was used).
Figure 17. Influence of bucket filling technique
6 Discussion
While the operator model presented in paper [2] focussed on the bucket filling phase
and used min/max relationships reminiscent of fuzzy-sets, the operator model presented
here focuses on the remaining elements of a short loading cycle and uses a pure dis-
crete-event approach. It is advantageous to use this modelling paradigm, but certain
elements such as path control are probably best modelled with a simple PI-controller.
Therefore, in the next version of this operator model the possibility of combining these
approaches needs to be explored. Because the chosen simulation package supports both
continuous and discrete-event simulation, this is a feasible prospect.
As mentioned earlier, one interesting aspect is the possibility of compiling the opera-
tor model into an executable file, which is referenced as a General State Equation in the
multi-body simulation program. With this, the engineer conducting a simulation can
focus on the wheel loader, rather than having to spend time defining how it needs to be
controlled to mimic reality.
7 Conclusion
An event-driven operator model has been developed, which is able to adapt to basic
variations in workplace layout and machine capability. With this, a “human element”
can be introduced into dynamic simulation of complete wheel loaders, giving more
relevant answers with respect to total machine performance, fuel efficiency, and possi-
bly even operability in complete loading cycles.
Acknowledgements
The financial support of Volvo Wheel Loaders AB and PFF, the Swedish Program
Board for Automotive Research, is hereby gratefully acknowledged.
References
[2] Filla, R., Ericsson, A. and Palmberg, J.-O. (2005) “Dynamic Simulation of Con-
struction Machinery: Towards an Operator Model”. IFPE 2005 Technical Confer-
ence, Las Vegas (NV), USA, pp 429-438.
[3] Filla, R. (2003) “Anläggningsmaskiner: Hydrauliksystem i multidomäna miljöer”.
Hydraulikdagar 2003, Linköping, Sweden.
sessment, and Implementation”. Final Report, Website of Kennedy Space Center,
NASA, USA.
156 Paper III
Volvo_Predicting_Digging.pdf

Paper IV
A Methodology for Modeling
the Influence of
Construction Machinery Operators
on Productivity and Fuel Consumption
Reno Filla
VOLVO CONSTRUCTION EQUIPMENT AB, ESKILSTUNA, SWEDEN
IV
Abstract
This paper is concerned with modeling the actions of a human operator of construction
machinery and integrating this operator model into a large, complex simulation model
of the complete machine and its environment. Because human operators to a large de-
gree affect how the machine is run, adaptive operator models are a necessity when the
simulation goal is quantification and optimization of productivity and energy efficiency.
Interview studies and test series have been performed to determine how professionals
operate wheel loaders. Two models using different approaches were realized and inte-
grated into a multi-domain model for dynamic simulation. The results are satisfactory
and the methodology is easily usable for other, similar situations.
Keywords: dynamic simulation, operator model, driver model

A déjà vu is usually a glitch in the Matrix.
It happens when they change something.
(Trinity in the film “The Matrix”)

Filla, R. (2009) “A Methodology for Modeling the Influence of Construction Machinery
Operators on Productivity and Fuel Consumption”. Proceedings of HCII 2009: Digital
Human Modeling, LNCS 5620, pp 614-623.
http://dx.doi.org/10.1007/978-3-642-02809-0_65
A Methodology… 159
1 Introduction
In this ongoing research on simulation in the conceptual design of complex working
machines, a wheel loader was chosen as the object of study, although others can be
found not only in the field of construction machinery, but also in other sectors such as
agriculture, forestry, and mining. Common factors are that these machines consist of at
least two working systems that are used simultaneously and that the human operator is
essential to the performance of the total system.
In the case of a wheel loader, drive train and hydraulics are both equally powerful
and compete for the limited engine torque. Figure 1 visualizes how the primary power
from the diesel engine is split up between hydraulics and drive train (outer loop) in or-
der to create lift/tilt movements of the bucket and traction of the wheels, but is con-
nected again when filling the bucket in e.g. a gravel pile. In this situation, the traction
force from the drive train, acting between wheels and ground, creates a reaction force
between gravel pile and bucket edge, which in turn counteracts lift and tilt forces from
hydraulics, and vice versa [1].
Lifting +
Breaking/Tilting
ECU
Auxiliaries
Gravel
pile
External (+/-)
ECU
Travelling/
Drive train Wheels
Penetration
Figure 1. Simplified power transfer and control scheme of a wheel loader

during bucket loading
The inner loop in Figure 1 shows how the human operator interacts with the wheel
loader. In order to fill the bucket, the operator needs to control three motions simultane-
ously: a forward motion that also exerts a force (traction), an upward motion (lift) and a
rotating motion of the bucket to fit in as much material as possible (tilt). This is similar
to how a simple manual shovel is used. However, in contrast to a manual shovel, the
operator of a wheel loader can only observe, and cannot directly control these three
motions. Instead, he or she has to use different subsystems of the machine in order to
accomplish the task. The gas pedal controls engine speed, while lift and tilt lever control
valves in the hydraulics system that ultimately control movement of the linkage’s lift
and tilt cylinder, respectively.
The difficulty lies in that no operator control directly affects only one single motion.
The gas pedal controls engine speed, which affects both the machine’s longitudinal mo-
160 Paper IV
tion and via the hydraulic pumps the speeds of the lift and tilt cylinders. The linkage
and due to its design a lift movement will also change the buckets tilt angle and a tilt
movement affects the bucket edge’s height above the ground. In summary, there are
many of interdependencies and it thus takes a certain amount of training to be able to
use the machine efficiently.
In modern wheel loaders the operator does not control major components and sub-
systems directly, but via electronic control units (ECUs). This makes it possible to give
the operator support, e.g. by controlling the cylinder speeds in such a manner that the
non-linearity of the linkage is compensated for and thus the speed of bucket lift and tilt
is proportional to the angle of the tilt and lift lever. Certain aspects of machine opera-
tion, for instance a typical brake-and-reverse driving sequence, can also be developed to
be semi-automatic or fully automated.
If a simulation is required to capture the full scope of the interaction between the ma-
chine, its environment, and its operator, all three must be modeled at an appropriate
level of detail in order to give valid results as regards such complex total system proper-
ties as productivity and fuel consumption / energy efficiency. This is an aspect that is
traditionally neglected, because the modeling needs to be extended beyond the technical
system.
2 Literature study
2.1 Working cycle
The knowledge regarding wheel loader operation and working cycles presented in this
paper has been derived from the author’s own wheel loader experience and to a large
extent from discussions with colleagues, test engineers, product specialists, and profes-
sional operators at Volvo. Most of these discussions were not formally structured, but
rather conducted in an ad-hoc manner. However, some unstructured research interviews
were carried out and one interview, conducted with a professional test operator, was
recorded in the form of a semi-structured research interview. Furthermore, many meas-
urements were performed and the results and implications discussed. Most of these re-
ports are internal, but some MSc theses [2], [4], [5] and academic papers [6] are avail-
able in the public domain.
In [7], Gellerstedt published wheel loader operator’s thoughts and reasoning and also
documented some typical working cycles with photos and test data.
Furthermore, non-academic publications like operating manuals and instruction ma-
terial available from machine manufacturers contain useful information.
2.2 Operator models
Surprisingly little could be found in a literature review restricted to working machines

and designing operator models for the purpose of simulation. Zhang et al. validate con-
trol strategies by conducting human-operated experiments in their Earthmoving Vehicle
Powertrain Simulator. The corresponding paper [3] gives some insight into their reason-
ing regarding human-machine interaction. They acknowledge the difference between
task-oriented jobs (such as wheel loader operation in a short loading cycle) and refer-
ence-oriented jobs (e.g. driving a car).
Some more work can be found in the field of autonomous excavation. Hemami spe-
cifically examines bucket filling in [8] and also Wu [9], starting off with analyzing
wheel loader operation in general, later focuses on the bucket filling phase. Both handle
the problem as one that can be solved by following predefined trajectories.
An abundance of literature in the aerospace and in the automotive sector deals with
pilot models and parameter identification for the purpose of predicting pilot behavior
over the next seconds [10]-[14]. Among the techniques employed are path-following
controllers, Kalman filters, fuzzy logic, and neural networks. Such models are used for
advanced control problems like pilot or driver assistance systems and energy manage-
ment of hybrids. But the inverse problem is also considered: assessing handling quali-
ties in certain predefined maneuvers.
3 Development methodology
3.1 Required model features
Since the application of the operator model is in simulation in conceptual design, i.e.
before any physical prototype is available, it has been deemed important that the model
not be hard-coded in any way. Using fixed time, speed or position references and prede-
fined trajectories, there is a significant risk that these references are only valid for the
machine that was used during development of the operator model, but result in large
deviations for a new machine with as yet unknown properties. Therefore, any such ref-
erences must be weak ones and either constant for all machines of any size and architec-
ture, or possible to formulate parametrically, i.e. as a function of bucket length, loading
capacity, wheel distance or similar.
Also, in order to mimic the human operator as close as possible, the operator model
must also be strictly separated from both machine and environment model, and its in-
puts and outputs must be limited to those of a human operator.
3.2 Orientating interviews
These interviews are unstructured and aim to establish basic knowledge of the type of
cycle to model, its features, and its characteristics.
162 Paper IV
In our work, interviewing owners and operators of wheel loaders at different sites has
lead to the conclusion that each working place is unique in its parameters, but the short
loading cycle (Figure 2) is highly representative of the majority of applications.
Typical for this cycle is bucket loading of material on an adjacent load receiver
within a time frame of 25-35 seconds. The aforementioned problems with interactions
between subsystems are highly present.
Several phases can be identified (Figure 2), which Table 1 describes briefly. A more
detailed exploration of the short loading cycle can be found in [1].
Essentially all operators described their operation commands as triggered by events,
both from the machine and its environment, sometimes guided by weak speed or posi-
tion references. It has therefore been found meaningful to develop rule-based operator
models and all subsequent steps are based on that decision.
3.3 In-depth interviews
In this next step, semi-structured research interviews are conducted with professional
operators who are able to verbally express how they use the machine in the working
task at hand. It is important to go through each cycle phase in detail, noting what event
triggered a reaction, which controls are applied and how long, alternate scenarios, what
defines success, what defines failure etc.
In our work we were able to interview one product specialist who was also a profes-
sional test operator and, more importantly, experienced machine instructor and thus
used to teaching people how to operate wheel loaders in an efficient manner. Addition-
ally, many brief unstructured interviews were performed with colleagues of a similar
background. Table 2 lists one result: a guide to how to fill the bucket.
Table 1. Phases of the short loading cycle
# Phase Description
1 Bucket filling Bucket is filled by simultaneously controlling the
machine speed and lift and tilt functions.
2 Leaving bank Operator drives backwards towards the reversing
point and steers the machine to achieve the charac-
teristic V-pattern.
3 Retardation Is started some time before phase 4 and can be
either prolonged or shortened by controlling the
gas pedal and the service brakes.
4 Reversing Begins when the remaining distance to the load
receiver will be sufficient for the lift hydraulics to
achieve the bucket height necessary for emptying
during the time it takes to get there.
5 Towards The operator steers towards the load receiver, thus
load receiver completing the V-pattern. The machine arrives
perpendicular to the load receiver.
6 Bucket emptying The machine is driven forward slowly, the loading
unit being raised and the bucket tilted forward at
the same time.
7 Leaving Operator drives backwards towards the reversing
load receiver point, while the bucket is lowered to a position
suitable for driving.
8 Retardation Not necessarily executed at the same location as in
and reversing phases 3 and 4, because lowering an empty bucket
is faster than raising a full one.
9 Towards bank The machine is driven forward to the location
where the next bucket filling is to be performed,
the bucket being lowered and aligned with the
ground at the same time.
10 Retardation Often combined with the next bucket filling by
at bank using the machine’s momentum to drive the
bucket into the gravel pile.
Table 2. Bucket filling
# Description
1 Accelerate to a certain velocity
2 Shift to lowest gear (kick-down) when bucket edge meets gravel pile
3 Apply gas pedal to increase traction and push bucket into the pile
4 In case of slippage between tires and ground, lift bucket a little more to
increase load on front axle (increases traction)
5 If bucket gets stuck in pile, tilt bucket backwards a little
6 If bucket gets stuck in pile, reduce traction a little
7 Follow the slope of the pile in a carving manner
8 Lift continuously, apply tilt function without releasing lift lever
9 Leave pile with bucket below tipping height (straight lifting arms)
10 Tilt bucket fully back
164 Paper IV
As another example, Table 3 shows how phases 2 to 4 are performed, i.e. leaving
bank, retardation, and reversing.
Table 3. Leaving bank, retardation, and reversing
# Description
1 Put transmission into reverse gear
2 Start lifting
3 Apply gas pedal to accelerate, but keep machine speed below shifting
point for gear 3
4 Steer machine into right curve (if load receiver is standing to the left)
5 Adjust steering so that machine can arrive perpendicular to the load
receiver
6 Choose reversing point so that distance to load receiver is enough for
bucket to reach sufficient height for emptying, if continuously lifted
7 Retard economically by releasing gas pedal
8 Steer back machine into straight position
9 Apply brakes until machine almost stands still
10 Put transmission into forward gear
3.4 Recording and analyzing test series
Together with video recordings of the machine in operation and possibly also from in-
side the cab, measurement data can serve as an additional source of information.
In our work we equipped test machines with a data acquisitioning system and re-
corded control inputs and major machine variables, such as engine speed, traveling dis-
tance, bucket height and angle (etc). We also found video recordings of the machine in
operation to be a valuable complement.
3.5 Deriving general rules and constraints
The results from interview studies and possibly additional measurements are now to be
transformed into general, non-machine specific rules and constraints. This is the last
step before coding the operator model, which makes it necessary to construct the rules
and constraints in a way that is possible to implement.
In our work one result was a quantification of values for “a little” and “certain”, the
vague vocabulary used in Table 2. For instance, the initial velocity in rule #1 was set to
3 km/h and the trigger for kick-down in rule #2 was set to a bucket penetration depth of
200 mm. The height in rule #9 was set to 1/3 of maximum lifting height. Also, the
working place was set up parametrically, more details can be found in [16].
3.6 Implementation
The derived general rules and constraints are now to be implemented as execution paths
of a finite state machine.
In our work two operator models have been realized. In both cases the model of the
technical system and the working place were developed and simulated in ADAMS, a
three-dimensional multi-body system code. Both operator models are separate entities
and the information exchange with machine and environment is limited to operator in-
puts and outputs similar to a human being, i.e. the operator has neither insight nor influ-
ence at a deeper level.
The first operator model has focused on the bucket filling and has been realized as
state equations in ADAMS with an algorithm reminiscent of fuzzy logic. This is ex-
plained in more detail in [15].
The second operator model, with a focus on the remaining phases of the short load-
ing cycle, has been realized in Stateflow (Figure 3) and co-simulated with ADAMS
[16]. Table 4 shows how cycle phase 2 has been realized with two parallel paths.
Figure 3. Top level view of the operator model in Stateflow [16]
Table 4. Applied steps in phase 2 of the operator model
# Description
A1 Apply full lift and full gas pedal.
A2 Steer machine so that reversing will take place with machine pointing at
workplace origin (angle at ca 45°). Calculate required steering angle.
A3 Apply full steering, reduce and stop when required steering angle is reached.
A4 Begin steering back when machine is about to point at workplace origin.
A5 Keep driving backwards until path B terminates this phase.
B1 Wait for machine to pass load receiver (fist geometric possibility to reverse).
B2 Wait for extrapolated remaining lifting time to be lower than remaining ex-
trapolated driving time to load receiver.
B3 Calculate required steering angle for perpendicular arrival at load receiver.
Terminate this phase if within achievable limits.
B4 Go to phase 2a (extension, continued straight driving until steering possible).
166 Paper IV
3.7 Validation
With the operator model implemented, it can now be validated by running simulations
with varying parameter setups and comparing to results from test series.
In our work, both operator models have been tested by varying the machine’s techni-
cal parameters. For instance, changing the machine’s torque converter to a weaker char-
acteristics (as we also have done in our measurement series), leads to a slightly different
operating style with in general higher engine speeds, since the operator adapts and
quickly finds the necessary amount of gas pedal angle to control traction force. Exactly
this phenomenon has also been shown to occur in our simulations, without any explicit
coding of it (see Figure 4, results from first operator model).
Figure 4. Engine load duty for machines with different torque converters
(first model) [15]
Figure 5. Adaptation to lifting speed (second model) [16]
Another example is that of changed speed in the bucket’s lift hydraulics, emulating
insufficient pump capacity. A human operator adapts to this by reversing farther with
the wheel loader until finally driving forward to the load receiver to dump the bucket’s
load. This could also be demonstrated in our simulations without any modification of
the operator model. Figure 5 shows traces from experiments with the second operator
model (black curve with circular markings: lower lifting speed).
4 Conclusion and outlook

With both presented operator models, a “human element” has been introduced to dy-
namic multi-domain simulation of complete construction machinery. Both operator
models have been derived from interview studies in a fairly straightforward manner.
They prescribe the machine’s working cycle in a more generic way, independently of
the machine’s technical parameters. Due to this, whole components or sub-systems can
be changed in their characteristics without compromising the validity of the simulation.
This gives more relevant answers with respect to total machine performance, productiv-
ity, and fuel consumption. The results are satisfactory and the methodology is easily
usable for other, similar situations.
In the future, we will try to develop operator models with which a machine’s oper-
ability can be predicted using simulation. As noted earlier, in automotive and aerospace
simulations this is usually achieved by analyzing the control effort required to perform
predefined maneuvers. A weighed, piece-wise analysis of prominent cycle phases (e.g.
bucket filling, reversing and bucket emptying) might be a way forward.
In order to validate any such simulations, we will also work with quantification of
work load by simultaneously performing in-depth measurements on a wheel loader in
operation and physiological measurements on the operator controlling the machine.
References
[1] Filla, R. (2005) “Operator and Machine Models for Dynamic Simulation of Con-
[2] Vännström, J. and Lindholm, S. (2007) “Educational Interface for Reducing Fuel
Consumption”. Master thesis, Luleå University of Technology, Luleå, Sweden.
http://epubl.luth.se/1402-1617/2007/114/
[4] Stener, P. and Snabb, R. (2008) ”Körbarhetskvantifiering av hjullastare”. Master
thesis, Linköping University, Linköping, Sweden.
168 Paper IV
[5] Boman, M. (2006) “On predicting fuel consumption and productivity of wheel
loaders”. Master thesis, Luleå University of Technology, Luleå, Sweden.
http://epubl.ltu.se/1402-1617/2006/009/
[6] Filla, R. (2003) “Anläggningsmaskiner: Hydrauliksystem i multidomäna miljöer”.
Hydraulikdagar 2003, Linköping, Sweden.
[7] Gellersted, S. (2002) ”Manövrering av hjullastare (Operation of Wheel Loaders)”.
Technical report, JTI – Institutet för jordbruks- och miljöteknik, Uppsala, Sweden.
Oct 1994, pp 1333-1338.
http://dx.doi.org/10.1109/28.315248
[10] Macadam, C.C. (2003) “Understanding and Modeling the Human Driver”. Vehicle
System Dynamics, vol. 40, pp 101-134.
[11] Summala, H. (2000) “Automatization, automation, and modeling of driver’s be-
havior”. Recherche – Transports – Sécurité, vol. 66, pp 35-46.
http://dx.doi.org/10.1016/S0761-8980(00)80004-7
[12] Plöchl, M. and Edelmann, J. (2007) “Driver models in automobile dynamics ap-
plication”. Vehicle System Dynamics, vol. 45, nos. 7-8, pp 699-741.
http://dx.doi.org/10.1080/00423110701432482
[13] Bengtsson, J. (2001) “Adaptive Cruise Control and Driver Modeling”. Licentiate
Thesis, Lunds Institute of Technology, Lund, Sweden.
http://www.cs.clemson.edu/~johnmc/courses/cpsc875/resources/acc/6.pdf
[14] Anderson, M. R., Clark and C., Dungan, G. (1995) “Flight test maneuver design
using a skill- and rule-based pilot model”. Conference paper, IEEE International
Conference on Intelligent Systems for the 21st Century, vol. 3, pp 2682-2687.
[16] Filla, R. (2005) “An Event-driven Operator Model for Dynamic Simulation of

Paper V
Hybrid Power Systems
for Construction Machinery:
Aspects of System Design and Operability
of Wheel Loaders
Reno Filla
Abstract V
This paper will examine the wheel loader as a system with two parallel energy conver-
sion systems that show a complex interaction with each other and with the power
source. Using a systematic design approach, several principle design solutions for hy-
bridization can be found.
Furthermore, the human operator with his/her control actions needs to be considered
as part of the total system. This paper will therefore also connect to results from ongo-
ing and previous research into operator workload and operability.
Keywords: hybrid systems, operator workload, adaptive automation

