Post on 31-Dec-2019
Aircraft and Rotorcraft Pilot Couplings – Tools and Techniques for
Alleviation and Detection
ACPO-GA-2010-266073
Deliverable No. D2.3
State-of-the-art pilot model for RPC
prediction report
Contractual delivery date:
March/2011
Actual delivery date:
April/2011
Partner responsible for the Deliverable: TUD
Author(s):
Deniz YILMAZ (TUD),
Michael JUMP (UoL), Lu LINGHAI (UoL), Michael JONES (UoL)
Dissemination level
PU Public X
PP Restricted to other programme participants (including the Commission Services)
RE Restricted to a group specified by the consortium (including the Commission
Services)
CO Confidential, only for members of the consortium (including the Commission
Services)
ACPO-GA-2010-266073 Deliverable D2.3
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Document Information Table
Grant agreement no. ACPO-GA-2010-266073
Project full title ARISTOTEL – Aircraft and Rotorcraft Pilot
Couplings – Tools and Techniques for Alleviation
and Detection
Deliverable number D2.3
Deliverable title State-of-the-art pilot model for RPC
prediction report
Nature R1
Dissemination level PU2
Version 1.0
Work package number WP2
Work package leader ONERA
Partner responsible for Deliverable TUD
Reviewer(s) Michael Jump, UoL
M.D. Pavel, TUD
The research leading to these results has received funding from the European Community's
Seventh Framework Programme (FP7/2007-2013) under grant agreement no 266073.
The author is solely responsible for its content, it does not represent the opinion of the
European Community and the Community is not responsible for any use that might be made
of data appearing therein.
1 R=Report, P=Prototype, D=Demonstrator, O=Other
2 PU=Public, PP=Restricted to other programme participants (including the Commission Services),
RE=Restricted to a group specified by the consortium (including the Commission Services), CO=Confidential,
only for members of the consortium (including the Commission Services)
ACPO-GA-2010-266073 Deliverable D2.3
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Revision Table
Version Date Modified Page/Section Author Comments
1.0 11.04.2011 Deniz Yilmaz
Executive Summary
This document describes a review of pilot modelling techniques. A summary of the
literature survey is presented to indicate up-to-date status of pilot modelling
techniques and their applications for Aircraft Rotorcraft Pilot Couplings (ARPCs)
phenomena. Recent innovative pilot modelling techniques are introduced and
interpretations of possible adaptation of these new models, as well the present
models, into the ARPCs research of ARISTOTEL are discussed.
ACPO-GA-2010-266073 Deliverable D2.3
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Table of Contents
Document Information Table ................................................................................................. 2
Revision Table ...................................................................................................................... 3
Executive Summary .............................................................................................................. 3
1 Introduction .................................................................................................................. 5
2 Up-to-date Pilot Modelling ............................................................................................ 6
2.1 Human Operator Behaviour 6
2.2 Pilot Modelling Techniques 7
2.2.1 Human Sensory Models ........................................................................................... 7
2.2.2 Physiological Models ............................................................................................... 9
Biodynamical Models 9
Neuromuscular System 10
2.2.3 Control Theoretic Models ........................................................................................10
2.2.3.1.1 Crossover Model 11
2.2.3.1.2 Precision Model 12 2.2.3.1.3 Quasi-Linear Model 12
2.2.3.1.4 Hess structure model 13
2.2.3.1.5 Hossman model 14
2.2.3.1.6 Modified Hess model 15
2.2.3.1.7 Optimal Control Models 16
3 Pilot Models based on soft-computing techniques .......................................................17
3.1 Hybrid Neural and Fuzzy Pilot Models 17
3.2 OCM Extensions 19
3.2.1 Modified Optimum Control Model (MOCM) .............................................................19
3.2.2 Revised Optimum Control Model (ROCM) ..............................................................20
3.2.3 Variable Strategy Pilot Model ..................................................................................21
3.2.4 Kalman Filter Models ..............................................................................................22
3.2.5 Sensitivity Pilot Model (H∞ and H2) ..........................................................................23
3.3 Gray‟s BAT Model 23
4 Application of Pilot Models into PIO/PAO studies ........................................................24
4.1.1 Categorisation of APCs Based on Different Theories ..............................................24
4.1.2 Criteria for Workload Build-Up Flight Test Technique .............................................25
4.1.2.1. Pilot BAT Tracking Performance 26
4.1.2.2. Pilot inceptor Workload - Pilot Duty Cycle and Aggressiveness 26
4.1.3 Application of Optical Tau for APC Prediction .........................................................28 5 Discussion and future plan ..........................................................................................29
6 References ..................................................................................................................31
7 List of Abbreviations ....................................................................................................35
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 5 of 35
1 Introduction
Work Package 2 (WP2) is related to rotorcraft and pilot modelling, and merging both
modelling aspects for A/RPCs assessments. As a part of the WP2, deliverable 2.3 is
dedicated to review of state-of-the-art pilot modelling techniques and discussing their
suitability for ARISTOTEL project in means of A/RPCs studies.
A detailed review of the up-to-date pilot modelling methods is presented in section 2.
Starting from the closed loop behaviour types of a human operator, a review of pilot
modelling techniques is presented. These methodologies are categorized within three main
headings as modelling the human sensory, physiological and control theory originated
systems. Each system is investigated with presentations of the corresponding models
involved.
Innovative recent pilot modelling techniques are presented in sections 2 and 3.
The traditional view of how pilots perform a wide range of flying tasks involves an initial acquisition, followed by point tracking (PT) of aircraft flight path or attitude. Having acquired the desired flight path or vehicle attitude, the pilot tries to maintain it at some fixed value. One category of Aircraft Pilot Couplings (APCs), traditionally called pilot induced oscillations (PIOs), result from an increased pilot task gain during the tracking phase [Ref. 1-4]. Situations when pilots may operate aircraft within attitude and flight-path constraints using high feedback gain include air-to-air refuelling, formation flying, target tracking and operations in confined areas. However, Gray[5] noted that there are times when pilots deviate from this classical PT behaviour and adopt a strategy whereby they monitor and avoid defined boundaries (e.g. such as when trying to avoid ground impact while, at the same time, preventing low altitude departure from controlled flight). He proposed a piloting strategy to explain a class of PIO that can occur in these situations, termed Boundary Avoidance Tracking (BAT) PIOs. When PIOs occur in these situations, they are distinct from classical PT PIOs, in that they are triggered by the pilot's need to manage the aircraft approaching potentially opposing boundaries. Current understanding and knowledge about PIOs are considered inadequate to explain these BAT control strategies.
[Ref. 6] has shown the close relationship between the BAT concept and optical tau. Optical
tau was introduced by Lee[7] as a development of Gibson‟s optical flow theory of visual
perception [Ref. 8]. The development of tau theory is based on the premise that purposeful
actions are accomplished through coupling the motion with either external or internal sources
– the so-called motion guides [Ref. 7, 9-11]. Motivated by its basis in visual perception, tau
has been applied to flight control and handling qualities [Refs 1,12-15], with the hypothesis
that, in terms of a visual guidance strategy, the overall pilot‟s goal is to overlay the perceived
optical flow-field over the required flight trajectory; the pilot then works directly with optical
variables to achieve prospective control of the aircraft‟s future trajectory. In it, the application
of optical tau is extended to Boundary Avoidance Tracking using a „roll-step‟ maneuver [Ref.
16], developed to evaluate lateral-directional handing qualities of rotorcraft, as an extension
to the ADS-33 mission task element family [Ref. 17].
Final section is a summary of discussions on the present pilot models and new candidate
models. This section is responsible for pointing out the main features of the reviewed models
and their possible applications into ARISTOTEL.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 6 of 35
2 Up-to-date Pilot Modelling
2.1 Human Operator Behaviour
Variations of human operator control strategies compose mainly three categories of
behaviours according to closed loop manual control systems; compensatory, pursuit and
precognitive pilot behaviours [Ref. 34].