It's not what you are underneath but what you do, that defines you.
(from the film “Batman Begins”)

Filla, R. (2009) “Hybrid Power Systems for Construction Machinery: Aspects of
IMECE 2009, Vol. 13, pp 611-620.
Hybrid Power Systems… 173
1 Introduction
Working machines in construction, mining, agriculture, and forestry are not only be-
coming ever more sophisticated in terms of digital control, but are also complex in ar-
chitecture even in their conventional form. Many of these machines consist of at least
two working systems that are used simultaneously and the human operator is essential
to the performance of the machine in its working place.
This paper will focus on wheel loaders, giving first an overview of the technical sys-
tem and then describing possibilities for hybridization with some examples of how as-
pects of operator workload are linked to system design. Methods to assess operator
workload will also be reviewed.
2 Conventional Wheel Loaders

Wheel loaders are good examples of complex working machines. Drive train and hy-
draulics are both equally powerful and compete for the limited engine torque. Figure 1
shows how in the case of bucket loading the primary power from the diesel engine is
essentially split up between hydraulics and drive train to create lift/tilt movements of
the bucket and deliver traction to the wheels.
Lifting +
Breaking/Tilting
Auxiliaries
Gravel
Σ Engine
pile
External (+/-)
Travelling/
Drive train Wheels
Penetration
Figure 1. Simplified power transfer scheme of a conventional wheel loader during

bucket loading
In addition to auxiliary systems such as cooling systems, there may also be external
systems connected via PTO’s (power take-outs).
3 Working Cycles
Wheel loaders are versatile machines and each working place is unique, yet common
features can nonetheless be found.
The short loading cycle shown in Figure 2, sometimes also dubbed V-cycle or Y-
cycle for its characteristic driving pattern, is highly representative of the majority of
applications. Typical for this cycle is bucket loading of granular material (e.g. gravel)
174 Paper V
on an adjacent load receiver (e.g. a dump truck) within a time frame of 25-35 seconds,
depending on working place setup and how aggressively the operator uses the machine.
The interaction between hydraulics and drive train is one reason for choosing the short
loading cycle as a kind of standard test cycle when productivity, fuel consumption and
operability are to be assessed. A detailed description with identification of all phases
can be found in [1] and [2].
Load & carry cycles, sometimes also called long loading cycles, are commonly used,
too. They resemble short cycles, but involve two longer transport phases of up to 400 m
in forward gear (Figure 3).
Figure 3. Load & carry cycle

4 Hybrid Wheel Loaders

As mentioned before, wheel loaders are versatile machines used in a variety of applica-
tions, which makes it hard to find a globally optimal setup. This problem is even more
accentuated when it comes to hybridization of these working machines. Customers will
expect great gains in terms of energy efficiency when using the hybrid wheel loaders for
the same variety of applications they used their conventional machines for.
In order to systematically examine hybridization opportunities, it is meaningful to
first consider the parallel working systems hydraulics and the drive train separately.
n1
ICE n2
ICE
Figure 4. Conventional drive train and hydraulics
The left-hand diagram in Figure 4 shows the conventional drive train, where the in-
ternal combustion engine drives a transmission via a torque converter (here simplified
as a clutch) and further on rotates the machine’s four equally large wheels, passing
through differentials along the way. The hydraulics are driven by one or several parallel
hydraulic pumps, directly connected to the engine (Figure 4, right).
In this paper the focus will be on electric hybrids, but also hydraulic hybrids are un-
der development in the industry and academia [3][4][5].
M/G n1
ICE 3~ n2
ICE M/G
3~
M/G n1
ICE 3~ n2
Figure 5. Parallel hybrid drive train and hydraulics

176 Paper V
In a parallel hybrid system, another power source is mechanically connected to the

same drive shaft, enabling torque to be added or subtracted. As shown to the left in
Figure 5, in the case of the parallel hybrid-electric drive train the question arises of
whether the electric machine should be placed upstream or downstream of the torque
converter. Upstream placement permits cranking and torque support of the engine,
while a downstream position offers higher recuperation efficiency. A compromise can
be found by using a lock-up torque converter and an additional clutch. However the
design of a parallel hybrid hydraulic system is fairly straight forward (Figure 5, right).
In a series hybrid configuration (Figure 6) there is no purely mechanical connection
between engine, drive train, and hydraulics. The power flows through electric machines
where it is first converted into the electric domain and then back to the mechanical do-
main.
M/G M/G n1 M/G

ICE 3~ 3~ n2 3~
ICE M/G
3~
Figure 6. Series hybrid drive train and hydraulics
A complex or power-split hybrid system can be seen as a combination of parallel and

series hybrid. There is both a mechanical connection between the engine and the system
to be driven and an electrical connection via two electric machines (Figure 7).
M/G
3~
M/G n1 M/G
ICE 3~ n2 Σ 3~
ICE M/G
3~
Figure 7. Complex hybrid drive train and hydraulics
The amount of power to be transferred either way can be chosen according to the
situation at hand. This offers the possibility to combine the advantages and to avoid the
disadvantages of the topologies described above.
An abundance of literature deals with various types of complex hybrid drive trains
and power-split, continuously variable (CVT) or infinitely variable (IVT) transmissions.
It is also possible to let the road act as a summation device, as can be seen in Figure 8.
M/G n1 M/G n1
ICE 3~ n2 3~ n2
Figure 8. Road as summation device
Power-split hydraulics can in principle be seen as conventional hydraulics with elec-

trically driven boost pumps, but might also be designed so that both pumps share the
generation of hydraulic power in a larger time frame.
conventional Hydraulics
conventional
n1
ICE n2
parallel hybrid complex (power-split) series hybrid
M/G
3~
M/G
parallel hybrid
3~
M/G n1 M/G n1 M/G n1

ICE 3~ n2 ICE 3~ n2 ICE 3~ n2
M/G n1
ICE 3~ n2
complex (power-split)
M/G
3~
M/G
3~
M/G n1 M/G M/G n1 M/G

ICE 3~ n2 Σ 3~ ICE 3~ n2 Σ 3~
M/G n1 M/G
ICE 3~ n2 Σ 3~
Drive train
M/G
3~
M/G
3~
series hybrid
M/G M/G n1 M/G M/G n1

ICE 3~ 3~ n2 ICE 3~ 3~ n2
M/G M/G n1
ICE 3~ 3~ n2
Figure 9. Combination matrix of hybrid topologies
Combining all the variants, we find a matrix of system possibilities (Figure 9), from
conventional systems over parallel and complex hybrids all the way to series hybrid
systems. The topology of the hydraulic system changes with the column, while the drive
train topology varies between the rows. Some combinations are not drawn because they
178 Paper V
are not logically possible: if e.g. a series hybrid drive train is installed, then naturally
any installed conventional hydraulics are automatically a parallel hybrid because the
electric machine connected to the engine is at the same time also connected to the hy-
draulic pumps. The same reasoning can be applied to all five excluded combinations.
The system sketches shown are to be understood as simple examples. Neither energy
storage nor power electronic converters are included. Also, several principle solutions
can be found for each combination, each with distinctive sub-variants.
For instance, the full series hybrid design shown in the lower right corner of Figure 9
may be the simplest possible solution in terms of design effort, but not necessarily so in
terms of system complexity. It is a relatively naïve design to drive only the hydraulic
pump and the transmission with an electric motor and keep the rest of the working sys-
tem unchanged. Using electric machines usually opens up for radical new possibilities
in system design, as well as system size (downsizing) and system control.
For hydraulics, one natural path of evolution might be towards pump-controlled
“valveless” systems [6][7], possibly with a separate hydraulic circuit for each individual
function (lift, tilt and steering). It is then a question of whether the flow adaptation
should still be made by means of pumps with variable displacement (as shown in all
figures) or if this should be accomplished varying only the motor speed and using fixed
displacement pumps.
M/G
3~
M/G
3~
M/G M/G n1
ICE 3~ 3~ n2
M/G M/G n1
ICE 3~ 3~ n2
Figure 10. Designs of full series hybrid wheel loaders
In the case of the drive train, electric machines could be integrated directly into the
wheel hubs. Both solutions are shown on the right in Figure 10, in comparison to the
naïve design shown on the left.
In general, all electro-hybrid system solutions give rise to new system properties be-
cause electric machines offer a different torque envelope and a faster response in a
wider speed range compared to combustion engines. Their introduction should therefore
also include an adaptation and redesign of the system to be driven, rather than just a
substitution of the propulsion source. For a hybrid wheel loader, this gives new possi-
bilities to achieve an optimal compromise between productivity, energy efficiency, and
operability.
5 Operability
While there are usable definitions for productivity (for a wheel loader: material loaded
per time) and energy efficiency (preferably: fuel consumed per loaded unit, rather than
per time unit), a generally agreed definition of the human operator’s difficulty in work-
ing with the machine is still to be found. For vehicles, the concept of drivability has
been the main focus for quite some time. Recently, however, the operability of working
machines has also become a major research subject (the word operability acknowledg-
ing the fact that while the drive train is the only major power-consuming system in a
vehicle, working machines like wheel loaders have at least two working systems that
are used simultaneously).
In [8], a definition is offered that also works well for working machines: “Operability
is the ease with which a system operator can perform the assigned mission with a sys-
tem when that system is functioning as designed”. The limitation to states where the
system is functioning as designed effectively excludes somewhat related yet separate
properties such as robustness and reliability.
The challenge in designing a wheel loader, from a manufacturer’s point of view, is to
find an appropriate, robust, and maintainable balance between productivity, fuel effi-
ciency, and operability over the complete area of use. At Volvo, the term Harmonic
wheel loader has been coined, describing a machine possessing a high degree of ma-
chine harmony which makes it intuitively controllable and able to perform the work
task in a straightforward manner without much conscious thought or strategy. The lower
the degree of machine harmony, the more effort the operator has to put in in order to
perform the work task (Figure 11) and thus the higher the workload.
Operator
effort
y
Secondary harmony
on
rm
(added control systems)

ha
e
hin
ac
M
Primary harmony
(matched and appropriately specified hardware)
Figure 11. Machine harmony and operator effort
In general, the assumption made in this research is that the operator’s impression of
the working machine’s operability stems from the amount of workload the operator is
subjected to. We exclude physical workload and therefore also operator comfort with
aspects like exposure to vibration, ergonomics etc, and concentrate on the mental work-
180 Paper V
load of the operator, which is affected by the operator’s control efforts and thus by the
following requirements (the fourth item probably being the most prominent):
• fast response
• high precision of machine and bucket positioning
• several operator controls to be actuated simultaneously in order to achieve
a specific effect
• several machine functions to be synchronized.
In a short loading cycle as shown in Figure 2, as well as in a load & carry cycle ac-
cording to Figure 3, there are three phases in which the operator is challenged to a
higher degree:
• Bucket filling (phase #1)
• Reversing (phase #4)
• Bucket emptying (phase #6 in Figure 2, #9 in Figure 3)
Each of these phases will be examined in more detail with regard to operability and
operator workload in the following sections.
6 Operability Aspects in the Bucket Filling Phase

Extending the schematics of the technical system (Figure 1), Figure 12 depicts how the
human operator interacts with the wheel loader during bucket filling.
Lifting +
Breaking/Tilting
ECU
Auxiliaries
Gravel
pile
External (+/-)
ECU
Travelling/
Drive train Wheels
Penetration
Figure 12. Simplified power transfer and control scheme of a wheel loader during
bucket loading
In order to fill the bucket, the operator needs to control three motions simultane-
ously: a forward motion that also exerts a force (traction), an upward motion (lift) and a
rotating motion to fit in as much material as possible (tilt). This is similar to how a sim-
ple manual shovel is used. However, in contrast to a manual shovel, the operator of a
wheel loader can only observe, and cannot directly control these three motions. Instead,
he or she has to use different subsystems of the machine in order to accomplish the task.
The gas pedal controls engine speed, while it is lift and tilt lever control valves in the
hydraulics system that ultimately control movements of the linkage’s lift and tilt cylin-
ders, respectively.
One difficulty is that no operator control directly affects only one single motion. The
gas pedal controls engine speed, which together with the selected transmission gear
affects the machine’s longitudinal motion. But engine speed also relates to the speed of
the hydraulic pumps, which in turn affects the lift and tilt cylinder speed. The linkage
and due to its design a lift movement will also change the bucket’s tilt angle and a tilt
movement affects the bucket edge’s height above the ground. Furthermore, when simul-
taneously using lift and tilt function, a reduction in the lift cylinder speed can be ex-
pected as the displacement of the hydraulic pump is limited and the tilt function is pri-
oritized over lift due to its lower pressure demand. The more the tilt lever is actuated,
the higher the tilt cylinder speed, while the speed of the lift cylinder decreases.
Finally, there is also a strong interaction between traction and lift/tilt forces: penetrat-
ing the gravel pile with the bucket requires tractive force, which is transferred through
the wheels to the ground (Figure 13). In accordance with Newton’s Third Law of Mo-
tion, the Law of Reciprocal Actions, the gravel pile exerts an equal and opposite force
upon the loader bucket. A typical sequence for actually filling the bucket is then to
break material by tilting backwards a little, lifting, and penetrating even further.
Figure 13 shows how these two efforts work against each other: in order to achieve a
lifting force, the cylinders have to create a counter-clockwise moment around the load-
ing unit’s main bearing in the front frame. At the same time, the reaction force from the
traction effort creates a clockwise moment that counteracts the lifting effort.
Figure 13. Force balance during bucket filling (simplified)
The ground and the gravel pile thus connect drive train and hydraulics, forcing them
to interact with each other, and to experience each other’s effort as an external load.
Since both working systems are already mechanically connected via the engine, an in-
teresting situation arises in which engine torque is transferred to the wheels to accom-
182 Paper V
plish traction, but at the same time counteracts that part of the engine torque that has
been transferred through hydraulics to accomplish lifting and tilting – in turn requiring
even more torque to be transferred.
The momentary torque or power distribution to drive train and hydraulics is specific
for the working task at hand and is controlled by the operator, who ultimately balances
the complete system and actively adapts to both the machine, the task at hand, and the
working place.
The traditional way of decreasing the severity of these phenomena has been to design
the wheel loader with a high load margin by utilizing the internal combustion engine at
a high speed (often near governed speed), backed up by a steep torque rise, and to em-
ploy weak torque converters that require a high rate of slip in order to transfer torque.
This creates a certain decoupling effect in the drive train, but it is unfortunately prohibi-
tively costly in terms of fuel efficiency.
In modern working machines the main idea is to utilize the engine at as low speeds
as possible by employing a stiffer torque converter and hydraulic pumps with larger
displacement. Also, major components and subsystems are controlled by electronic con-
trol units (ECUs), all connected within a network, as shown schematically in Figure 12.
This makes it possible to give the operator support in controlling the wheel loader. In a
hybrid working machine, this opportunity is even greater. Depending on the chosen
topology, several of the difficulties discussed above can be avoided.
While the gravel pile will still connect drive train and hydraulics downstream via the
bucket, in a series hybrid these systems would not be connected upstream via the en-
gine. This enables the operator to control only one motion at a time via only one con-
trol: the gas pedal will only control the machine’s longitudinal motion and the hydraulic
levers can control flow independently, all independently of each other and of the current
engine speed.
The responsiveness of electric machines also makes it possible to quickly ramp
torque up or down such that tractive force and lift/tilt force can be actively controlled.
This can be used in essentially all hybrid variants (Figure 9).
Furthermore, the cylinder speeds can be controlled in such a manner that the non-
linearity of the linkage is compensated for and thus the speed of bucket lift and tilt is
proportional to the angle of the tilt and lift lever.
However, it is important to observe that not all limitations that arise because of con-
ventional wheel loader design are experienced as negative by the operator. During their
work with a pump-controlled hydraulic system with a separate hydraulic circuit each for
the lift and tilt function, the authors of [6] and [7] noticed that professional operators
use the aforementioned prioritization of the tilt function over the lift function to their
advantage.
Instead of operating the lift and tilt lever separately, the operators pull the lift lever to
a certain level, hold it there and then use the tilt lever to control the speed of both tilt
and lift cylinder. By using this procedure during bucket filling, the operator has only to
actively handle two controls (tilt lever and gas pedal) instead of three (because the lift
lever has been locked in a fixed position). This behavior has been confirmed in the au-
thor’s own interview studies [1].
During measurements on a research prototype with the pump-controlled hydraulic
system described above, it was observed that the fuel consumption in short loading cy-
cles was unusually high during the bucket filling phase. It could be shown that the op-
erators tried, but were not able, to use the new system just like to old one. With the
speeds of the lift and tilt cylinders now independent of each other, they could no longer
be controlled by just one lever. No record exists as to whether the operators themselves
commented on this and complained about the new functionality (in the author’s own
studies on a similar machine one did), but it is clear that the prototype’s operability was
reduced and thus the workload of the operator increased. To solve this, a similar behav-
ior was implemented by software, artificially decreasing the flow demand for the lift
function when the tilt lever was activated (Figure 14).
1.0
0.8
0.6
Lift speed
0.4
0.2
1.0
0.8 0
0.6 0
0.2
0.4 0.4
0.2 0.6
Lift 0 0.8
command 1.0 Tilt command
Figure 14. Reduction of lift speed due to prioritization
Learning from the example above, the question arises of how many changes one can
make in the design of the operator interface and the behavior of a hybrid wheel loader
without professional operators experiencing this in a non-positive way – and then, how
long it would take for the operators to adapt.
7 Operability Aspects in the Reversing Phase

In this phase of a short loading cycle the operator changes the direction of the machine
from reverse to forward and aims at the load receiver, while at the same time continuing
to raise the bucket using the lift function. As examined in [9], there is again a conflict
between the working systems:
In order to reverse the machine, the operator lowers the engine speed (otherwise gear
shifting will be jerky and the transmission couplings might wear out prematurely). Less
184 Paper V
torque is available at lower speeds. Engine response is also worse at lower speeds,
mainly due to the inertia of the turbo-charger and smoke limiter settings.
When the operator shifts from reverse to forward the loader is still rolling backwards.
This forces an abrupt change in the rotational direction of the torque converter’s turbine
wheel, greatly increasing the slip, which leads to a sudden increase in torque demand
from the already weakened engine.
Continuing to lift the bucket while reversing requires high oil flow, which is propor-
tional to the pump’s displacement and shaft speed (and thus engine speed). In order to
satisfy the flow requirement at lower speed, pump displacement is usually at maximum
in this phase, which together with a high hydraulic pressure due to a full bucket and the
loading unit’s geometry leads to a large demand for driving torque.
Thus, both drive train and hydraulics apply high load to the engine, which at lower
speed has less torque available and a longer response time. In a conventional wheel
loader it is the operator’s task to control the machine correctly. In order to lower fuel
consumption, operators of conventional wheel loaders with torque converters are usu-
ally advised to release the gas pedal and use the brakes in order to lower machine speed
almost to stand-still before shifting into forward. This greatly reduces torque converter
slip and thus saves fuel.
In a wheel loader with a hybrid drive train, it would actually be better, both operabil-
ity- and energy efficiency-wise, to go back to the traditional non-optimal method of
reversing and just command a change of direction at a certain point in time and then let
the machine automatically initiate regenerative braking by using the electric machine in
generator mode, followed by a smooth acceleration forward. Requiring the operator to
use the brake pedal leads to a higher workload and ideally requires the hybrid working
machine to incorporate brake blending, i.e. use the electric machine for regenerative
braking in combination with the mechanical brakes.
Related to this, the author witnessed another example of how productivity, energy
efficiency (in the form of fuel consumption) and operability (in the form of operator
workload) are intertwined. In a research prototype machine a system had been imple-
mented that automatically lowered engine speed and applied the brakes when the opera-
tor commanded a change of direction the traditional way by just putting the transmis-
sion gear lever from reverse into forward. The initial expectation was to save fuel,
according to a chain of reasoning as shown in Figure 15.
Reduced slip Reduced
uses brakes when
Figure 15. Expected effect of the automatic reversing feature
Comparing fuel consumed per time unit with and without the automatic reversing
feature, it showed that the new feature actually led to higher fuel consumption. The rea-
son for this was the operator using the automatic reversing feature to increase produc-
tivity, which also increases fuel consumption per time unit. Figure 16 shows the actual
course of events.
Reduced slip Increased
uses brakes when
Reduced Increased
operator work load productivity
for lower work load
and drives harder
Figure 16. Actual effect of the automatic reversing feature
Besides showing how productivity, energy efficiency, and operability are connected
in a working machine, the example above also proves how clearly inappropriate it is to
use fuel consumption measured per time unit in such comparative testing. Fuel effi-
ciency expressed in fuel consumed per loaded mass unit of material would have been a
much better choice.
8 Operability Aspects in the Bucket Emptying Phase