Compensatory behaviour
Compensatory behaviour is characteristically present
when the pilot command input signals (desired state of
the aircraft) is random-appearing and when the only
information the pilot receives is system errors or aircraft
outputs. Under full-attention conditions the pilot exerts
continuous closed-loop control on the aircraft so as to
minimize system errors in the presence of commands
and disturbances.
Figure 1 presents a compensatory display. Only the
error is displayed. The system state is displayed by the
solid dot; the command input signal is represented by
the line. The difference is visible as error (e).
Pursuit behaviour
When the pilot command input signal can be distinguished from the system outputs by virtue
of the display or by preview, pursuit behaviour is possible. From experiments it is known that
human operator performance during pursuit behaviour improves with respect to the
compensatory tracking behaviour [Ref. 34].
Figure 2 presents a pursuit display. Here the command signal (c) and the system state (y)
are visible and both have the same reference. The error is the difference between y and c.
Figure 3 presents a preview display. It is basically the same as a pursuit display, but here the
near future for the command input signal is also visible.
Precognitive behaviour
When complete familiarity with the controlled element dynamics and the entire perceptual
field is achieved, the highly-skilled human pilot can, under certain conditions, generate
neuromuscular commands which are properly timed, scaled and sequenced so as to result in
machine outputs which are almost exactly as desired.
A special case of precognitive behaviour is synchronous behaviour. This means that when
the command input signal is sinusoidal the pilot can, after intermediate adaptation (which can
include pursuit behaviour), duplicate the sinusoid without phase lag. The pilot dynamics
could be modelled as a pure gain for synchronous behaviour. McRuer [Ref. 34]:
Figure 1: Compensatory display
Figure 2: Pursuit display Figure 3: Preview display
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 7 of 35
“Synchronous behaviour is, perhaps, the most important type of pilot action for large
amplitude severe PIO‟s.” [Ref. 35] gives an overview of such pilot models relevant for
PIO/PAO simulation.
2.2 Pilot Modelling Techniques
Pilot modelling techniques can be split into three main areas of interest: Human Sensory
modelling, physiological modelling and control theoretic modelling. Although these
branches are complicated within their own research areas, a comprehensive overview of the
evolvement of pilot modelling is important to understand the state-of-the-art pilot modelling
aspects. An overall block diagram representation of pilot-vehicle system is shown in Figure 4.
Figure 4: Block diagram representing the pilot-vehicle-system under manual control [36].
2.2.1 Human Sensory Models
Human sensory mechanism receives aircraft states visually through flight displays and
natural sensory organs. Although this mechanism represent a highly sophisticated sensor
network in conjunction with central nervous system, it is more suitable for moderate angular
rotations of short durations, whereas most flight envelopes are consist of low intensity and
long duration rotations, which may end up with a diverged perception. One of the most
important parameter within human sensory modeling is Spatial Disorientation (SD), which
defines the failure of the correct perception of position, motion and attitude of the aircraft
within a fixed coordinate system defined by Earth fixed coordinate systems. SD is being
investigated within three categories: [Ref. 36]
Type 1: where the pilot is unaware that the perceived orientation is wrong
Type 2: where there is a conscious awareness of disorientation due to conflict
of what is sensed and what is present on the displays
Type 3: where high level of confusion about the actual orientation arises with
growing levels of stress and inability to maintain the adequate control activity
Spatial Disorientation simulations should inherently include three sub-models; visual,
vestibular and proprioceptive systems. These signals are merged and interpreted in the
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 8 of 35
central nervous system, thus the recognized final orientation cues are the result of these
interpretations.
The Visual System
The visual cue system is one of the crucial information source to maintain orientation in Earth
fixed space. Typically, the visual system is divided into two mode: ambient and focal.
Ambient mode refers to low frequency visual transmission to central nervous system, e.g.
effects of slant on pilot altitude judgements. On the other hand, the focal mode of vision is
concentrated on central field of vision, particularly for high spatial frequencies with high levels
of consciousness. A good example of this mode is the pilot perception activity during an
instrument flight. The visual cues that are perceived by the pilot can be modelled at various
levels, an example visual cue model of Hess [Ref. 35] is shown in Figure 5.
Figure 5: Block diagram representing visual cue perception model proposed by Hess [35].