Having driven the machine from the gravel pile to the load receiver and having raised
the full bucket to dumping height at the same time, the operator now needs to coordi-
nate four controls in order to successfully empty the bucket‘s contents:
• Tilt lever in order to dump the material
• Lift lever, gas pedal and brake pedal in order to safely guide the bucket
over the edge of the load receiver and to evenly distribute the material
In this phase, the aforementioned difficulty of no operator control affecting just one
motion is again apparent. The connection between desired increase in hydraulic flow
and thus commanded increase in engine speed now leads to a possibly unwanted in-
crease in tractive force, propelling the machine forward at a faster rate. With the wheel
loader already standing in front of the load receiver and only small forward movements
required, this increases the operator’s workload since the machine may risk driving into
the load receiver, which can only be prevented by also applying the brake pedal.
In a series of measurements on a research prototype with a stiffer torque converter
(i.e. one that can create the same torque at a lower slip rate than that of a baseline ma-
chine) the author again witnessed the connection between productivity, energy effi-
ciency, and operability. Originally it was expected to see a decrease in fuel consumption
due to the stiffer torque converter in the research prototype machine theoretically lead-
ing to less slip and thus lower drive train losses (Figure 17).
186 Paper V
Stiffer torque converter, Reduced slip Increased

yet same traction demand over torque converter fuel efficiency
Figure 17. Expected effect of a stiffer torque converter
Comparing fuel efficiency, the operator when using the research prototype machine
with the stiffer torque converter was shown to require more time per loading cycle and
thus consumed more fuel in total and per loaded unit of material (but not per time unit,
as the additional consumption per time unit was close to the average and thus led to
approximately the same mean fuel consumption figure). It was concluded that the main
reason for the longer cycle time was difficulties in bucket filling and bucket emptying
due to the stiffness of the torque converter. Figure 18 shows the course of events.
However, same pump

Stiffer torque converter, Reduced
speed and thus same
yet same traction demand fuel efficiency
engine speed demand
Higher traction force Increased cycle

propels machine faster time and reduced
for same pedal angle productivity
by applying brakes
during bucket emptying
Figure 18. Actual effect of a very stiff torque converter
Especially during bucket emptying the operator found it difficult to handle the higher
tractive force at lower slip rates and thus lower engine speeds if compared for the same
propeller shaft speed. In order to increase hydraulic oil flow, operators apply the gas
pedal to increase pump speed, which due to a conventional wheel loader’s system de-
sign also increases converter speed and thus increases tractive force. To avoid colliding
with the load receiver, the operator had to apply the brake pedal, which resulted in both
increased workload and increased drive train losses (because the torque converter essen-
tially acted as a retarder brake).
In most hybrid wheel loader designs (Figure 9) the above described course of events
can be avoided, since the gas pedal would be used to control the machine’s traveling
speed rather than the internal combustion engine’s rotational speed.
9 Assessing Operability
In summary it can be concluded that operability issues must be factored in when design-
ing working machines, and more so when designing hybrids.
Virtual Prototyping has been generally adopted in product development in order to
minimize the traditional reliance on testing of physical prototypes. The complex archi-
tecture of working machines with tightly coupled, non-linear subsystems in different
engineering domains make simulation of the complete system’s dynamic behavior diffi-
cult. But in early design stages such as conceptual design, simulation is the only viable
option.
In order to capture the full scope of the interaction between the machine (the techni-
cal system), its environment (working place, gravel pile), and its operator, all three must
be modeled at an appropriate level of detail if the simulation is to give valid results.
Usually, the technical systems are modeled at a very intricate level of detail, while its
environment is modeled crudely (for instance by various types of rolling resistance),
and the operator model is virtually non-existent, following a predefined trajectory with
control inputs at given points, and thus capturing none of the essential phenomena.
Advances in simulation-related research, among those the author’s own work
[10][11] and the work of others inspired by it [12][13] as well as the many advances in
research on autonomous bucket loading [14][15][16], have resulted in a growing body
of knowledge on the operator’s behavior and how to approximate his or her control ac-
tions in a simulation. Using for example a weighed, piece-wise analysis of control effort
in the different cycle phases might be one way of quantifying the (simulated) operator’s
workload and measuring the (simulated) machine’s operability. However, some work
still remains to be done in this regard.
Meanwhile, research in the field of Human-Computer Interaction has resulted in the
development of Cognitive Architectures that can be used for modeling cognitive aspects
of human beings while interacting with technical systems [17]. However, for tasks like
operating a machine within a simulated environment the operator model needs to have a
spatial awareness in order to orient itself in space, detect objects to interact with etc.
One promising approach to address this has been reported in [17].
It can also be argued somewhat philosophically that the closer a simulation comes to
reality, the more complex it necessarily has to be until it is finally as complex and diffi-
cult to understand as reality itself, also introducing challenges to its validity in terms of
repeatability. That means that there will be some limit to the extent of how operability
can be simulated using operator models.
The other approach then would then be to still simulate the technical system yet use a
human operator to control it in so-called human-in-the-loop simulations. Examples can
be found in [19][20][21]. The question then is how to quantify a (simulated) machine’s
operability or the operator’s workload leading to an impression of operability.
In the examples given earlier, a physical prototype already exists, yet the same ques-
tion of quantification of operator workload still needs to be answered. Fortunately,
much research has been done on this topic, initially in aeronautics but later also in on-
road vehicles and other applications.
In the beginning, workload was quantified using self-report measures like the modi-
fied Cooper-Harper scale or the NASA Task Load Index. These methods rely on the
operator trying to be objective, i.e. not consciously trying to skew results. However, due
to the fact that such self-report assessments are often performed afterwards, the time
delay itself leads to a certain amount of unreliability. Also if for example a wheel
loader’s operability in the reversing phase is to be rated, the operator may find it hard to
188 Paper V
concentrate on this and not let his/her impression of the machine’s total behavior in the
complete loading cycle influence the rating.
It has therefore been found better to assess workload by measuring psychophysi-
ological parameters, predominantly heart rate variability and finger temperature. A
comprehensive overview of the subject and references for further study can be found in
[22][23][24][25].
In the research published in [26], a real-time flight simulator is used with a human
pilot in the loop and the results from these simulator flights are compared to real flights.
We will conduct similar measurements, yet not focused on the operator’s training effect
but on using the operator’s workload in order to assess the tested machine’s operability.
This does not need to be limited to the development of advanced working machines like
hybrids, but can also play an important role in rating selected features of conventional
machines.
10 Workload-adaptive Machines
An earlier chapter discussed how the introduction of hybrid power systems can lead to
machines that are easier to use due to a different system design.
But there are also a great many opportunities in new human-machine interfaces,
made possible by hybrid systems and extensive computer control. The question was
earlier raised of how many changes one can make in the design of the operator interface
and the behavior of a hybrid wheel loader without professional operators experiencing
this in a non-positive way – and then, how long it would take for the operators to adapt.
This issue can be made less prominent by adapting machine behavior and operator
interface to the individual operator’s needs and preferences. This could be as relatively
easy to do as redefining the resolution and mapping of various operator controls to suit
the individual operator, either online or offline.
Augmented Cognition in order to raise the operator’s situation awareness is another
solution, for example by giving advanced visual, haptic or acoustic feedback [27][28].
Instead of requiring the operator to adapt, one could also move the adaption task to
the technical system and use adaptive automation to give the operator help in certain
situations [29]. Possibilities to do this are especially high in hybrid machines whose
systems have greater degrees of freedom than is the case for conventional machines.
11 Conclusions
Working machines like wheel loaders are complex in architecture and behavior, espe-
cially since the operator plays an important role in controlling the total system. This
paper has shown several examples where the operator’s influence has led to unexpected
results. It has been discussed that in future measurements, assessment of operator work-
load needs to be included in order to fully understand the course of events. Methods for
assessing workload in the context of this research have also been discussed.
Hybrid systems give the opportunity to not only vastly improve energy efficiency
and possibly productivity of the working machine, but also increase operability. Trying
to optimize a machine for such different properties as power, productivity, fuel effi-
ciency, operability, initial purchase cost, and total cost of ownership is difficult due to
conflicting demands. Examples have been shown on how hybrid systems can help to
find a better compromise.
Acknowledgments
The financial support of Volvo Construction Equipment and Energimyndigheten, the

Swedish Energy Agency, is hereby gratefully acknowledged.
References
[1] Filla, R. (2009) “A Methodology for Modeling the Influence of Construction Ma-
chinery Operators on Productivity and Fuel Consumption”. Proceedings of
HCII 2009: Digital Human Modeling, LNCS 5620, pp 614-623.
http://dx.doi.org/10.1007/978-3-642-02809-0_65
[3] Achten, P. (2008) “A serial hydraulic hybrid drive train for off-road vehicles”.
IFPE 2008 Technical Conference, Las Vegas (NV), USA, pp 515-521, March 12-
14, 2008.
http://www.deltamotion.com/peter/Hydraulics/Innas/HYDRID/Technical%20Papers%20H
ydrid/HD03%20HyDrid%20for%20off%20road%20vehicles.pdf
[4] Macor, A., Tramontan, M., 2007, “Hydrostatic Hybrid System: System Definition
and Application”, International Journal of Fluid Power, 8 (2), pp 47-62.
http://direct.bl.uk/bld/PlaceOrder.do?UIN=212705688&ETOC=RN
[5] Achten, P., 2007, “The Hydrid Transmission”, SAE 2007-01-4152.
http://dx.doi.org/10.4271/2007-01-4152
[6] Heybroek, K. (2008) “Saving energy in construction machinery using displace-
ment control hydraulics”, Licentiate thesis, Department of Management and Engi-
neering, Linköping University, Linköping, Sweden.
190 Paper V
[7] Lugnberg, M. and Ousbäck, M. (2008) “System Integration of a Pump Controlled

Open Circuit Solution in a Wheel Loader Application”. Master thesis, Department
of Management and Engineering, Linköping University, Sweden.
NASA, USA.
[11] Filla, R. (2005) “An Event-driven Operator Model for Dynamic Simulation of
[12] Elezaby, A.A., Abdelaziz, M., Cetinkunt, S. (2008) “Operator Model for Con-
struction Equipment”. Proceedings of IEEE/ASME 2008 International Conference
on Mechatronic and Embedded Systems and Applications, MESA 2008, pp 582-
585.
[13] Worley, M. D. and La Saponara, V. (2008) “A simplified dynamic model for
front-end loader design”. Proceedings of the Institution of Mechanical Engineers,
Part C: Journal of Mechanical Engineering Science, vol. 222, no. 11, November
2008, pp 2231-2249.
http://dx.doi.org/10.1243/09544062JMES688
[14] Sarata, S., Koyachi, N., Tubouchi, T., Osumi, H., Kurisu, M. and Sugawara, K.
(2006) “Development of Autonomous System for Loading Operation by Wheel
Loader”. Proceedings of the 23rd International Symposium on Automation and
Robotics in Construction, pp 466-471.
http://www.iaarc.org/publications/fulltext/isarc2006-00150_200608211032.pdf
Experiments, Modelling, Identification and Control”. Master Thesis, Queen's Uni-
versity, Kingston (Ontario), Canada.

[17] Gluck, A. and Pew, R., 2005, “Modeling Human Behavior with Integrated Cogni-
tive Architectures: Comparison, Evaluation, and Validation”. Lawrence Erlbaum
Associates, USA.
[18] Robbins, B., Carruth, D. and Morais, A. (2009) “Bridging the Gap between HCI
and DHM: The Modeling of Spatial Awareness within a Cognitive Architecture”.
Proceedings of HCII 2009: Digital Human Modeling, LNCS 5620, pp 295-304.
http://dx.doi.org/10.1007/978-3-642-02809-0_32
[20] Norris, W. R., Zhang, Q. and Sreenivas, R. S. (2006) “Rule-base Reduction for a
Fuzzy Human Operator Performance Model”, Applied Engineering in Agriculture,
vol. 22, no. 4, pp 611-618.
[21] Brown, T., Moeckli, J. and Marshall, D. (2009) “Use of High-Fidelity Simulation
to Evaluate Driver Performance with Vehicle Automation Systems”, Proceedings
of HCII 2009: Engineering Psychology and Cognitive Ergonomics, LNAI 5639,
pp 339-348.
http://dx.doi.org/10.1007/978-3-642-02728-4_36
son, K. (2003) “GARTEUR Handbook of Mental Workload Measurement”, Final
Report for Flight Mechanics Group of Responsibles (FM GoR) of the Group for
Aeronautical Research and Technology in Europe (GARTEUR) by GARTEUR
Action Group FM AG13.
[23] De Waard, D. (1996) “The Measurement of Driver’s Mental Workload”. Doctoral
thesis, University of Groningen, The Netherlands.
military Scenarios. Criteria for Sensor Technologies”. Proceedings of HCII 2009:
Foundations of Augmented Cognition - Neuroergonomics and Operational Neuro-
science, LNCS 5638, pp 304-310.
http://dx.doi.org/10.1007/978-3-642-02812-0_36
192 Paper V

http://dx.doi.org/10.1007/978-3-642-02574-7_16
Sweden.
[27] Marzani, S., Tesauri, F., Minin, L., Montanari, R. and Calefato, C. (2009) “De-
signing a Control and Visualization System for Off-Highway Machinery Accord-
ing to the Adaptive Automation Paradigm”, Proceedings of HCII 2009: Aug-
mented Cognition, LNAI 5638, pp 42-50.
http://dx.doi.org/10.1007/978-3-642-02812-0_6
[28] Minin, L., Marzani, S., Tesauri, F., Montanari, R. and Calefato, C. (2009), “Con-
text-Dependent Force-Feedback Steering Wheel to Enhance Drivers’ On-Road
Performances”, Proceedings of HCII 2009: Augmented Cognition, LNAI 5638, pp
51-57.
http://dx.doi.org/10.1007/978-3-642-02812-0_7
[29] De Greef, T., Lafeber, H., van Oostendorp, H. and Lindenberg, J. (2009) “Eye
Movement as Indicators of Mental Workload to Trigger Adaptive Automation”,
Proceedings of HCII 2009: Augmented Cognition, LNAI 5638, pp 219-228.
http://dx.doi.org/10.1007/978-3-642-02812-0_26

Paper VI
Study of a method for
assessing operability of working machines
in physical and virtual testing
Reno Filla
Abstract
In this study of eighteen wheel loader operators, test-driving a machine in three dif-
ferent traction force settings, we found strong support for the hypothesis that the opera-
VI
tor’s control commands can be used to assess the machine’s operability, at least in form
of ease of bucket filling.
The methods chosen to derive the control effort worked well and were computation-
ally efficient.
Keywords: operator; control effort; mental workload; operability; human-machine in-

teraction; operator model; simulation
Whatever they say about reality, it's the only place where you can get a good steak.
(Woody Allen)
This paper has been submitted for publication as:

Filla, R. (2011) “Study of a Method for Assessing Operability of Working Machines in
Physical and Virtual Testing”. Submitted in July 2011 for publication in International
Journal of Vehicle Systems Modelling and Testing.
http://www.inderscience.com/browse/index.php?journalID=169
Study of a method for assessing operability … 195
1 Introduction
The trend of replacing traditional testing of physical prototypes with virtual testing of
simulated vehicles and machines is continuing. Thanks to increased power and capabil-
ity of both computer hardware and software, simulation models are becoming larger and
more detailed, and integrated simulation of subsystems of different technical domains is
now common practice.
Working machines in construction, mining, agriculture, and forestry are complex in
architecture with various subsystems that are used simultaneously. Optimisation with
traditional development methods is difficult; the trend towards increased virtual product
development is especially apparent. With the help of co-simulation, complete working
machines have been simulated for about ten years [1].
However, problems arise when a simulation needs to account for the actions of a
human driver or operator. For example, the operator of a working machine is essential
to the performance of the machine in its working place and can influence productivity
and fuel efficiency to a large degree, positively or negatively, depending on the way of
using the machine. It is therefore essential to include the operator’s behaviour in the
simulation.
Since such an operator model would be used in simulation in conceptual design, i.e.
before any physical prototype is available, it is important that the model not be hard-
coded in any way. Using traditional methods like fixed time, speed or position refer-
ences and predefined trajectories, there is a significant risk that these references will
only be valid for the machine that was used during development of the operator model,
but result in large deviations for a new machine with as yet unknown properties (for
example a hybrid machine with a new drive train that gives the machine a different
characteristic behaviour). Therefore, any such references in an operator model must be
weak ones and either constant for all machines of any size and architecture, or possible
to formulate parametrically, i.e. as a function of bucket length, loading capacity, wheel
distance or similar.
Using a wheel loader as an example, such models were reported in [2], using one ap-
proach similar to fuzzy-logic and another one with discrete events. Later, these were
combined and further explained in [3], which presented a methodology for modelling
the influence of construction machinery operators on productivity and fuel consump-
tion. Others, inspired by this work, either copied and again validated the discrete-event
approach [4] and [5], or presented less flexible models using a blend of path-tracking
and velocity-tracking [6]. Still others use a fuzzy-rules based approach [7].
Nonetheless, one aspect had been neglected: the operator’s physical and mental
workload which strongly influences how the machine is used. This aspect is part of the
machine system’s inherent operability, which in [8] is defined as “the ease with which a
system operator can perform the assigned mission with a system when that system is
functioning as designed”. In [9] examples are given of operating situations of a wheel
loader where knowledge of the operator’s workload is of importance when testing of
both physical and virtual prototypes. If this cannot be simulated due to a lack of fully
196 Paper VI
cognitive and information-processing models of human beings, then at least the control
effort in the operator models must somehow be correlated to the workload a human
operator would have had, performing the same actions. Once established, such measure
would not only be useful in virtual testing, but also enable a more objective assessment
of operability in physical testing by offering a complement to test operators’ subjective
evaluation.
It is thus desirable to have some method of quantifying workload other than by ask-
ing the operator for a subjective assessment after test-driving. This article will report on
the setup and results of a larger study conducted to find such method for workload
quantification. The method found is relatively simple and can be applied in both physi-
cal and virtual testing.
2 Wheel loaders and short loading cycles

In this ongoing research, a wheel loader was chosen as the object of study, although
working machines of similar complexity can be found in construction, agriculture, for-
estry and mining. Common factors are that these machines consist of at least two work-
ing systems that are used simultaneously and that the human operator is essential to the
performance of the total system.
Wheel loaders are versatile machines and each working place is unique, yet common
features can nonetheless be found. The short loading cycle shown in Figure 1 is highly
representative of the majority of applications.

Typical for this cycle is bucket loading of granular material (for instance gravel) on
an adjacent dump truck (or other load receiver, mobile or stationary) within a time
frame of 25-35 seconds, depending on working place setup and how aggressively the
operator uses the machine. A detailed description with identification of all phases can
be found in [2].
Visualising test results obtained in earlier studies, Figure 2 shows that the fuel con-
sumption rate (expressed in volume or mass per time unit) is approximately 60% higher
during bucket filling than the cycle average. Expressed in absolute values, bucket filling
accounts for 35-40% of the mean total fuel consumption per cycle, yet the time spend
for filling the bucket is only 25% of the average cycle time.
reversing point and steers the wheel loader to achieve the characteristic V-pattern of a
short loading cycle. The lifting function is engaged the whole time. The operator
without delay. Figure 2 shows that the fuel consumption rate for phases 2 to 6 is ap-
proximately constant and close to the average fuel consumption rate of the complete
short loading cycle.
In the remaining phases, the bucket needs to be lowered and the operator steers the
wheel loader back to the initial position in order fill the bucket again in the next cycle.
Phases 7-10 are less energy-demanding and the fuel consumption rate is therefore lower
than the average for the complete short loading cycle.
198 Paper VI
The higher fuel consumption rate during bucket filling warrants a closer look at this
phase. The inner loop in Figure 3 shows how the human operator interacts with the
wheel loader. In order to fill the bucket, the operator needs to control three motions si-
multaneously: a forward motion that also exerts a force (traction), an upward motion
(lift) and a rotating motion of the bucket to fit in as much material as possible (tilt). This
is similar to how a simple manual shovel is used. However, in contrast to a manual
shovel, the operator of a wheel loader can only observe, and cannot directly control
these three motions. Instead, he or she has to use different subsystems of the machine in
order to accomplish the task. The gas pedal controls engine speed, while lift and tilt
lever control valves in the hydraulics system that ultimately control movement of the
linkage’s lift and tilt cylinder, respectively.
Lifting +
Breaking/Tilting
ECU
Auxiliaries
Gravel
pile
External (+/-)
ECU
Travelling/
Drive train Wheels
Penetration
tion and via the hydraulic pumps the speeds of the lift and tilt cylinders. The linkage
and due to its design a lift movement will also change the buckets tilt angle and a tilt
movement affects the bucket edge’s height above the ground.
Furthermore, Figure 3 also shows how in the outer loop the primary power from the
diesel engine is split up between hydraulics and drive train in order to create lift/tilt
movements of the bucket and traction of the wheels, but is connected again when filling
the bucket in e.g. a gravel pile. Figure 4 shows that in this situation, the traction force
from the drive train, acting between wheels and ground, creates a reaction force be-
tween gravel pile and bucket edge, which in turn counteracts lift and tilt forces from
hydraulics, and vice versa.
Figure 4. Force balance during bucket filling
The normalised diagram to the right visualises that depending on lifting height and
applied traction force, the lifting force can be cancelled out totally, and the bucket can-
not be moved upwards any further. The operator then has to either reduce the traction
force by reducing the engine speed or apply the tilt function.
In summary, there are many interdependencies and it thus takes a certain amount of
training to be able to use a wheel loader efficiently. It is therefore of great interest to be
able to quantify operability as precise and as early in the development cycle as possible.
3 Method
3.1 Working cycle
The aforementioned challenges are one reason for choosing the short loading cycle as a
kind of standard test cycle when productivity, fuel consumption and operability are to
be assessed. In our study it would therefore have been natural to do the same.
However, with the intriguing facts visualised in Figure 2, Figure 3, and Figure 4 in
mind, we specifically wanted to study the bucket filling phase. We therefore chose to
use a modified short loading cycle, where the operator was instructed to still go through
most of the motions, but not to use steering and thus empty the bucket at the same spot
where it was filled. Since the material used was sorted gravel which does not stick to-
gether like e.g. clay and is thus fairly easy to handle, this procedure did not introduce
any skewing of the test results as far as the bucket filling phase is concerned.
The study also included another part, not further presented here, in which most of the
psychophysiological signals that are of interest with regard to the operator’s workload,
i.e. finger temperature, heart rate, respiration rate, CO2 concentration in exhaled air, and
galvanic skin response (skin conductance) were recorded. The attachment of the sensors
required the operator to use the joystick for steering, instead of the steering wheel. In-
experienced operators are unused to this, which in turn means that forcing them to do so
200 Paper VI
would introduce additional mental workload, thus potentially skewing the test results.
The solution was not to use the steering function at all.
3.2 Machine setup

The wheel loader used was a Volvo L70F equipped with a general purpose bucket with
a load capacity of 2.3m3. The total operating weight of the machine is approximately
13.4t, according to [10].
bucket filling, the wheel loader software used in this study was modified to limit engine
speed, and thus traction force, whenever the transmission’s 1st gear was automatically
engaged, which occurs only during bucket filling (all machines normally start in 2nd
gear).
reduced to 62% in 1st gear (see the top left diagram in Figure 5), which resulted in a
maximum engine speed that was comparable to what expert operators on average used
the wheel loader in. This setting also avoided wheel spin. There was a slight impact on
speed of the hydraulic functions (due to the rotational speeds of the engine, torque con-
verter and hydraulic pump being connected, see Figure 3 and the explanation in an ear-
lier section).
was reduced to 47% in 1st gear (see the bottom left diagram in Figure 5). We deliber-
ately used an engine speed limit below what was needed for the working task planned,
which also affected the speed of the hydraulic functions more noticeably. Furthermore,
the maximum obtainable traction force in 1st gear was even lower than what was avail-
able in 2nd gear, all of which an operator should experience as clearly negative.
Figure 5. Effect of engine speed limitation on traction force and on resulting lifting force
The 3D diagram to the right in Figure 5 shows that the counter-acting effect of the
traction force is clearly less pronounced for the 62% and 47% settings, and that even at
maximum use of the traction force available the lifting force is never cancelled out
completely.
In this article the three tested machine variants will be referred to by their respective
limitation of maximum traction force as “100%”, “62%”, and “47%”.
3.3 Apparatus
During all sessions various data were recorded off the wheel loader’s CAN bus, using
Vector Informatik’s CANcaseXL equipment and CANalyzer software.
Not all signals of interest were readily available, which in some cases required modi-
fication of the software for engine and transmission control, while in other cases an ex-
ternal ECU (Parker IQAN MDL) was used to place calculated signals on the CAN bus.
Additional data from externally mounted sensors for lift and tilt cylinder stroke and
lever position for lift and tilt were acquired and placed on the CAN bus. A modified
LoadTronic system from AADI was used to measure the net weight of the bucket. The
data were transmitted to the machine controller via SAE J1587 protocol and then
merged into the data stream on the CAN bus.
All tests were recorded on video using an externally placed digital video camera and
later synchronised with the acquired data from the CAN bus.
3.4 Participating operators

In all, eighteen people, all male, agreed to participate in this study; all of them close
colleagues of the author. Their data have been recorded and handled in a way that pre-
vents drawing conclusions as to their identities (this was especially important for the
part of the study not reported here, where psychophysiological body signals such as
heart rate and respiration rate were recorded). In all documents, as in this article, an
operator is referred to by the day of his test driving, from “04” (tested on October 4th,
2010) to “02” (tested on November 2nd, 2010).
The people asked to participate as test operators in this study are all male, mostly in
order not to subject the other part of the study to possible skewing due to gender differ-
ences in the way the human body reacts to mental workload. The potential operators
were not randomly chosen, but pre-selected according to the author’s preliminary
judgement of their wheel loader operating skills. However before starting the test each
operator was asked to make a self-evaluation, using a visual analogue scale without
values, but with some helpful guiding statements, see Figure 6 to the left.
One operator commented that the last two skill statements should change places, so
that “professional” is the highest ranking. The author’s intention with the scale as
shown in Figure 6 was to beat the innate modesty of most people, which leads to the
phenomenon that almost nobody ever chooses 100% (or close to) as a skill level. Also,
202 Paper VI
there might be several people who have machine testing as their profession, yet they
know of others who they perceive to be even better at operating a wheel loader.
To the right, Figure 6 shows the division of the scale into four skill groups and how
the individual operators are placed in these groups, using their self-evaluated skill ranks.
These turned out to be similar to the author’s preliminary judgement, resulting in nine
operators in skill group 1 (inexperienced), six in skill group 2 (experienced) and three
operators in skill group 3 (expert). This also gives the possibility to combine groups 2
and 3 at a later stage into one “experienced and above” group that matches the “inexpe-
rienced” group in size, which would benefit the statistical power of direct comparisons.
In our pre-selection of potential test operators for this study we chose not to focus on
expert operators alone, but to include operators of all skill groups in order to test
whether there is a difference in how operators of these groups handle and evaluate the
different machine variants. However, we deliberately avoided skill group 0 (newcom-
ers) in order to guarantee at least a theoretical knowledge of how to operate a wheel
loader, and in order to exclude practice effects as much as possible. The ten minutes of
self-training before each live testing session would not have sufficed for newcomers,
but were deemed sufficient for operators of skill group 1 and above.
3.5 Design of the study and procedure

Each operator was given an exclusive 2.5 hours session after lunch, during which all
testing of all three machine variants was performed. In this study, the traction force set-
ting was the independent variable. The two limited settings of 47% and 62% were tested
against the unmodified software version with 100% maximum traction force, in random
order, unknown to the operator. The operators were not told what had been modified
and how, but most of them, especially the more experienced ones, were able to deduce
this fairly quickly.
In order to minimise the skewing influence of learning effects, before testing a ma-
chine version each operator was given ten minutes’ self-training just to familiarise him-
self with the specific characteristics of the current machine version. After the training
followed five minutes’ live test-driving, where the operator was asked to make sure to
fill the bucket completely and use the machine at a normal production tempo, resulting
in a cycle duration of approx. 25 seconds (both of which most of the inexperienced op-
erators did not manage to achieve completely, but at least this demand created a certain
pressure). The session ended with a subjective evaluation where, among other things,
the operator was asked to judge the tested machine version’s ease of bucket filling and
perceived power on a visual analogue scale (without printed values). The operator was
also asked to rank his overall impression of the machine in comparison to all previously
tested variants on a nominal scale (“better”, “worse” or “same”).
Since the study was performed during the autumn, comparable conditions for the op-
erators were ensured by performing tests only on days with good weather (no rain) and
at the time of the highest ambient temperature, i.e. immediately after lunch. The latter
also guaranteed that the operators were rested.
4 Analysis
The measurements conducted during this study resulted in several gigabytes of data,
making automated analysis necessary.
The main parts of the calculations were executed in MathCad worksheets that link to
data files originating from CANalyzer (machine data) and cStress (psychophysiological
data), exported in Matlab and Excel format, respectively.
4.1 Detection of the bucket filling phase

Among other things, it was necessary to develop a function for automatic identification
of bucket filling. While this phase of a short loading cycle clearly ends with the operator
tilting back fully and lifting the bucket out of the pile, then putting the transmission into
reverse and backing away from the gravel pile, the question of when this phase actually
begins is almost philosophical.
In earlier papers [2] it was stated that bucket filling begins when the bucket’s cutting
edge meets the gravel pile. But how can one measure this without involving some com-
204 Paper VI
plicated measurement of absolute position or video analysis? Merely integrating the

rotary speed of the wheels is not sufficient, because the wheels can spin, fully or par-
tially, in both cases, which changes the relationship between machine speed and wheel
speed. Also, the position of the front edge of the gravel pile will change with each load-
ing cycle. This definition is therefore not practical.
It has also earlier been stated that the beginning of the bucket filling phase is marked
by the operator with putting the transmission into 1st gear. While such manual kick-
down is still available, most operators of modern wheel loaders utilise the automatic
function, which among other parameters uses the torque converter slip to calculate
when 1st gear is to be activated. But by then the bucket has already begun to penetrate
the gravel pile, making also this definition impractical.
Some operators use the momentum of the moving machine to put the bucket into the
material; some even deliberately accelerate the machine to higher speeds on their way
towards the place of loading. The energy consumed during this phase is of course not
free. If fuel consumption during bucket filling were to be compared, we would need to
consider this in order to avoid skewed results. Therefore, we can use the shift from 2nd
to 1st gear as a sign that bucket filling is probably ongoing, but the real start might al-
ready have occured, so we have to look back in time in order to include some of the
earlier events.
If the operator is using the machine in continuous cycles, then the earliest the bucket
filling phase can begin is when the machine reverses direction (see phase 9 in Figure 1).
But the operator might also have used the machine from stand-still. Common to both is
that at some point in time the operator steps on the gas pedal in order to accelerate the
machine towards the place of loading. Relying on this, however, can give erroneous
results, in particular for longer return distances.
The method used in this study is based on the experience that the operator, returning
to the gravel pile from the last reversing phase, always lowers the bucket to the ground
only immediately before the beginning of each filling phase. This is utilised in the fol-
lowing algorithm that has been found to be of practical use (Algorithm 1):
Algorithm 1. Detect start of bucket filling
# Description
1 Begin at the point in time where the transmission is shifted from 2nd
gear to 1st
2 Go back in time to where the lift and tilt function begin to be used (be-
fore that, they were not activated)
3 Go back in time even further to where lift and tilt function are just be-
ginning to not be used any more
Figure 7 shows how the algorithm performs on a typical sequence. The grey areas do
not belong to the identified bucket filling phase, which can be seen to start some time
before the transmission is shifted into 1st gear.
Figure 7. Automatically identified bucket filling phase (white area)
Some operators clean the working place by reversing farther away and then putting
the bucket on the ground to scrape off excess material all the way until right in front of
the gravel pile. The algorithm above therefore needs to also specify some maximum
distance from the gravel pile. This can be done by taking the point in time where 1st
gear is activated and integrating wheel speed backward in time. If the resulting distance
reaches the pre-determined threshold, then the algorithm will be stopped. Also, if while
going back in time a gear shift from reverse to forward is encountered, the algorithm
also needs to be stopped.
This algorithm can even be used in real-time by utilising a buffer that after reversing
starts counting from when lift and tilt lever are both no longer used until the transmis-
sion shifts into 1st gear, at which point the bucket filling is confirmed and the previously
buffered time is used. Possibly the stroke of lift and tilt cylinder could also be taken into
account to increase the detection quality, because in rare cases an operator may start to
use the lift function and then stop it again, all before the automatic kick-down function
activates 1st gear.
4.2 Operability measure from machine signals

The main idea in this part of the study was to find some measure that can also be recre-
ated in a simulation and that in the tests conducted in this study is shown to correlate
well with the test operators’ subjective evaluations of the machine versions’ operability,
with the focus on the bucket filling phase.
There are some seemingly obvious choices for such measure, one being time on task,
i.e. how long the operator needs to fill the bucket. But two simple thought experiments
disprove this hypothesis. To begin with, merely by reducing the speed of the lifting
function we could accomplish longer bucket filling phases. However, this would not
206 Paper VI
mean that the wheel loader would have been harder to operate; it would just have taken
longer to lift the bucket out of the gravel pile until the transmission can be set to re-
verse, thus concluding the bucket filling phase. Secondly, it should mean a difference in
operability to a human operator if a bucket can be filled in for example ten seconds just
by activating and holding the levers or if during the same period of time the positions of
the levers require frequent readjustment.
Another seemingly obvious choice would have been task completion, i.e. how much
material the operator manages to load into the bucket. However, up to a certain point
the operator always manages to fill the bucket completely; it is just a matter of effort.
This also invalidates bucket load as a measure of operability.
Another easily acquired measure would have been fuel consumption, which indeed
should correlate fairly well with operability, since it is indicative of how much energy
was required to fill the bucket. However, in the future we also want to compare oper-
ability between different machines. Fuel consumption is impacted to a very large degree
by the component and system losses, which are affected by size and the state of tech-
nology. For instance, fuel consumption for the same size of an engine differs when
comparing versions that comply with different exhaust emission legislations (Tier II vs.
Tier III vs. Tier IV). Also, the drive train can feature similarly strong transmission but
of vastly different technologies (e.g. hydrodynamic or hydrostatic, planetary or counter-
shaft) and therefore different internal power losses. Finally, different hydraulic systems
can be employed that deliver similar power but with different internal losses (e.g. open
centre or load sensing systems, employing fixed or variable displacement pumps). Fuel
consumption during bucket filling is therefore also not a usable measure of a wheel
loader’s operability.
4.3 Operability measure from operator control commands

All of the above measures derived from the machine itself, bucket filling time, bucket
load and fuel consumption during bucket filling, are just secondary effects of the opera-
tor’s way of using the wheel loader. All of them are also profoundly impacted by the
machine’s system design and the choice of components.
A much better way of deriving a measure indicative of operability is to measure
close to the operator. In one part of this study we therefore measured signals on the op-
erators’ bodies: heart rate, finger temperature, respiration rate, CO2 concentration in the
exhaled air, and galvanic skin response. None of these signals are (today) possible to
generate in a dynamic simulation; we need instead to involve human operators in the
process. In the future, this could be done by performing studies in using human-in-the-
loop simulators, for instance like reported in [11].
In the following we will focus on the operator’s three most important control com-
mands: the way the gas pedal, lift lever and tilt lever are used during bucket filling. If
by using the data from the physical tests performed in this study a measure could be
established that shows good correlation with operability, then this would also enable
virtual operability testing by means of operator models, as reported in [2] and other
works.
Some variants of such measures can be dismissed by a group of simple constructed

examples such as in Figure 8.
3 5
2 4 6
Figure 8. Constructed examples for operator control commands
To begin with, one idea similar to measuring time on task was to measure the arc
length of each control command, perhaps also normalised with respect to duration. Us-
ing the examples in Figure 8, this would lead to different values in examples 1 and 2,
which could be acceptable, since in example 1 the operator does not have to do anything
at all. However, though comparable in terms of difficulty, example 3 and 4 would result
in different values, but both lower than example 2, which is less difficult to perform
than both 3 and 4. Furthermore, even though the effort in example 5 seems quite high,
the calculation would give the same value as example 2. On the other hand, even though
less difficult to accomplish, example 6 would result in a higher value than example 5.
Arc length, absolute or normalised to duration, can therefore not be considered useful.
Another idea was to use the area below the control command curve, also this possi-
bly normalised to duration. Using the same examples from Figure 8 we can show that
this is not useful either: example 2 is easiest to accomplish (other than not doing any-
thing, as in example 1), but will result in the highest value possible. Examples 5 and 6
are both more difficult to perform, but will result in lower values than 2. The more dif-
ficult to accomplish, rectangular pulses are added to example 6, the more its calculation
result will approach example 2’s value. Also, the calculation for example 3 and 4 again
will provide different values, even though the degree of difficulty is obviously similar.
What is it then that can be used to describe the difficulty of an operator’s control
command in the time domain? In earlier work, referenced in [2], it was speculated that
calculating an RMS-based “control work dose” similar to the dose of whole body vibra-
tions could be one way forward. In this study we extended this line of thinking and ex-
amined whether the control commands’ variance, skewness or kurtosis might correlate
with the operators’ subjective evaluation of the ease of bucket filling. As it turned out,
none did.
The question above and the mentioning of the time domain also trigger the idea of
using Fourier analysis. Example 2 in Figure 8, while easy to accomplish obviously con-
tains an infinite amount of frequencies. A perfect half sine wave would be difficult to
perform for the operator, but only result in one frequency being detected. In addition to
this mismatch, the question arises of how to weigh the coefficients for the various fre-
208 Paper VI
quencies. The idea of somehow using the frequency spectrum of a signal is indeed in-
triguing, yet definitely not a trivial task.
Another interesting idea comes from the above frequently used variants of the word
“difficulty”, which leads to musing on whether “complexity” would not have been a
better choice. In computer science, the Kolmogorov complexity (or algorithmic en-
tropy) of an object is the least amount of information needed to accurately describe the
object. The optimal description of an object may be virtually impossible to find without
prior knowledge of the object itself. For instance, saving a rendered image of the Man-
delbrot set fractal in PNG file format will result in the use of a compression algorithm
that reduces the file size considerably compared to a raw image file. But smaller yet
would be the representation of the Mandelbrot with the iterative equation that leads to
the image. However, without the knowledge that one simple equation produces such an
intricate image, the most efficient description might have been the PNG file mentioned
above. Is then algorithmic entropy usable to describe the intrinsic difficulty of control
commands? The answer is no. This is not due to the possible failure to find the most
compressed representation, but due to the fact that most of the descriptions will be too
efficient. If we take example 6 in Figure 8, then one way of representing this would be
“rectangular pulse from 0 to 1 with duration t1, repeated after t2.” This would already be
too compressed, because a human operator works serially and cannot just save a pattern
and easily retrieve it later for reuse. Another example would be the aforementioned half
sine wave, hard for an operator to perform but quite efficiently described as “half sine
wave from 0 to 1 with duration t”. Not wanting to rely on theoretical considerations
alone we have nevertheless tried to examine this in practice by measuring the size of
ZIP-files containing control commands exported to text files and binary files. No good
correlation was found.
In all the reasoning above we human beings somehow seem to be able to intuitively
evaluate the performance difficulty of the simple examples in Figure 8. Example 2
seemed easiest of all the time series where the operator control actually was activated.
Examples 3 and 4 seemed comparable and both quite easy to perform. Example 6 was
not too difficult either, while example 5 involved many steps and was therefore hardest.
It thus seems that the number of steps or more precisely the number of break points in
the graph correlates with our understanding of performance difficulty. This measure
will in the following be called control effort. In reality, time series data of recorded op-
erator control commands will not be composed of long straight lines such as in Figure
8. Rather, measurement noise and shaking of the operator’s seat will introduce many
small variations that are not intended by the operator. We cannot therefore merely count
the number of break points, but must first apply some algorithm that simplifies the time
series data by approximating the control command with a minimum of straight lines
without too great an approximation error. Because it is more difficult for humans to
control movements with variable speed, using polynomials of higher order than one (i.e.
lines) is not proposed. Such algorithms will be presented in the next section. Having
applied these, each cycle results in three figures: the number of break points for the sim-
plified control commands for gas pedal, lift lever and tilt lever. As the control com-
mands are essentially orthogonal, a natural way of combining their associated control
effort would be to calculate the Pythagorean sum.
2 2 2
nsum = n gas + nlift + ntilt
(1)
where n is the number of break points, also called control effort.
4.4 Approximation of control commands with graphs composed of

straight lines
The goal is to approximate the operators’ control commands as shown in Figure 9, here
using the gas pedal signal from Figure 8 as an example.
Figure 9. Gas pedal signal, measured and approximated
Researching this sub-problem, we found that a similar task is typical within the field
of mobile robotics: line extraction from a cloud of points resulting from a 2D laser scan.
Several algorithms have been developed and presented, one of the most common being
versions of Split and Merge, as for example presented in [12].
Algorithm 2. Split and Merge (SM)
# Description
1 Start with one line, connecting first and last point in data set
2 Create list of all points in-between, in descending order by vertical dis-
tance from the line currently under examination
3 Check list of potential break points, beginning from top until a valid
break point is found, that splits the original line into two lines, whose
combined residual sums of squares plus a pre-defined threshold is
smaller than the residual sum of squares of the original line
4 If valid break point found, then split up original line into two
5 Repeat 2 – 3 recursively for all line segments until no more valid break
points can be found for any line segment
6 Start with break point between first and second line
7 Check if break point is invalid and allows for merging the lines to the
left and right of it to one line. Invalid if the original lines’ combined
residual sums of squares plus a pre-defined threshold is larger than the
residual sum of squares of the merged line
8 If invalid break point found, then merge the two line segments into one
9 Repeat 7-8 recursively for all line segments until no more invalid break
points can be found
210 Paper VI
In addition to a variant with least-squares line fitting, which proved to be superior in

terms of speed and accuracy, five other algorithms are presented and benchmarked in
[13]. In our work we used the SM variant shown in Algorithm 2.
Steps 1 – 5 in Algorithm 2 are the Split phase, where all line segments are examined
as to whether they can be split up into two segments. A pre-defined threshold added to
the combined residual sums of squares of the two new lines regulates this process. A
split is only allowed when this sum is smaller than the residual sum of squares of the
original line segment. After this, steps 6 – 9 perform the Merge phase where all previ-
ously found line segments are examined to determine whether they can be merged into
one line. The same pre-defined threshold regulates this process in the same way as be-
fore. We used different values for the operator’s gas pedal and lever commands: 1000
and 500, respectively. These settings were manually derived by experimenting with
some sample signals and visually inspecting the overall quality of the approximation.
In our work we also developed and tested an algorithm of our own, called Stretch
and Relax. In contrast to our version of the Split and Merge algorithm (Algorithm 2),
the line-fitting in Stretch and Relax makes use of the least squares method. The princi-
pal steps for the SR-A variant can be seen in Algorithm 3 (which actually performed
slightly faster than SM in our testing).
Algorithm 3. Stretch and Relax – Absolute (SR-A)
# Description
1 Start with first data point (S1)
2 Use next point as break point candidate BP1
4 Use BP1 as preliminary start point S2 for next line
5 Use next point as future break point candidate BP2
8 If index of BP2 ≥ BP2max then BP1max = BP1 and BP2max = BP2
10 Use BP1max as break point for line L1 and as new start point S1 for
next line segment L1
11 Repeat steps 2-10 until end of data set is reached
Also for this algorithm, the threshold values used for gas pedal and lever commands
of the operator differed. We used 2000 and 500, respectively. These settings were de-
rived in the same way as before, i.e. by experimentation with samples and visual inspec-
tion.
We further examined two additional variants of the Stretch and Relax algorithm,
called SR-RL (Relative, Linear) and SR-RS (Relative, Square root). The idea here was to
encourage the algorithm to create longer line segments by normalising the residual sum
of squares to the length of the line, either proportionally (SR-RL) or to the square root of
the segment length (SR-RS). However, this invalidated some optimisations that were
included in the original SR-A version and resulted in very long execution times, with
only minor improvements in the quality of the approximation. While all operator con-
trol commands had still been processed also using these algorithm variants, the results
are not essentially different from SM and SR-A, and are therefore not presented in this
article.
After finishing all calculations and signal processing, during writing of this article,
we also discovered another variant of achieving straight-lines approximations of time
signals [14], but limited to break points being stationary points. In their work, the au-
thors introduced the steering wheel reversal rate as a measure for assessing the impact
of secondary tasks (e.g. the interaction with in-vehicle information systems) on the lat-
eral control performance of vehicle drivers (e.g. lane keeping). The proposed measure
captures the number of steering wheel reversals per time unit and is derived by first
low-pass filtering the signal and then applying algorithms to find stationary points and
then reversal points.
4.5 Statistical analysis

A brief summary of the setup of our study reveals the possibility for both sub-studies
with between subjects design and sub-studies with within subjects design:
- Eighteen operators participated.
- Each operator can be sorted into one of three skill groups according to his
self-evaluated skill level.
- If only two skill groups are to be used then there is the same number of op-
erators in skill group 1 and the combined skill group 2+3
- Each operator was tested under similar conditions.
- There was no particular order between the operators; they were booked ac-
cording to availability.
- Each operator successively tested three different software settings of the
otherwise unchanged machine.
- The order of the tested machine versions was randomised.
- Each machine version was first tested for ten minutes to minimise practice
effects, followed by five minutes of live test-driving.
- Five minutes of live testing resulted in at least six bucket fillings, often
more, depending on operator
- As practice effects are excluded by the ten minutes of self-training, there is
no particular order for the cycles, i.e. the measurements taken can be seen
as random.
212 Paper VI
- A normal distribution can be assumed due to good repeatability for the

same combination of machine variant and operator.
- Subjective evaluations were captured on visual analogue scales with free
comments, performed directly after each live test-driving session.
One interesting aspect to study was how time on task (i.e. bucket filling time), task
completion (i.e. bucket load) and fuel consumption vary for different operators, but also
for different skill groups (a typical between subjects design).
However, this article is concerned with finding a measure indicative of operability,
with the hypothesis that the operator’s control commands can be used. For this, we re-
gard each operator as a sub-group and do not compare between operators, but only be-
tween machine variants test-driven by the same operator. It is thus a within subjects
design with repeated measurements.
At first glance, this appears to be a case for a Repeated Testing ANOVA. But then
again, we have only three conditions (100%, 62% and 47% traction force), and we
would need to perform a post-hoc analysis anyway, so it seems to involve less work to
just calculate t-tests for the three combinations 100%-62%, 100%-47% and 62%-47%
for each operator. One could then assume that a paired (related) t-test would be appro-
priate, but the design of our study does not really fit the usual examples from medicine
with “pre treatment” and “post treatment”, or from economics/agriculture with the
amount of goods produced per factory/field in successive years. A paired t-test would
operate on the difference between sample 1 of condition 1 and sample 1 of condition 2,
then the difference between sample 2 of condition 1 and sample 2 of condition 2, and so
on. But in our study the measurements within subject (operator) and for the same test
condition (machine variant) are not related and not in any particular order; there is no
practice effect nor any other causality between the cycles. It was therefore decided to
perform unrelated t-tests for all three combinations of machine variants per operator.
The goal was not to derive a regression function with the operators’ skill level and
control efforts as independent variables and the operators’ subjective evaluation results
for ease of bucket filling etc as dependent variables, but rather to show that there is a
monotone relationship that can be used for comparative studies. This can be done visu-
ally or as simple comparison, without the need to model the influence of operator skill.
As statistical analysis it is sufficient to show that the data for the different conditions
within the subjects do not share the same mean – which a common t-test provides.
As mentioned before, the operator’s control commands, and thus the control efforts,
can be considered orthogonal, comparable to Cartesian coordinates. If for example two
three-dimensional locations are to be compared against each other, then three separate
comparisons can be made for each coordinate, and in order to prove that a difference
exists between the locations it is sufficient to show a difference in one such comparison.
The same reasoning applied to the data from our study means that two sets of operator
control commands (each set consisting of three signals: gas pedal, lift lever and tilt
lever) can not be considered to be different only if we fail to reject the null hypothesis
for all three individual t-tests. The combined value for α, the probability of having
made at least one Type I error (a false positive; having declared some difference being
significant (unequal means), when in fact there was no difference) is the product of all
three individual probabilities:
α sum = α gas ⋅ α lift ⋅ α tilt (2)
Since the decision was made to calculate t-tests for the three combinations 100%-
62%, 100%-47% and 62%-47% for each operator, the problem of repeated measure-
ments needs to be addressed. Generally, performing m t-tests on the same data set will
give a family-wise error rate of m times the error rate per comparison. If we test three
times for α = 0.05 on the same set of data, then in total there is a 15% chance of having
made at least one Type I error, provided comparisons are independent; if they are not
independent, then 15% would be the upper boundary. In conclusion, we need to prove a
lower α in the individual t-tests.
Instead of checking against a pre-decided confidence level we used an explorative
approach and inversely calculated the α value in each individual t-test. We combined
these values according to equation (2) and added these combined probabilities in order
to derive the upper-most boundary for the probability of a Type I error anywhere in the
result set of the examined sub-group.
3
α total = ∑ α sum i (3)
i =1
where i is the index of combination (100%-62%, 100%-47% and 62%-47%)
We set the upper limit for αtotal to 0.01, which gives a 99.9% statistical confidence
per sub-group of this study.
5 Results
This article is concerned with finding a measure indicative of operability, hypothesising
that the operator’s control commands can be used and that a monotonic relationship
between the operators’ total control effort and their subjective evaluation of ease of
bucket filling (etc) can be shown. This section will therefore not discuss other interest-
ing results, for instance how bucket filling time, bucket load, and fuel consumption vary
between operators depending on their skills.
5.1 Subjective machine evaluations

ants, especially in light of their operating skill. Figure 10 shows that the operators in
group 3, i.e. the experts, share the clear opinion that the original machine (100% trac-
214 Paper VI
tion force) is the best, the wheel loader limited to maximum traction force of 62% next
best, and the extremely limited machine worst. This opinion is predominant even in
skill group 2, the experienced operators, but two had already ranked the 62% variant as
same or better than the 100% machine.

100% 62% 47% 100% 62% 47% 100% 62% 47%
Ranking
Ranking
Ranking

2+3 Machine variant

Ranking
2 6
Worst 9 2+3 9
Figure 10. Results of the operators’ machine ranking, skill groups 1, 2, 3, and combined 2+3
The operators in skill group 1, the inexperienced operators, also follow the same vot-
ing pattern to a certain degree, but deviate more often. Two operators in this group ac-
tually ranked the 47% machine as best, and one ranked the 100% variant as worst.
In this context it is of interest to analyse whether a relationship exists between the
for ease of bucket filling. Figure 11 shows that such a relationship is given most of the
time, but is somewhat less distinct for operators in skill group 1.
1 Machine rank 2 Machine rank 3 Machine rank

Best Med. Worst Best Med. Worst Best Med. Worst
Ease of bf
Ease of bf
Ease of bf
Highest 8 2 Highest 6 Highest 3

Lowest 1 1 7 Lowest 1 5 Lowest 3
2+3 Machine rank

Best Med. Worst Matrices show operators’ added assessments
Ease of bf
Highest 9 1 9 of ease of bucket filling for ranked machines.

2 6
Lowest 1 8 2+3 9
Figure 11. Comparison of operators’ assessment of ease of bucket filling vs. machine ranking
The same analysis made for machine power, also evaluated by the operators using a
visual analogue scale, is shown in Figure 12. The same picture emerges as before; the
expert operators in skill group 3 are most distinct in their judgements, the experienced
operators in group 2 are quite distinct with the exception of one, while the inexperi-
enced operators in skill group 1 show more deviations from the expected outcome of the
best ranked machine also having the highest score for machine power.

M. power
M. power
M. power
Highest 7 1 1 Highest 6 Highest 3

Lowest 1 1 7 Lowest 6 Lowest 3
2+3 Machine rank

M. power
Highest 9 1 9 of machine power for ranked machines.

2 6
Lowest 9 2+3 9
Figure 12. Comparison of operators’ assessment of machine power vs. machine ranking
From the spontaneous comments, noted during the study, we know that several op-
erators were unsure how to rank a machine variant. Several, especially inexperienced
operators seem to correlate machine ranking with their impression of the machine’s
power, even though they gave a lower score for ease of bucket filling, presumably
blaming themselves for not being able to utilise the power available. One operator
commented on this and reasoned that bucket filling is an important phase in a loading
cycle and that he will therefore give one particular machine variant a lower rank due to
him finding it more difficult to fill the bucket, even though the machine received the
highest score for power.
Three operators in skill group 1 also made the interesting comment that the machine
with the least power, the 47% variant, seemed to be more powerful than the other vari-
ants in the moment right after finishing bucket filling, i.e. when backing away from the
gravel pile. Analysing the test data, it can indeed be seen that both machine speed and
engine speed ramp up faster for the 47% machine than for the others. But we can also
see that in these cases the operator activates the gas pedal sooner, while activating the
lift lever later than for the other machines. In some cases the operator never releases the
gas pedal, which is understandable given that the traction force of the 47% variant is so
low that wheel spin and total cancellation of lifting force are virtually impossible, thus
eliminating the need to modulate engine speed. It is plausible that the operator in this
machine learns to be less sensitive with the gas pedal during bucket filling, thus using it
more aggressively also in the moment immediately following. Furthermore, with less
traction force available the bucket load is usually also the lowest in the 47% machine,
thus requiring less engine torque to lift, and therefore having more torque available to
accelerate the engine and the whole machine.
5.2 Control effort

Earlier we discussed measurable signals that might correlate to the operators’ subjective
assessments of the tested machine variants’ operability. Their usefulness in this context
was argued against by theoretical reasoning; however we still examined most of them
more closely, using the test data to prove the reasoning correct, i.e. the correlation was
indeed limited or non-existent. In this section we will not present these results but in-
stead focus on the method that actually provided the monotone relationship we were
seeking, the use of control effort according to equation (1) and application of Algorithm
216 Paper VI
2 (SM) or Algorithm 3 (SR-A). Both of these algorithms gave similar results, we will
therefore focus on SR-A.
Figure 13 shows that a strong monotone relationship between control effort and the
operators’ subjective evaluation of the ease of bucket filling could be found in five
cases, all of them being operators with higher skills. In five more cases, most of them
operators with medium skills, we were able to find at least a partial monotone relation-
ship, meaning that the control effort data for the machine ranked in-between highest and
lowest ease of bucket filling were not significantly different from one of their
neighbours.
Strong monotone relationship between 27 22

control commands and self-assessed 28
ease of bucket fill or machine ranking 14 25
Partial monotone relationship 26 20

(e.g. only between best and 29
worst machine or only for machine rank 08 06
but not for ease of bucket filling)
11
18
04 13 21
12 02
Operator skill 05
(self-assessed)
0 25 50 75 100

Figure 13. Existence of a relationship between operator control effort and ease of bucket filling
Figure 14 shows details for all operators where a strong monotone relationship be-
tween control effort and the operability-related results of the subjective evaluations was
found. The statistical analysis reveals a significance of above 99.9% (α< 0.001).
Figure 15 shows details for all operators where a partial monotone relationship be-
tween control effort and the operability-related results of the subjective evaluations was
found. In these cases we either see that the control effort data from one machine variant
is too close to another, thus giving less statistical significance in that specific compari-
son (e.g. operator 20), or we see a monotone relationship between control effort and
machine ranking and machine power, but not ease of bucket filling (operator 08). While
not perfectly good, there are nevertheless signs of the same type of relationship that the
operators shown in Figure 14 exhibit fully.
Figure 14. Operators showing a strong monotone relationship btw. control effort and operability
218 Paper VI
Figure 15. Operators showing a partial monotone relationship btw. control effort and operability
We chose to look for a relationship between control effort and primarily ease of
bucket filling, secondarily to machine rank, because these are closest connected to oper-
ability. In some cases control effort shows a clearer monotone relationship with ma-
chine power, but not so in other. Bearing Figure 10, Figure 11, and Figure 12 in mind,
we were also curious as to whether we could see a clearer relationship between control
effort and traction force, which would show that several operators were making errone-
ous subjective assessments, but their bodies responded “correctly”. Figure 16 shows the
results of this analysis: two operators from skill group 1 (operators 05 and 13) now
show a partial monotone relationship and one operator from group 2 (operator 02) even
exhibits a strong monotone relationship between control effort and traction force, while
earlier there was none at all between control effort and subjectively evaluated ease of
bucket filling. While interesting, this presumably only shows that some operators are
not really accustomed to assessing machine properties via visual analogue scales.
27 28
Strong monotone relationship between 22
control commands and traction force
14 02 25
Partial monotone 26 13 20
relationship 05
(only between best and worst machine) 08 29 06
11
18
04 21
12
Operator skill
(self-assessed)
0 25 50 75 100

Figure 16. Existence of a relationship between operator control effort and traction force
220 Paper VI
6 Discussion of possible sources of error

During the testing phase of this study we tried to keep the environmental conditions as
similar as possible. We only tested on days with good weather (no rain) and at the time
of the highest ambient temperature, i.e. immediately after lunch (which also ensured
that the operators were rested). But since this study was performed in the autumn, tem-
peratures were generally lower; during the testing around 5-10°C. The machine was
parked outside overnight and had therefore cooled quite substantially by the morning.
Together with 20-30 minutes’ engine warm-up each day (about two hours before test-
ing) and a 1.5km drive to the testing site, the first ten minutes of self-training should
provide enough waste heat to warm up the machine (engine, transmission, axles, and
hydraulics), so that testing conditions were similar for all cycles, independent of order.
The test machine’s automatic functions were fixed at gear-dependent BSS (Boom
Suspension System), automatic APS (Automatic Power Shift), and with Auto Kick-
down enabled. Of course, the BSS is only deactivated when in 1st gear, which means it
is activated during the first seconds of the bucket filling phase, until the transmission
automatically shifts from 2nd to 1st gear. This is done via an adaptive algorithm, which
might differ between the operators. Perhaps one of the pre-defined fixed APS settings L
(low power), M (medium power) or H (high power) should have been used instead. Or
perhaps the automatic kick-down function might have better been disabled altogether, in
order to give the operator the freedom to choose when to shift gear.
The operators were asked to use the wheel loader for three times 10 + 5 minutes, in
total 45 minutes. There is a risk of fatigue, especially for the less experienced operators.
However, the most extreme evaluations of choosing the 100% machine as worst or the
47% machine as best are hard to explain by this. Of course, in spite of the ten minutes
of self-training before each live testing session, a practice effect might still be present,
especially for the less experienced operators.
As mentioned earlier, it could also be the case that some of the inexperienced opera-
tors were also unaccustomed to using visual analogue scales. Their memory of the other
tested machine variants might not have been correct and they could thus have skewed
their own evaluation results. Though operators were allowed to look at their previously
assessments for reference, hardly anyone asked for this opportunity.
While all the possible sources of error listed should be avoided next time, their total
impact on the results of this study is considered to be minor.

In this study of eighteen wheel loader operators, test-driving a machine in three differ-
ent traction force settings, we found ten individual cases that support the hypothesis that
the operator’s control commands can be used to assess the machine’s operability, at
least in the form of ease of bucket filling. The methods chosen to derive the control ef-
fort, our Split and Merge variant and the Stretch and Relax, Absolute algorithm, worked
well and were computationally efficient. The statistical analysis gave a confidence of
above 99.9%.
This provides us with a tool to evaluate test data not only from physical testing, but
also from virtual testing by using the control commands of an operator model similar to
the ones presented in [2], [5] and [6]. Future studies will examine this in more detail.
The findings of this study are not limited to wheel loaders, but should be generally
applicable, in particular for working machines.
Acknowledgments
It is gratefully acknowledged that this research has been supported by the Energy &
Environment programme within the Swedish Vehicle-Strategic Research and Innova-
tion programme FFI, administered by the Swedish Energy Agency, Energimyndigheten.
References

Construction Machinery”. Licentiate thesis, Division of Fluid and Mechanical En-
gineering Systems, Department of Mechanical Engineering, Linköping University,
Sweden.
struction Machinery”. Licentiate thesis, Department of Mechanical Engineering,
Linköping University, Sweden.
[3] Filla, R. (2009) “A Methodology for Modeling the Influence of Construction Ma-
chinery Operators on Productivity and Fuel Consumption”. Proceedings of
HCII 2009: Digital Human Modeling, LNCS 5620, pp 614-623, 2009.
http://dx.doi.org/10.1007/978-3-642-02809-0_65
[4] Elezaby, A.A., Abdelaziz, M. and Cetinkunt, S. (2008) “Operator Model for Con-
struction Equipment”, Proceedings of IEEE/ASME 2008 International Conference
on Mechatronic and Embedded Systems and Applications, MESA 2008, pp 582-
585.
[5] Elezaby, A. A. (2010) “Virtual Autonomous Operator Model for Construction

Equipment Applications”. Doctoral thesis, University of Illinois at Chicago, Chi-
cago, IL, USA.
http://gradworks.umi.com/34/17/3417331.html
222 Paper VI
[6] Thiebes, P. and Vollmer, T. (2011) “Modellierung des Fahrers zur Untersuchung
von Antriebssträngen in der 1D-Simulation am Beispiel eines Radladers mit
Hybridantrieb”, Proceedings of 3. Fachtagung Hybridantriebe für mobile Ar-
beitsmaschinen, pp 47-59.
[7] Norris, W. R., Screenivas, R. S., Zhang, Q. (2002) “A Novel Real-Time Human
Operator Performance Model for Performing Adaptive System Design”. Proceed-
ings of the 2002 Conference of American Society of Agricultural Engineers,
Automation Technology for Off-Road Equipment, pp 287-306.
NASA, USA.
[9] Filla, R. (2009) “Hybrid Power Systems for Construction Machinery: Aspects of
IMECE 2009, vol. 13, pp 611-620, 2009.
[10] Volvo CE (2009) “Product brochure L60F, L70F, L90F”.
http://www.volvoce.com/constructionequipment/europe/en-
gb/products/wheelloaders/wheelloaders/L70F/Pages/specifications.aspx
[11] Löfgren, B. (2009) “Kinematic Control of Redundant Knuckle Booms with

Automatic Path-Following Functions”. Doctorial thesis, Royal Institute of Tech-
nology, Stockholm, Sweden.
[12] Zhang, L. and Ghosh, B. K. (2000) “Line Segment Based Map Building and Lo-
calization Using 2D Laser Rangefinder”, Proceedings of 2000 IEEE International
Conference on Robotics & Automation, pp 2538-2543.
http://dx.doi.org/10.1109/ROBOT.2000.846410
[13] Nguyen, V., Martinelli, A., Tomatis, N. and Siegwart, R. (2005) “A Comparison
of Line Extraction Algorithms using 2D Laser Rangefinder for Indoor Mobile Ro-
botics”, Proceedings of IROS 2005: IEEE/RSJ International Conference on Intel-
ligent Robots and Systems.
http://dx.doi.org/10.1109/IROS.2005.1545234
[14] Markkula, G. and Engström, J. (2006) “A Steering Wheel Reversal Rate Metric
for Assessing Effects of Visual and Cognitive Secondary Task Load”, Proceedings
of the 13th ITS World Congress.
http://trid.trb.org/view.aspx?id=847195

Paper VII
A Case Study on Quantifying the Workload
of Working Machine Operators by Means of
Psychophysiological Measurements
Reno Filla (1), Erik M. G. Olsson (2), Bo H. C. von Schéele (3),
and Kjell Ohlsson (4)
(1)VOLVO CONSTRUCTION EQUIPMENT AB, ESKILSTUNA, SWEDEN
(2)DEPT. OF PUBLIC HEALTH AND CARING SCIENCES, UPPSALA UNIVERSITY, SWEDEN
(3) STRESS MEDICINE AB, BERGVIK, SWEDEN
(4) DEPT. OF MANAGEMENT AND ENGINEERING, LINKÖPING UNIVERSITY, SWEDEN
Abstract
In this study of eighteen wheel loader operators, test-driving a machine in three differ-
ent traction force settings, we examine if a workload index derived from psychophysi-
ological measurements of heart rate, finger temperature, skin conductance, respiration
rate and end-tidal CO2-concentration in exhaled air can be used to assess operator
VII
workload in sufficient detail to use it as a complement to traditional subjective evalua-
tions, and also to use such measurements in a workload-adaptive operator assistance
system in a longer perspective.
Keywords: operator, working machines, mental workload, physical workload, operabil-

ity, human-machine interaction
If real is what you can feel, smell, taste and see,
then 'real' is simply electrical signals interpreted by your brain.
(Morpheus in the film "The Matrix")
This paper has been submitted for publication as:

Filla, R., Olsson, E. M. G., von Schéele, B. H. C. and Ohlsson, K. (2011) “A Case
Study on Quantifying the Workload of Working Machine Operators by Means of
Psychophysiological Measurements”. Submitted in August 2011 for publication in
International Journal of Human Factors and Ergonomics.
http://www.inderscience.com/browse/index.php?journalID=377
A Case Study on Quantifying Workload… 227
1 Introduction
Working machines in construction, mining, agriculture, and forestry are not only be-
coming ever more sophisticated in terms of digital control, but are also complex in ar-
chitecture. Many of these machines consist of at least two working systems that are
used simultaneously and the human operators are essential to the performance of the
machines in their working place.
With their productivity and energy efficiency being linked to the operators’ work-
load, it is essential to make working machines easier to use. There are a great many op-
portunities in new human-machine interfaces, made possible by extensive computer
control of the machine and its sub-systems. One promising idea is to make the whole
system adaptive to the operator’s workload and offer assistance in critical situations.
Contrary to the function of Active Safety systems, such workload-adaptive operator
assistance functions ought to be non-intrusive in order to not disturb the workflow. This
requires online methods for automatically detecting critical workload situations robustly
and in time.
Another goal of our research is to support product development by complementing
the traditional subjective evaluation of a prototype machine’s operability (involving
asking the operator to rate the machine according to a specific scale) with a less subjec-
tive measure based on the operator’s workload. Ideally, this measure would also be cal-
culated automatically and online in order to be available during or immediately after
testing.
This article will report on the setup and results of a larger study conducted to investi-
gate if common psychophysiological measurements of finger temperature, heart rate,
respiratory rate, CO2 concentration in exhaled air, and galvanic skin response can be
used to quantify the workload of wheel loader operators during operation of the ma-
chine.
2 Wheel loader application

In our ongoing research, a wheel loader was chosen as the object of study, although
working machines of similar complexity and with similar challenges can be found in
construction, agriculture, forestry and mining.
The short loading cycle shown in Figure 1 is highly representative of the majority of
a wheel loader’s application. Typical of this cycle is bucket loading of granular material
(for instance gravel) on an adjacent dump truck (or other load receiver, mobile or sta-
tionary) within a time frame of 25-35 seconds, depending on working place setup and
how aggressively the operator uses the machine. A detailed description with identifica-
tion of all phases can be found in [1].
228 Paper VII
Visualising test results obtained in earlier studies, Figure 2 shows that the fuel con-
sumption rate (expressed in volume or mass per time unit) is approximately 60% higher
during bucket filling than the cycle average. Expressed in absolute values, bucket filling
accounts for 35-40% of the mean total fuel consumption per cycle, yet the time spent on
filling the bucket is only 25% of the average cycle time.

reversing point and steers the wheel loader to accomplish the characteristic V-pattern of
a short loading cycle. The lifting function is engaged the whole time. The operator
without delay. Figure 2 shows that the fuel consumption rate for phases 2 to 6 is ap-
proximately constant and close to the average fuel consumption rate of the complete
short loading cycle.
In the remaining phases, the bucket needs to be lowered and the operator steers the
wheel loader back to the initial position in order to fill the bucket again in the next cy-
cle. Phases 7-10 are less energy-demanding and the fuel consumption rate is therefore
lower than the average for the complete short loading cycle.
The higher fuel consumption rate during bucket filling warrants a closer look at this
phase. The inner loop in Figure 3 shows how the human operator interacts with the
wheel loader. In order to fill the bucket, the operator needs to control three motions si-
multaneously: a forward motion that also exerts a force (traction), an upward motion
(lift) and a rotating motion of the bucket to load as much material as possible (tilt). This
is similar to how a simple manual shovel is used. However, in contrast to a manual
shovel, the operator of a wheel loader can only observe, and cannot directly control
these three motions. Instead, he or she has to use different subsystems of the machine to
accomplish the task. The gas pedal controls engine speed, while the lift and tilt levers
control valves in the hydraulics system that ultimately control movement of the link-
age’s lift and tilt cylinder, respectively.
Lifting +
Breaking/Tilting
ECU
Auxiliaries
Gravel
pile
External (+/-)
ECU
Travelling/
Drive train Wheels
Penetration
tion and via the hydraulic pumps also affects the speeds of the lift and tilt cylinders. The
linkage between the hydraulic cylinders and the bucket acts as a non-linear planar
230 Paper VII
transmission and due to its design a lift movement will also change the bucket’s tilt an-
gle and a tilt movement affects the bucket edge’s height above the ground.
Furthermore, Figure 3 also shows how in the outer loop the primary power from the
diesel engine is split up between hydraulics and drive train in order to create lift/tilt
movements of the bucket and traction of the wheels, but is connected again when filling
the bucket from for example a gravel pile. Figure 4 shows that in this situation, the trac-
tion force from the drive train, acting between wheels and ground, creates a reaction
force between gravel pile and bucket edge, which in turn creates a moment of forces
around the loading arm’s centre of rotation that counteracts the moment of forces cre-
ated by the push of the lift cylinder.
Figure 4. Force balance during bucket filling
In effect, the traction force, indirectly controlled by the operator through the engine
speed, reduces the available lifting force. Depending on lifting height and applied trac-
tion force, the lifting force can be completely cancelled out, and the bucket cannot be
moved upwards any further. The operator then has to either apply the tilt function or
reduce the traction force by reducing the engine speed – which is rather counter-
intuitive as a common mental model of vehicle drivers is that an increase in gas pedal
deflection leads to more torque from the engine until any external resistance can be
overcome.
Another challenge for the operators exists in that the hydraulic levers control the
speed of the hydraulic cylinders, not the force exerted by them. In an old-fashioned less
energy-efficient hydraulic system of the Open Centre type a higher external load would
result in a lower cylinder speed for the same amount of lever actuation. This would give
the operator an indication of the external load acting on the cylinder. This is no longer
the case in a modern load-sensing hydraulic system.
In summary, there are many interdependencies and it thus takes a certain amount of
training to be able to use a wheel loader efficiently. It is therefore of great interest to
develop operator-assisting functions.
Contrary to the function of Active Safety systems, any support system that helps an
operator to use the machine in a highly productive and efficient manner ought to be
non-intrusive in order to not disturb the workflow. As with all such software features,
timing and intensity of activation are essential and requirements will vary for different
people and applications. Highly experienced operators might not be the primary target
group and will (at least initially) probably even turn off any operator-assisting function
(compare this with professional truck drivers’ attitude to automatic gear shifting in the
USA). However, the less experienced an operator is, the higher the gain will be from
using such support systems.
It can be argued that the optimum would be a system that is not only aware of the
state of the machine, its environment and application (mission), but also of the opera-
tor’s state and capabilities. Such a system would then be able to provide individually
tailored support in a guiding, rather than an interfering manner. This was proposed
some time ago but has still not been fully realised even though progress in this regard is
being made in aviation [2]. Among other things, this requires online methods for detect-
ing critical operator workload situations robustly and in time.
3 Quantifying human workload with

psychophysiological measurements
There are many aspects of how our physiological systems interact with physical and
mental workload. Heart rate (HR) and respiration rate (RR) are classical measures of
physical workload where a higher HR and RR result in a better supply of oxygen to the
working body. Both HR and RR also interact with mental workload.
Other physiological measures involved mainly in mental workload are skin conduc-
tance level (SCL) derived from palm sweat gland release and peripheral temperature
(typically measured in fingers), derived from smooth muscles constricting the periph-
eral blood vessels. A higher workload is associated with higher SCL and lower finger
temperature. However, the magnitude of changes is dependent on individual vasocon-
striction and sweat gland activities.
All physiological parameters are highly influenced by autonomic nervous system
(ANS) activity and mainly by the sympathetic branch but heart rate variability (HRV)
and respiratory sinus arrhythmia (the portion of HRV that is related to breathing) are
associated with both sympathetic and parasympathetic nervous system activity.
Respiration is regulated in a more complex way which includes cognitive, chemical,
and metabolic systems as well as the ANS. When subjected to mental or physical chal-
lenges, ventilation increases, which in turn results in a drop in plasma pCO2. The pCO2
level can therefore be used to assess whether breathing is in excess of metabolic de-
mands resulting in a low pCO2, which is the definition of hyperventilation [3]. A change
in pCO2 level is the result of changes in RR and/or tidal volume. Some researchers have
pointed out that CO2 level can be an advantageous measure of psychosocial stress, since
it can differentiate between metabolic and psychosocial demands, which in most other
measures generate the same response [4]. The pCO2-level is often, as in our study,
232 Paper VII
measured as end-tidal CO2 (EtCO2) which is satisfactorily correlated to pCO2 and per-
mits assessment of breath by breath variation. The physiological systems mentioned
interact with each other in a complex way, cooperating with and compensating for each
other [5]. In case of immense stressful events people tend to stop breathing, with a di-
rect impact on several regulatory systems as a consequence.
Operating a wheel-loader involves both physical and mental workload. Even though
you sit still and the controls do not demand a large amount of power you have to keep
your balance and be prepared. Mentally you have to maintain attention and respond
appropriately all the time. Measuring this workload therefore calls for a measure that
handles several physiological signals.
4 Method
4.1 Machine setup
The wheel loader used was a Volvo L70F equipped with a general purpose bucket with
a load capacity of 2.3m3. The total operating weight of the machine is approximately
13.4t [6].
bucket filling, the wheel loader software used in this study was modified to limit engine
speed, and thus traction force, whenever the transmission’s 1st gear was automatically
engaged, which occurs only during bucket filling (all machines normally start in 2nd
gear).
reduced to 62% in 1st gear, which resulted in a maximum engine speed that was compa-
rable to what expert operators on average used the wheel loader at. This setting also
prevented wheel spin. There was a slight impact on the speed of the hydraulic functions
(due to the rotational speeds of the engine, torque converter and hydraulic pump being
connected, see Figure 3 and the explanation in an earlier section).
was reduced to 47% in 1st gear. We deliberately used an engine speed limit below what
was needed for the working task planned, which also affected the speed of the hydraulic
functions more noticeably. Furthermore, the maximum obtainable traction force in 1st
gear was even lower than what was available in 2nd gear, all of which an operator would
experience as clearly negative.
In both conditions the counteracting effect of the traction force is clearly less pro-
nounced so that even at maximum use of the traction force available the lifting force is
never cancelled out completely. This should be experienced positively, especially by
less experienced operators (but the question is whether this positive aspect outweighs
the artificial limitation of maximum available traction force).
The test machine’s automatic functions were set to gear-dependent BSS (Boom Sus-
pension System is deactivated when in 1st gear), Automatic Power Shift (automatic gear
shifting of the transmission) and Auto Kick-down enabled (automatically changing base
gear from 2nd to 1st when needed). We also disabled the function for automatic bucket
tilt-in stop at level position.
The radio and most of the external displays were turned off in order to not distract
the operator from the working task. However, the display for the load weighing system
had to stay turned-on and since there were other people and machines present at the
proving ground, we also decided to turn on the rear camera monitor for increased safety.
4.2 Apparatus and sensor placements

Our study also included another part, not further presented here, in which the hypothesis
was tested that the operator’s use of the controls available (gas pedal, lift lever and tilt
lever) is indicative of a working machine’s operability and that a measure for compara-
tive operability can thus be derived based on the operator’s control commands. That
study required essentially the same type of data and equipment (and operators) as the
one presented in this article, which led to the decision to run both studies concurrently.
Figure 5. Apparatus for measuring and recording of the psychophysiological data
During all sessions various data were recorded off the wheel loader’s CAN bus, onto
which data from externally mounted sensors were also merged (thus providing signals
of interest that were not readily available in a standard machine). Vector Informatik’s
CANcaseXL equipment and CANalyzer software were used for the recording of all
traffic on the machine’s CAN bus to a laptop placed inside the cab. The need for par-
ticular signals required in some cases modification of the software for engine and
transmission control, while in other cases an external ECU (Parker IQAN MDL) was
used to place calculated signals on the CAN bus. The latter was also used to acquire
additional data from externally mounted sensors for lift and tilt cylinder stroke and lever
234 Paper VII
position for lift and tilt. A modified LoadTronic load weighing system from AADI was
used to measure the net weight of the bucket. The data were transmitted to the machine
controller via SAE J1587 protocol and then merged into the data stream on the CAN
bus.
The psychophysiological measurements were conducted using the cStress software
from PBM Stressmedicine Systems, running on a second laptop installed in the cab of
the wheel loader. Finger temperature, heart rate (via Electrocardiogram, ECG) and skin
conductance were acquired using a C2 physiological monitoring system from J&J En-
gineering and a LifeSense AirPas oxycapnograph handled data acquisition and pre-
analysis of the operator’s respiration (Figure 5).
Figure 6. Right hand controls and sensor placement on right wrist
Figure 7. Left hand controls and sensor placement on left wrist and fingers
The ECG sensors were placed on both wrists (see Figure 6 and Figure 7) using iso-
tonic gel. We knew from earlier measurements that these sensors need to be secured
against relative movement in order to avoid misreadings. Since the operator in a wheel
loader cab is exposed to vibrations and accelerations of various frequencies and magni-
tudes, potentially rendering ECG measurements impossible to perform in this environ-
ment, we were glad to find that good data could be acquired when using the suspended
chair’s integrated arm rests to support the operator’s lower arms and asking the operator
to control the machine’s steering angle and forward/reverse gear not through the steer-
ing wheel and main gear shift lever but by using the controls on the left side of the chair
instead (see Figure 7).
Figure 7 also shows the placement of the other sensors. The operator’s finger tem-
perature was acquired by a thermistor taped to the top of the left little finger (the tape
did not restrict the blood flow). The sensors for measuring galvanic skin response were
attached to the inner phalanges of the ring and middle fingers of the left hand, again
using isotonic gel. The operator’s respiration was sampled by a nasal sample line, in-
serted some millimetres into the left nostril; in connection with this the operator was
asked to exclusively breathe through the nose.
All tests were also recorded on video using an externally placed digital video camera
and later synchronised with all other acquired data. Synchronising video to CAN data
was possible by using stationary points in the recorded data for bucket height and wheel
speed. Synchronisation of machine data (recorded in CANalyzer) and psycho-
physiological data (recorded in cStress) proved a little more challenging. In the end we
chose an offline semi-manual approach: at the beginning and end of each test recording
we would press two buttons simultaneously, where activation of the first button was
logged on the CAN bus and the other logged as an event in cStress. These synchronisa-
tion markers were later used to automatically merge both data streams.
A simple kitchen timer counted backwards and sounded an alarm when the required
operating time was reached. Both the timer and all measuring equipment were placed so
that the operator was unable to look at the displays.
4.3 Working cycle

With the intriguing facts visualised in Figure 2, Figure 3, and Figure 4 in mind, we spe-
cifically wanted to study the wheel loader operator during the bucket filling phase. We
therefore chose to modify the short loading cycle such that the operator was instructed
to still go through most of the motions, but not to use steering and thus empty the
bucket at the same spot where it was filled. Since the material used was sorted gravel
which does not stick together like for example clay, and is thus fairly easy to handle,
this procedure did not introduce any skewing of the test results as far as the bucket fill-
ing phase is concerned.
Not using steering also solved another problem that would have had an impact on the
measurement results. As mentioned before, the attachment of the ECG sensors required
the operator to use the joystick for steering, instead of the steering wheel. However,
inexperienced operators are unused to this, and since the use of the joystick for steering
236 Paper VII
requires some practice, forcing them to do so would have introduced additional work-
load, thus potentially skewing the test results. The decision not to use the steering func-
tion at all solved this dilemma.
4.4 Psychophysiological Stress Profile

Every human being has an individual response to workload, which means that some sort
of reference or calibration is needed in order to be able to correctly evaluate the results
of psychophysiological measurements. Establishing a Psychophysiological Stress Pro-
file (PSP) has proven to be valuable in clinical work with patients with stress-related
dysfunctions.
The PSP shown in Table 1 is essentially taken from [7] and has been implemented in
the cStress software. It contains 15 minutes of data recording, guiding the patient (or in
our case the wheel loader operator) through six steps.
Table 1. Psychophysiological Stress Profile (PSP)
# Designation Observation Description

time
1 Base line 3 min Relaxed silent reading of a neutral text
2 Deep breathing 2 min Deep breathing under guidance, approx. 6 bpm
3 Nonverbal stress 2+2 min Two periods of thinking about a stressful situation,
feedback and guidance in-between
5 Math stress 2 min Counting aloud backwards from 2500 in steps of 7
In step 1 we provided the operator with a neutral text to read, informing him that this
was not an exam and while we asked him to read carefully and try to understand the
text, he was not required to read all of it or even remember anything. This put some
workload on the operator, but far from enough to induce stress. (In our case the text
handed out was a newspaper article about the Nobel Prize for Literature and whether
receiving it means the end of an author’s career – it is fair to state that this text was neu-
tral as none of the operators got overly excited by reading it).
During step 2 the person was guided to breathe slowly and deeply (using the stom-
ach) with about 6 breaths per minute (with small individual deviations), and by this
means attained a very relaxed state.
Step 3 is called “Verbal stress” in [7] and according to that paper in the first half of
this step the person is to be instructed to relate some stressful event that they have ex-
perienced. During the second half, the person shall then, instead of talking, just think
about the same negative and stressful event. However, in our case the test leader is a
close colleague and personal friend of all of the operators, who in turn were not patients
with psychological problems to be treated. Therefore, modifying this step into a phase
of non-verbal stress seemed more appropriate. At the beginning of this step we therefore
asked the operator to think of a negative and stressful event, real or imaginary, and keep
thinking of it for two minutes. After this time the test leader approached the person
again, gave feedback and asked him to continue for two more minutes, but this time to
also breathe irregularly as if was stressed.
In the following phase, step 4, the operator was asked to close his eyes and just relax
thoroughly, using individual techniques without any other requirement e.g. on breath-
ing. The results show how quickly the person recovers from stress.
In step 5 the person was tested for reaction to directly induced stress, coming from
the requirement to count aloud backwards from 2500 in steps of 7. To increase the pres-
sure the operator was told that it would be noted how many subtractions he would man-
age to perform during the two minutes of testing and that each mistake would be noted,
announced and required to correct before being allowed to continue. (It was interesting
to see that this indeed induced an enormous amount of stress in some individuals, while
hardly bothering others.)
Step 6, the last step within the PSP, is a repetition of step 4, i.e. relaxing without any
other requirement.
After each step the operator was asked to fill in a questionnaire (Figure 8).
Self-evaluation
How do you evaluate your own performance, did you cope?

(Put an X somewhere on the line.)
It did not OK, It went

go well at all as required very well
I felt OK, I felt

very stressed no major problems no stress at all
Do your hands feel warmer or colder than before – or the same?

(Circle the answer)
Estimate your finger temperature: ________°C
Figure 8. Questionnaire during PSP
4.5 Participating operators

In all, eighteen people, all male, agreed to participate in this study; all of them close
colleagues of the author. Their data have been recorded and handled in a way that pre-
vents drawing conclusions as to their identities. In all documents, as in this article, an
238 Paper VII
operator is referred to by the day of his test driving, from “04” (tested on October 4th,
2010) to “02” (tested on November 2nd, 2010).
The people asked to participate as test operators in this study are all male, mostly in
order to exclude possible skewing due to gender differences in the way the human body
reacts to mental workload. The potential operators were not randomly chosen, but pre-
selected according to the author’s preliminary judgement of their wheel loader operat-
ing skills. However before starting the test each operator was asked to make a self-
report with respect to their skill, using a visual analogue scale without values, but with
some helpful guidance, see Figure 9 left.
One operator commented that the last two skill statements should change places, so
that “professional” is the highest ranking. The author’s intention with the scale as
shown in Figure 9 was to beat the innate modesty of most people, which leads to the
phenomenon that almost nobody ever chooses 100% (or close) as a skill level. Also,
there might be several people who have machine testing as their profession, yet they
know of others whom they perceive to be even better at operating a wheel loader.
To the right, Figure 9 shows the division of the scale into four skill groups and how
the individual operators are placed in these groups, using their self-reported skill ranks.
These turned out to be similar to the author’s preliminary judgement, resulting in nine
operators in skill group 1 (inexperienced), six in skill group 2 (experienced) and three in
skill group 3 (expert). This also gives the possibility to combine groups 2 and 3 at a
later stage into one “experienced and above” group that matches the “inexperienced”
group in size, which would benefit the statistical strength of direct comparisons.
In our pre-selection of potential test operators for this study we chose not to focus on
expert operators alone, but to include operators of all skill groups in order to test
whether there is any difference in how operators in these groups handle and evaluate the
different machine variants. However, we deliberately avoided skill group 0 (newcom-
ers) in order to guarantee at least a theoretical knowledge of how to operate a wheel
loader, and in order to exclude learning effects as much as possible. The ten minutes of
self-training before each live testing session would not have sufficed for newcomers,
but were deemed sufficient for operators in skill group 1 and above.
4.6 Design of the study and procedure

Each operator was given an exclusive 2.5 hours session after lunch, starting with the
Psychophysiological Stress Profile (PSP) described earlier. For this, we took the opera-
tor of the day to a secluded place (as far as this is possible in the triangle of the R&D
workshop’s prototype parking area, the proving ground and a motorway nearby). The
PSP was conducted with the operator sitting in working position in the cab of the wheel
loader and the person administering the test standing on the ladder, instructing through
the open door, and monitoring through the closed door when recording was in progress.
After the PSP, the operator took the machine to the proving ground, where a smaller
area was reserved for this study. The testing of all three machine variants was then per-
formed, with the traction force setting as the independent variable. The two limited set-
tings of 47% and 62% were tested against the unmodified software version with 100%
maximum traction force, in random order, unknown to the operator. The operators were
not told what had been modified and how, but most of them, especially the more ex-
perienced ones, were able to deduce this fairly quickly.
In order to minimise the skewing influence of learning effects, before testing a ma-
chine version each operator was given ten minutes’ self-training just to familiarise him-
self with the specific characteristics of the current machine version. After the training
followed five minutes’ live test-driving, where the operator was asked to make sure to
fill the bucket completely and use the machine at a normal production tempo, resulting
in a cycle duration of approx. 25 seconds (both of which most of the inexperienced op-
erators did not manage to achieve completely, but at least this demand created a certain
pressure). The session ended with a subjective evaluation where, among other things,
the operator was asked to judge the tested machine version’s ease of bucket filling and
perceived power on a visual analogue scale (without printed values). The operator was
also asked to rank his overall impression of the machine in comparison to all previously
tested variants on a nominal scale (“better”, “worse” or “same”). See Figure 10 for the
questionnaire used after testing of the third machine variant.
240 Paper VII
Evaluation of machine variant #3
How do you rank the machine in this setup,

especially during bucket filling? (Put an X somewhere on the line.)
Very hard OK, Very easy

to fill the bucket about normal to fill the bucket
The machine feels OK, The machine feels

powerless, slow about normal powerful, quick
Does the machine feel better or worse than the first one – or the same?
(Circle the answer)
Does the machine feel better or worse than the second one – or the same?
(Circle the answer)
How do you evaluate your own performance, did you cope?

(Put an X somewhere on the line.)
It did not OK, It went

go well at all as required very well
I felt OK, I felt

very stressed no major problems no stress at all
Do your hands feel warmer or colder than before – or the same?

(Circle the answer)
Estimate your finger temperature: ________°C
Figure 10. Questionnaire, to be filled in after testing of each machine variant
Since the study was performed during the autumn, comparable conditions for the op-
erators were ensured by performing tests only on days with good weather (no rain) and
at the time of the highest ambient temperature, at 13:00 local time. The latter also guar-
anteed that the operators were rested and had eaten recently.
5 Analysis
5.1 General setup
The measurements conducted during this study resulted in several gigabytes of data,
making at least semi-automated analysis necessary.
The main parts of the calculations were executed in MathCad worksheets that
merged data originating from CANalyzer (machine data) and cStress (psychophysio-
logical data) by using the aforementioned synchronisation markers embedded in both
data streams. A function then automatically tagged each bucket filling phase within
each loading cycle and exported part of the data stream back to cStress for expert re-
view (see next section). This result of this review was a consensus figure for each op-
erator’s absolute workload level in each condition. These data and the results from the
various questionnaires each operator had to fill in were then imported into MathCad and
merged with all other recorded data for further analysis.
5.2 Expert review of psychophysiological data

As mentioned in the introduction, our longer-term goal is the development of an opera-
tor assistance system that is adaptive to the human operator’s workload. For this we
need to develop online methods to detect critical workload situations automatically,
robustly and in time. Our shorter-term goal is to support product development, in which
the operability of prototype machines is traditionally evaluated subjectively by asking
the operator to compare the tested machine with a base line and rate it according to a
specific scale. Here, too, a measure of operator workload is required, ideally calculated
automatically either during or immediately after testing.
However, we are aware of the magnitude of these goals. There is a quite reliable rela-
tionship between finger temperature and mental workload in that increased workload
leads to decreased finger temperature and vice versa [8]. Such relationships can also be
found other signals, e.g. heart rate, heart rate variability, respiratory rate etc [9][10][11].
However, these data have to be evaluated in context and in relation to the PSP and each
other, which today requires expert knowledge. Furthermore, a reliable algorithm to
automatically remove measurement artefacts in data for ECG and respiration has yet to
be developed.
It is precisely this problem of analysing the output of interacting and compensating
systems in context that driving the development of Artificially Intelligence systems for
such tasks. The ongoing research reported in [12] and [8] is promising, but the results
can not yet easily be implemented into a system as described above. Instead, we simu-
lated the output of such an automatic system: three of us (Olsson, von Schéele, and
Ohlsson) used their expertise in this field to manually review and analyse the measure-
ment data and derive an absolute workload index for each operator in each of the condi-
tions (PSP and three machine variants), without knowledge of the operator’s skill level,
his subjective evaluation, and which machine variant (setting of traction force limita-
tion) corresponded to which condition for this particular operator. After each of us ana-
242 Paper VII
lysed the data individually, we compared our results and agreed on a consensus work-
load index for each operator in each condition.
Our manual analysis uses all of the recorded data (not just from the bucket filling se-
quences) with the following criteria applied. The ratings ranged between 1 and 5 in
every time period. The highest ratings (5) should be applied if all the criteria listed be-
low showed a very functional pattern well within known reference values for each
physiological signal. The lowest rating (1) should be applied when the workload seem
very high in all signals. If a physiological signal contained many measurement errors,
this signal was not used in the rating of that period.
The physiological signals were rated as described below and in the priority order
listed:
1. Level of HR, RR, EtCO2 and peripheral temperature. A HR below 80, RR be-
low 20, EtCO2 above 4.5% and peripheral temperature above 25ºC indicates
low workload.
2. Noticeable recovery mainly in HR, RR and SCL between bucket fillings and
during relaxation in the PSP indicates a more effort-economic behaviour and
therefore low workload.
3. A clear connection between breathing and heart rate variations, so called Respi-
ratory Sinus Arrhythmia (RSA), indicates low workload.
4. Extreme reactivity or extreme rigidity in mainly HR and SCL indicates high
workload
Signs of habituation in any system, with priority to HR by bucket fillings, indicate a

more effort-economic behaviour and therefore low workload.
5.3 Statistical analysis

A brief summary of the setup of our study reveals the possibility for both sub-studies
with between subjects design and sub-studies with within subjects design:
- Eighteen operators participated.
- Each operator can be sorted into one of three skill groups according to his
self-reported skill level.
- If only two skill groups are to be used then there is the same number of op-
erators in skill group 1 and the combined skill group 2+3
- Each operator was tested under similar conditions.
- There was no particular order between the operators; they were booked ac-
cording to availability.
- Each operator successively tested three conditions, which consisted of dif-
ferent software settings of the otherwise unchanged machine.
- The order of the tested machine versions was randomised.
- Each machine version was first tested for ten to twelve minutes to mini-
mise learning effects, followed by five to six minutes of live test-driving.
- Five to six minutes of live testing resulted in at least six bucket fillings, of-
ten more, depending on operator.
- As learning effects are excluded by the ten minutes of self-training, the cy-
cles are in no particular order for, i.e. the measurements taken can be seen
as random.
- A normal distribution can be assumed due to good repeatability for the
same combination of machine variant and operator.
One interesting aspect to study was how operator workload, time on task (i.e. bucket
filling time), task completion (i.e. bucket load) and fuel consumption vary for different
operators, but also for different skill groups (a typical between subjects design).
However, this study was mainly concerned with exploring the hypothesis that the op-
erator workload index, derived by experts in consensus, is a measure that correlates
with the operators’ subjective workload assessments (or any of the objective machine
performance measures). For this, we regard each operator as a sub-group and do not
compare between operators, but only between machine variants test-driven by the same
operator. This is a within subjects design with repeated measurements.
The question is how meaningful it would be to conduct common statistical analysis.
The problem lies with the workload index: for the within subjects part there is too little
data for simple t-tests, because there are only three workload values per operator, one
per condition (discounting the PSP), aggregating all psychophysiological measurement
data per machine variant. For the between subjects part there would be sufficient data
for a t-test, but the relevance of its results is questionable.
A t-test provides an answer to the question of whether data from two samples share
the same mean. If this is not the case then these data can be considered different. This
works well for direct and discrete observations, e.g. sampled data of finger temperature.
But the workload index is an indirect observation, aggregated from continuous meas-
urements of five different data streams (heart rate, finger temperature, galvanic skin
response, respiration rate, end-tidal CO2 concentration in exhaled air), derived in con-
sensus by experts, who also had to clean the data from measuring artefacts and then
view the cleaned data streams in reference to each other and data from the operator’s
individual PSP. The workload index as a value is just too far removed from the raw
measurement data to make simple statistical analysis like a t-test meaningful.
The same value of workload for two different individuals does not mean that both
necessarily had the same heart rate, finger temperature (etc.) in the same condition. It all
depends on context, i.e. of all five data streams in relation to each other (also in time),
and on the individuals’ respective PSP results. The same heart rate, finger temperature
(etc.) can therefore easily lead to different workload index values, depending on the
PSP.
We have therefore not conducted any t-test for the results presented.
244 Paper VII
6 Results
6.1 Workload index, within subjects
Table 2 shows the expert’s consensus on the operators’ workload during the PSP and
the three test conditions. To the right this table, sorted by operator skill level, also
shows in which individual order the machine variants with their respective traction
force setting were tested by the operators.
With the exception of one, all operators show a higher workload during the Psycho-
physiological Stress Profile than during the first machine test. Comparing the different
conditions, it shows that the workload does not vary much within each subject, despite
the operators clearly commenting on experiencing the machine variants differently, as
shown in Table 3.
No clear overall trend emerges when comparing the operators’ answers in the ques-
tionnaire (see Figure 10) on how well they felt they had managed (“personal satisfac-
tion”) and how stress-free they felt during the operation with the experts’ consensus on
the operators’ workload. This is examined in more detail in Table 4 and Table 5.
These tables show for how many operators (overall and divided into skill groups) a
strict monotone relationship (positive, constant, negative or none at all) can be found
between workload and respective measure. In each table, the right side shows the pic-
ture if a tolerance of ±0.1 workload points was applied, so that for example the se-
quence 3.8 – 3.9 – 3.8 could be considered a constant monotone relationship rather than
none at all. Applying this tolerance, it can be seen that half of the operators are consid-
ered to show no change in workload.
Apart from this, no clear relationship between personal satisfaction and workload can
be established (Table 4). Without the (arbitrarily chosen) tolerance, a relative majority
of the operators would not even show any monotone relationship, be it positive, nega-
tive or constant.
Table 4. Operators showing a monotone relationship btw. personal satisfaction and workload
Table 5. Operators showing a monotone relationship btw. stress-free operation and workload
246 Paper VII
There seems to be some (rather weak) support in Table 5 for a negative relationship
between workload and the operators’ responses to whether they operated stress-free, i.e.
the higher the value for “stress-free operation” the lower the workload, which is what
one would expect. Without applying the tolerance, no such relationship can be claimed.
Table 6 shows the operators’ responses related to the tested machines, evaluated for
their ease of bucket filling, overall impression of power and comparative rank. Here
again, no clear trend is immediately obvious.
Table 6. Results: questionnaire responses on subjective machine evaluations
Even with applying the same (arbitrarily chosen) tolerance of ±0.1 workload points,
Table 7 shows as much support for the expected negative relationship between machine
rank and workload (i.e. the better a machine has been subjectively ranked, the lower the
workload) as for the opposite. The same can be stated for ease of bucket filling (Table
8) and machine power (Table 9).
Table 7. Operators showing a monotone relationship btw. machine rank and workload
Table 8. Operators showing a monotone relationship btw. ease of bucket filling and workload
Table 9. Operators showing a monotone relationship btw. machine power and workload
Not wanting to rely on the operators’ responses alone, Table 10 tests for the hypothe-
sis that the operators are unaccustomed to assessing machine properties via visual-
analogue scales and that a relationship exists between workload and how much a ma-
chine variant’s traction force was limited. However, no support for this could be found.
Table 10. Operators showing a monotone relation btw. traction force limitation and workload
Finally, in Table 11 it is examined if the operators simply reacted to the test se-
quence, i.e. that their workload over time either increased due to fatigue or decreased
due to trainings effects. Without applying the tolerance, there is some weak support for
fatigue, predominant among the inexperienced operators in skill group 1. With the tol-
erance applied as much support can be found for a positive relationship as for a negative
one.
Table 11. Operators showing a monotone relationship btw. test sequence and workload
248 Paper VII
In conclusion, the workload index, derived from the psychophysiological measure-

ments, mostly shows no variation with any of the independent or other dependent vari-
ables. Possible explanations will be discussed in the next chapter.
6.2 Workload index and performance, within subjects

With all recorded data available, it is tempting to look for similarities in the perform-
ance of the human operator (i.e. the workload) and the performance of the machine he
controls. Most important for a working machine in general are cycle productivity (e.g.
expressed in ton/hour) and cycle efficiency (e.g. expressed in ton/unit of energy) –
shown in Table 12 in normalised form. In the specific case of our study, average bucket
load, bucket filling time and energy required during bucket filling are also of interest
(normalised in Table 13).
Again, no clear trend can be spotted. For several operators the workload index is ap-
proximately constant even though the machine performance data vary (e.g. operators #4
and #21 to #28). For others the relative difference in performance between the machine
variants is small, yet the workload index varies (e.g. operators #8 and #2).
Table 12. Average machine performance data related to the working cycle (normalised)
Table 13. Average machine performance data related to bucket filling (normalised)
As described previously, the test leader instructed the operators to keep their pace
and try to fill the bucket completely each time and in each condition. Here we have con-
flicting requirements since the “47%” machine variant with severely limited traction
force simply does not permit a bucket to be filled as fast and as much in comparison
with the unmodified “100%” machine. Some operators chose to keep the bucket fill
factor constant, which led them to spend more time at the gravel pile. Others chose to
keep the pace with the consequence of less material in the bucket. This trade-off is visu-
alised in the first column of Figure 11 and Figure 12. The fourth column in each figure
shows the ratio of bucket load and bucket fill time, i.e. bucket fill productivity. How-
ever, for an entrepreneur total cycle productivity is important (the ratio of bucket load
and cycle time), shown in column six of the figure.
250 Paper VII
Figure 11. Operator workload over machine performance (mean values) for skill group 1
Figure 12. Operator workload over machine performance (mean values) for skill groups 2 and 3
252 Paper VII
It is peculiar that the four operators with the highest skill levels consistently have a
high workload for all machines, not affected by traction force limitation. This is pre-
sumably a ceiling effect. It is also interesting to see that while operators #21, #22 and
#25 accept that the machines with limited traction force settings do not easily permit as
high bucket loads, operator #28 works hard to achieve this – and in turn sacrifices pro-
ductivity by spending too much time on filling the bucket, especially when operating
the severely limited “47%” machine. We also see a slightly higher workload in this
condition.
Apart from these observations on an individual level it is hard to see any general
trend. In addition to productivity we have also more closely examined the relationship
of the operators’ workload to the fuel consumption and energy efficiency of both bucket
filling and the complete cycles (using the data from Table 12 and Table 13) – again, no
clear trend emerged.
6.3 Workload index and performance, between subjects

With the data presented in the previous section we can compare the workload index of
all operators together with their performance and their individual skill level.
Figure 13 shows all values for operator workload and all mean values for bucket
load, bucket fill time, bucket fill productivity, cycle time and cycle productivity for all
three measured conditions (traction force settings) by operator skill level. In the dia-
grams on the left side the lowest values per operator are marked (the other values
marked on the vertical line belong to the two conditions that did not lead to the overall
lowest mean value). Note that the vertical lines do not show standard deviation, but
simply connect the three values per operator. In the diagrams on the right in Figure 13
the highest values per operator are marked. Note that there is no correspondence be-
tween diagrams, i.e. the marked lowest values for bucket load, for example, do not cor-
respond to the marked lowest values for bucket fill time or operator workload (for such
visualisation see Figure 11 and Figure 12).
Analysing the diagrams in Figure 13 it appears that, except for a few individuals, the
lowest workload of an operator generally seems to increase with skill (top left diagram),
which was unexpected, because it means that the inexperienced operators in general
have a lower workload. These individuals also show a lower performance in compari-
son, but one hypothesis was that inexperienced operators would be stressed by too many
other factors (one example being noticeably reduced stability of the machine with a
loaded bucket in the top position) and would thus have a very high workload in all con-
ditions. Clearly, the data we gathered do not support this hypothesis.
Looking at highest workload per operator, the aforementioned ceiling effect around a
workload index of 4.0 is apparent. The expert operators in skill group 3 showed high
workload already in all conditions, while the experienced individuals belonging to skill
group 2 seem more able to relax, while being able to perform almost as well as the ex-
perts.
Figure 13. Lowest and highest values for workload and performance (mean values) by skill levels
254 Paper VII
Another, but expected general trend can be seen for the mean value of the bucket
load, both lowest and highest (Figure 13, second diagrams from top, left and right side),
which seems to increase with skill. Again, note that there is no correspondence between
the diagrams. From the analysis in the previous section we know that there is no general
trend between bucket load and workload on an individual level (within subjects), even
though both seem to increase with skill (between subjects).
As had been expected, the greatest variations in performance can be seen for the in-
experienced operators in skill group 1. It was, however, unexpected to find that the ex-
pert operators in group 3 focussed so intensely on filling the bucket even with the trac-
tion force being severely limited, to the extent that they spent comparatively more time
doing so and thus sacrificed productivity.
It is interesting to note that none of the operators were aware of their high workload.
In their opinion, the work performed was little demanding, especially in comparison to
other possible tasks such as loading of shot rock.
6.4 Subjective machine evaluations

ants, especially in light of their operating skill. Table 14 shows the same results previ-
ously presented in Table 6, but sorted according to traction force setting.
Table 14. Subjective machine evaluations sorted according to traction force setting
With this data Table 15 shows that the operators in group 3, i.e. the experts, share the
clear opinion that the original machine (100% traction force) is the best, the wheel
loader limited to a maximum traction force of 62% next best, and the extremely limited
machine worst. This opinion is predominant even in skill group 2, the experienced op-
erators, but two had already ranked the 62% variant as the same as or better than the
100% machine.
Table 15. Results of the operators’ machine ranking, skill groups 1, 2, 3, and combined 2+3

100% 62% 47% 100% 62% 47% 100% 62% 47%
Ranking
Ranking
Ranking
2+3 Machine variant

Ranking
2 6
Worst 9 2+3 9
The operators in skill group 1, the inexperienced operators, also follow the same vot-
ing pattern to a certain degree, but deviate more often. Two operators in this group ac-
tually ranked the 47% machine as best, and one ranked the 100% variant as worst.
In this context it is of interest to analyse whether a relationship exists between the
for ease of bucket filling. Table 16 shows that such a relationship is given most of the
time, but is somewhat less distinct for operators in skill group 1.
Table 16. Comparison of operators’ assessment of ease of bucket filling vs. machine ranking

Ease of bf
Ease of bf
Ease of bf
Highest 8 2 Highest 6 Highest 3

Lowest 1 1 7 Lowest 1 5 Lowest 3
2+3 Machine rank

Ease of bf
Highest 9 1 9 of ease of bucket filling for ranked machines.

2 6
Lowest 1 8 2+3 9
The same analysis made for machine power, also evaluated by the operators using a
visual analogue scale, is shown in Table 17. The same picture emerges as before; the
expert operators in skill group 3 are most distinct in their judgements, the experienced
operators in group 2 are quite distinct with the exception of one, while the inexperi-
enced operators in skill group 1 show more deviations from the expected outcome of the
best ranked machine also having the highest score for machine power.
From the spontaneous comments noted during the study, we know that several opera-
tors were unsure how to rank a machine variant. Several, especially inexperienced op-
erators seem to correlate machine ranking with their impression of the machine’s
power, even though they gave a lower score for ease of bucket filling, presumably
blaming themselves for not being able to utilise the power available. One operator
commented on this and reasoned that bucket filling is an important phase in a loading
256 Paper VII
cycle and that he will therefore give one particular machine variant a lower rank due to
him finding it more difficult to fill the bucket, even though the machine received the
highest score for power.
Table 17. Comparison of operators’ assessment of machine power vs. machine ranking

M. power
M. power
M. power
Highest 7 1 1 Highest 6 Highest 3

Lowest 1 1 7 Lowest 6 Lowest 3
2+3 Machine rank

M. power
Highest 9 1 9 of machine power for ranked machines.

2 6
Lowest 9 2+3 9
Three operators in skill group 1 also made the interesting comment that the machine
with the least power, the 47% variant, seemed to be more powerful than the other vari-
ants immediately after finishing bucket filling, i.e. when backing away from the gravel
pile. Analysing the test data, it can indeed be seen that both machine speed and engine
speed ramp up faster for the 47% machine than for the others. But we can also see that
in these cases the operator activates the gas pedal sooner, while activating the lift lever
later than for the other machines. In some cases the operator never releases the gas
pedal, which is understandable given that the traction force of the 47% variant is so low
that wheel spin and total cancellation of lifting force are virtually impossible, thus
eliminating the need to modulate engine speed. It is plausible that the operator of this
machine learns to be less sensitive with the gas pedal during bucket filling and therefore
also uses it more aggressively immediately afterwards. Furthermore, with less traction
force available the bucket load is usually also lowest in the 47% machine, thus requiring
less engine torque to lift, and therefore having more torque available to accelerate the
engine and the whole machine.
7 Discussion of possible sources of error

While performing the measurements, as well as during analysis and reporting, we sig-
nificantly improved our knowledge with regard to appropriateness of equipment and
procedures. This will be discussed in the following.
7.1 Procedure
We were able to secure a specific machine and a dedicated test area for our study, thus
ensuring that all operators were operating under the same conditions.
During the testing phase we tried to keep the environmental conditions as similar as
possible. We only tested on days with good weather (no rain) and at the time of the
highest ambient temperature, i.e. immediately after lunch (which also ensured that the
operators were rested). But since this study was performed in mid-autumn, temperatures
were generally lower; during the testing around 5-10°C. The machine was parked out-
side overnight and had therefore cooled quite substantially by the morning. Together
with 20-30 minutes’ engine warm-up each day (about two hours before testing) and a
1.5km drive to the testing site, the first ten minutes of self-training should provide
enough waste heat to warm up the machine (engine, transmission, axles, and hydrau-
lics), so that testing conditions were similar for all cycles, independent of order.
The significance of a Psychophysiological Stress Profile is to see the individual dy-
namic variation, as well as record the base line and review the subject’s ability to relax.
Before testing at the proving ground, we took the operator of the day to a secluded place
to establish the PSP. While this location was isolated visually, it was not so audibly,
since it was still within hearing distance of the proving ground, a nearby motorway and
the R&D workshop’s busy prototype parking area (with the backing alarm of a machine
driving in reverse being a particularly disruptive sound). This may have influenced the
PSP measurement data. A possible improvement may also be to establish a PSP in a
non-work place environment for comparison.
At the proving ground we had a dedicated area, but were at times surrounded by
other machines using the same facilities. This may have distracted the operator and gen-
erated higher workload.
During testing, the wheel loader’s automatic functions were fixed at gear-dependent
BSS (Boom Suspension System), automatic APS (Automatic Power Shift), and with
Auto Kick-down enabled. Of course, the BSS is only deactivated when in 1st gear,
which means it is activated during the first seconds of the bucket filling phase, until the
transmission automatically shifts from 2nd to 1st gear. This is done via an adaptive algo-
rithm, which might differ between operators. Perhaps one of the pre-defined fixed APS
settings L (low power), M (medium power) or H (high power) should have been used
instead. Or perhaps the automatic kick-down function might have better been disabled
altogether, in order to give the operator the freedom to choose when to shift gear.
The operators were asked to use the machine for three times 10 + 5 minutes, in total
45 minutes. There is a risk of fatigue, especially for the less experienced operators, but
we can at best only find weak support for this in the workload index data.
Furthermore, in spite of the ten minutes of self-training before each live testing ses-
sion, a learning effect might still be present, especially for the less experienced opera-
tors.
As mentioned earlier, it might be the case that some of the inexperienced operators
were also unaccustomed to using visual analogue scales. Their memory of the other
tested machine variants might not have been correct and they could thus have skewed
their own evaluation results. Though operators were allowed to look at their previously
assessments for reference, hardly anyone asked for this opportunity.
258 Paper VII
7.2 Measuring equipment and data acquisition

The measuring equipment used to acquire the psychophysiological data was originally
developed for clinical applications, i.e. use in a stationary, stable and clean environ-
ment. Using it, as we have done, in a mobile environment, exposed to large and fre-
quent accelerations might have introduced measurement errors other than what is dis-
cussed below.
As explained earlier, the operators were instructed to lay their lower arms on the arm
rests attached to the suspended chair. This dramatically lowered the amount of meas-
urement artefacts in the ECG signal, yet some still occurred. and had to be corrected
manually in the analysis phase.
Many measurement artefacts were found in the respiratory signal, which was ac-
quired with a nasal sample line, inserted some millimetres into the left nostril and se-
cured by a piece of tape. However, in a few cases the operator’s facial hair prevented
good adhesion, leading to a temporary detachment of the cannula and thus giving erro-
neous readings. For some operators we therefore used a nasal cannula that could be se-
cured via the operator’s ears. With only three such nasal cannulas available, we had to
decide in which cases to employ this alternative method.
During both the PSP session and machine operation the operator was asked to exclu-
sively breathe through the nose. However, we suspect that some of the artefacts in the
respiratory data are due to the operator unintentionally breathing through the mouth
(which itself is a sign of high workload and stress, but hard to clearly identify in retro-
spect).
The PSP was conducted with the operator sitting in working position in the cab of the
wheel loader and the person administering the test standing on the ladder, instructing
him through the open door, and monitoring through the closed door when recording was
in progress. With the colder ambient temperatures, yet still strong sunshine, this led to a
cooling of the cab’s interior while instructing the operator, yet gradual warm-up during
execution of the PSP steps – both of which had an impact on at least the finger tempera-
ture measurements. In some cases operators complained about low temperature and
wanted to turn on the heating and air conditioning system. But since this would have
required a running engine, which might have distracted the operator from the required
task, the decision was taken to only do so during the instruction phases before each PSP
step.
Furthermore, we discovered that the finger temperature readings were affected by
heat influx from the sun and by the flow of cold/warm air from the cabin’s heating and
air conditioning system, both during the PSP sessions and at the proving ground. Some
operators even deliberately directed air flow to their hands to cool/warm them during
operation of the machine. This discovery discredited the finger temperature readings
enough for us to decide to completely disregard them when determining the workload
index.
With regard to the above discussed we were still able to obtain satisfactory data from
all subjects to estimate their psychophysiological functioning based on ocular analysis.
7.3 General considerations

There are several general considerations to make with regard to this study. To begin
with, the workload index we derived is a compound measure, that does not only reflect
the necessary mental effort of the operator to handle different traction force settings of a
wheel loader. With the lack of distinct results at hand, it is thus probable that other ef-
fects had a larger influence on the psychophysiological data we acquired.
Operators themselves commented on distracting elements, such as poor gear shifting
quality of the automatic kick-down function, other machines nearby which required one
to look out when driving in reverse, or looking straight into the sun when turning the
head in order to check that nobody was standing behind the machine (even though we
specifically instructed him to check the rear-view camera monitor). These are normal
for a working place, but it might be that the impact of the different traction force set-
tings was too small to be registered in measurements with durations of a mere five to six
minutes. Next time we should consider aiming for 30 minutes or more.
Related to this the question arises of when to consider a workload to be different, e.g.
whether or not a difference of 0.1 in value is to be considered significant, and how this
relates to testing time.
In the future we need to consider measuring other data known to be indicative of
workload. The quality of our measurements of finger temperature was too poor to use
the data acquired. But as finger temperature is normally an important measure of SNS,
we should try to vary the sensor placement for increased reliability and validity. Al-
though heart rate and heart rate variability can change within seconds, reliable meas-
urements are obtained for periods with a minimum of 30-40 seconds (i.e. four times
longer than a typical bucket filling sequence), and a maximum of 5 minutes. It is
warned that sensitivity decreases for shorter or longer periods [10].
Respiratory Sinus Arrhythmia [13] can be analyzed dynamically and is the most so-
phisticated way to analyse ANS. This is very complex and at the moment there are no
automatic analysis methods available, but we included RSA in our manual estimation of
the subjects’ workload index, based on our understanding of the problem.
Finally, we need to reassess the working cycle. It could be that the workload index
was impacted so little by even the severe traction force limitation in the “47%” machine
because there were no immediate consequences of falling behind in pace or bucket load.
In reality, for instance at a saw mill or an asphalt processing plant, trucks would be ar-
riving continuously to be filled by the wheel loader. A decrease in productivity is
clearly visible in a constantly lengthening line of trucks waiting their turn. Furthermore,
when truck and wheel loader are properly matched it takes four buckets to fill a truck.
Again, the productivity decrease is clearly visible if an operator needs a fifth run. In our
study the operators would have to face none of these stress-increasing consequences.
We need to consider either more closely representing the situation at real working
places or introducing consequences by artificially adding some productivity-related
feedback by visual or acoustic means.
260 Paper VII
There was also the ambivalence of our instructions to keep “a normal cycle time of
20-25 seconds” (again, without consequences for not doing so) and the conflicting re-
quirements of both keeping pace and completely filling the bucket each time. As dis-
cussed previously, each operator handled this trade-off individually.

At the beginning of this article we described our vision of using psychophysiological
measurements to assess operator workload in sufficiently detail to use it as a comple-
ment to traditional subjective evaluations, but also use such measurements in a work-
load-adaptive operator assistance system in a longer perspective. Even though we were
able to make some interesting observations at both individual and skill group level, the
results of this study indicate that measures other than ones based on heart rate, finger
temperature, skin conductance and respiratory air flow should be considered. All of
these parameters seem hard to measure reliably in the tough environment of a mobile
working machine, but will be less prone to artefacts in the controlled environment of a
human-in-the-loop simulator. This of course requires the use of a simulator that has
both the fidelity required to produce correct results on the machine side and the immer-
siveness required to produce correct results on the side of the operator. The research
summarised in [14] shows that this is possible in principle.
In a simulator as well as in reality it is important to provide relevant test scenarios.
We suspect that the operators in our study did not experience enough pressure, because
there were no immediate consequences for not keeping the set cycle pace or not filling
the bucket completely. In the future we will try to add this pressure by either more
closely representing a real working place or by adding productivity-related feedback
artificially (e.g. by visual or acoustic means).
We also found that the sensitivity of the psychophysiological measurements con-
ducted appears to be too low for our requirements. On the other hand, promising re-
search is been done on the use of Artificial Intelligence techniques (such as Case-Based
Reasoning) to evaluate data from psychophysiological measurements. With proper in-
put, perhaps such systems can provide the sensitivity we require and might then also be
able to recognise measurement artefacts. The need to automatically analyse all acquired
data in context is the major driver for such AI-based analyses, since a manual review of
multiple channels with data from interacting and compensating systems can today only
be performed by experts, relying on clinical data which can not very easily be trans-
formed into complex work situations.
We and others, for example [15], also work with the development of new wireless
sensors that can be deployed without distracting the operator from the working task,
while at the same time being able to acquire data without the amount of measuring arte-
facts we have to cope with today.
Acknowledgments
The authors gratefully acknowledge the support received from the Energy & Environ-
ment programme within the Swedish Vehicle-Strategic Research and Innovation pro-
gramme (FFI), administered by the Swedish Energy Agency (Energimyndigheten) in
this research.
References
[2] Wilson, G. F., Russell, C. A. (2007) “Performance Enhancement in an Uninhab-
ited Air Vehicle Task Using Psychophysiologically Determined Adaptive Aid-
ing”. Human Factors, vol. 49, no. 6, pp 1005-1018.
http://dx.doi.org/10.1518/001872007X249875
[3] Ley, R. (1999) “The modification of breathing behavior. Pavlovian and operant
control in emotion and cognition”. Behavior Modification. Vol. 23(3), pp 441-479.
http://dx.doi.org/10.1177/0145445599233006
[4] Schleifer, L. M., Ley, R., Spalding, T. W. (2002). “A hyperventilation theory of
job stress and musculoskeletal disorders”. American Journal of Industrial Medi-
cine. Vol. 41(5), pp 420-432.
http://dx.doi.org/10.1002/ajim.10061
[5] Von Schéele, B. H. C. and von Schéele, I. A. M. (1999) “The Measurement of
Respiratory and Metabolic Parameters of Patients and Controls Before and After
Incremental Exercise on Bicycle: Supporting the Effort Syndrome Hypothesis?”.
Applied Psychophysiology and Biofeedback. Vol. 24(3), pp 167-177.
http://dx.doi.org/10.1023/A:1023484513455
[6] Volvo CE (2009) “Product brochure L60F, L70F, L90F”. Volvo Construction
Equipment.
http://www.volvoce.com/constructionequipment/europe/en-
gb/products/wheelloaders/wheelloaders/L70F/Pages/specifications.aspx
[7] Begum, S., Ahmed M. U., Funk, P., Xiong, N. and von Schéele, B. (2006) “Using
Calibration and Fuzzification of Cases for Improved Diagnosis and Treatment of
Stress”. Proceedings of the 8th European Conference on Case-Based Reasoning,
pp 113-122.
262 Paper VII
[8] Begum, S. (2011) “A Personalised Case-Based Stress Diagnosis System Using

Physiological Sensor Signals”. Doctoral thesis, Mälardalen University, Västerås,
Sweden.
[9] De Waard, D. (1996) “The Measurement of Driver’s Mental Workload”. Doctoral
thesis, University of Groningen, The Netherlands.
son, K. (2003) “GARTEUR Handbook of Mental Workload Measurement”, Final
Report for Flight Mechanics Group of Responsibles (FM GoR) of the Group for
Aeronautical Research and Technology in Europe (GARTEUR) by GARTEUR
Action Group FM AG13.
[11] Park, B. (2009) “Psychophysiology as a Tool for HCI Research: Promises and
Pitfalls”. Proceedings of HCII 2009: New Trends, LNCS 5610, pp 141-148.
http://dx.doi.org/10.1007/978-3-642-02574-7_16
[12] Ahmed, M. U. (2010) “A Case-Based Multi-Modal Clinical System for Stress
Management”. Licentiate thesis, Mälardalen University, Västerås, Sweden.
[13] Von Schéele, B. H. C. (1995) “Respiratory Sinus Arrhythmia as a Diagnostic
Tool, a Treatment Goal, a Rule for Treatment Planning, a Cool Biofeedback, and
an Out-Come Variable”. Biofeedback and Self Regulation. Vol. 20(3), pp 308-309.
Sweden.
http://dx.doi.org/10.1007/978-3-642-02812-0_36

263
"We'd run out of ironic things to say."
(Neil Tennant, The Pet Shop Boys)
View publication stats

V1 1 Complete

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

V1 1 Complete

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Quantifying Operability of Working Machines

Thesis · November 2011

Optimal Working Cycle View project

The user has requested enhancement of the downloaded file.

DIVISION OF FLUID AND MECHATRONIC SYSTEMS

“Quantifying Operability of Working Machines”

Linköping Studies in Science and Technology. Dissertations, No. 1390

Printed by: LiU-Tryck, Linköping

Only a few people in this world were lucky enough

(from “Blue Mars” by Kim Stanley Robinson)

To my parents Inge & Peter

There is one way to understand another culture.

(Agent Brown in the film "The Matrix")

I ARBETSMASKINER spelar föraren en avgörande roll för maskinens prestanda. Såväl

(from the film “American Beauty”)

[VIII] Filla, R. (2003) “Anläggningsmaskiner: Hydrauliksystem i multidomäna

1 Introduction and research background 1

8 Conclusions and outlook 87

W ORKING MACHINES in construction, mining, agriculture, and forestry have re-

Figure 1. Construction tools, past and present

1.1 Industrial objectives

1.2 Academic need

1.3 Research questions and hypotheses

1.4 Scope and delimitations

Figure 2. Research area of this thesis

1.5 Research approach

1.6 Thesis outline

WHEEL LOADERS are complex working machines. The interconnections of sub-

Figure 3. Wheel loader [4]

- Productivity (expressed for example in ton/hour)

Varying with the working task, other important parameters are:

2.2 Working cycles

2.2.1 Short loading cycle

Figure 4. Short loading cycle

Table 1. Phases of the short loading cycle

2.2.2 Load & carry cycle

Figure 5. Load & carry cycle

Table 2. Phases of the load & carry cycle

2.3 System architecture

2.4 Operability aspects

(added control systems)

Figure 7. Machine harmony in a working machine and operator workload

Secondary harmony Secondary harmony

Machine harmony Primary harmony Primary harmony

- Bucket filling (phase #1)

2.4.3 General operability aspects in short loading cycles

Another ingredient of machine harmony is the availability of power. Here, of course,

Figure 14. Fuel consumption during short loading cycle

Figure 15. Operator control commands in a calm short loading cycle

2.4.4 Operability aspects in bucket filling

Figure 17. Force balance during bucket filling (simplified)

Figure 19. Reduction of lift speed due to prioritisation

2.4.5 Operability aspects in reversing

Figure 20. Expected effect of the automatic reversing feature

Figure 21. Actual effect of the automatic reversing feature

2.4.6 Operability aspects of bucket emptying

- Tilt lever in order to dump the material

Stiffer torque converter, Reduced slip Increased

Figure 22. Expected effect of a stiffer torque converter

However, same pump

Higher traction force Increased cycle

Figure 23. Actual effect of a very stiff torque converter