During fixed wing aircraft studies, it is shown that given a perfect internal model, the visual
cues provides more information than motion cues in means of pilot orientation perception,
especially for judging altitude, flight path angle and speed [Ref. 37].
The Vestibular System
The Vestibular System is located within the inner ear and contains the olotihs and the semi-
circular canals, both functioning as an internal reference systems. The olohits are
responsible for sensing tilt and specific forces, whereas the semi-circular canals are about
perceiving angular accelerations. Correct modelling of this whole system is an essential
phase of pilot perception modelling, especially for degraded visual environments. The fluid
,endolymph, in the inner ear semi-circular canal provides the head and body angular
perception cues and the dynamics of this fluid inside the canal, is a remarkable source of SD.
As for an A/RPC case, the wrong sensation of pitch during a coordinated turn, may provide a
divergent input from the pilot to correct the situation, and in return the pilot would be
tightening the turn and increasing the rate of decent.
The Proprioceptive System
The proprioceptive systems, known as the somatosensory systems, is responsible of
providing feedback of aircraft/rotorcraft motion through the forces or displacements felt via
the control devices. The principle of this system is to determine body position and spatial
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 9 of 35
limb movements according to signals received from sensors ( such as muscle spindles)
within the muscle structure. The famous phrase “ seat-of-the-pants sensation” of the pilots is
a result of this perception system. The complexity of properly combining proprioceptive
feedbacks with other sources of cueing inputs lies in the crucial fact that “somatosensory
systems has evolved to give a sense of relative body part motion and not a sense of
orientation in space” [Ref. 36]. Moreover, most studies treat proprioceptive systems as an
comprehensive sense that results from various sensory organs, unlike an unitary sensory
information source like visual system.
The structural model of Hess includes a transfer function responsible for representing
proprioceptive sense, which is a candidate to trigger A/RPCs. The methodology of the
structural model is present in further sections.
2.2.2 Physiological Models
Physiological models are composed of biodynamical and neuromuscular models, which
compromise the representation of involuntary pilot input due to aircraft/rotorcraft acceleration
motion. There are two model approaches to physiological area; Biodynamical and
Neuromuscular models.
Biodynamical Models
The fact that pilots of aircraft can be subjected to accelerating or vibrating environments that
may adversely affect their performance has led to the formulation of biodynamic models of
human pilot behaviour. In these models, one can investigate the effects of an accelerating or
vibrating environment on pilot control capabilities. Depending on the representation of the
spine, there are three categories of biodynamical models; continuum, discrete and lumped-
parameter [Ref. 36]. Briefly, continuum models take the spine into account as a flexible
beam, of which parameters are tuned via comparison with empirical data. Discrete model
treats spine as a set of connected rigid bodies via springs and dampers. Finally, lumped-
parameter is the approach for modelling the body dynamics by developing an equivalent
mass-spring-damper system. To illustrate an example of discrete biodynamical model, Figure
6 shows the biodynamical model of Kohler and Hohne for roll ratchet investigation [Ref.41] .
Figure 6: Discrete biomechanical models used by Kohler and Hohne [41]
ARISTOTEL has an extensive workpackage (WP4) regarding to biodynamical modelling and
corresponding simulator tests. Detailed descriptions and modelling methodologies are
present in deliverable documents of WP4.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 10 of 35
Neuromuscular System
Neuromuscular models are the representation of both extrafusal and intrafusal muscle
activity with regards to body motions. Extrafusal muscles are responsible for generation of
force and they make up most of the muscle structure. Intrafusal muscles are scattered
throughout the muscle via spindles and they deliver the degree of feedback information to
the central nervous system. McRuer and Magdelano used the ensemble model to represent
the dynamics of the neuromuscular system [Ref. 42], as shown by the block diagram in
Figure 7.
Figure 7: Proposed neuromuscular mode of Magdaleno et al. [42]
In this neuromuscular model, muscle/manipulator block dynamics is modelled as the
following transfer function to fit the data obtained from experiments [Ref. 42]
(1)
For the spindle feedback block, a transfer function is used with delayed equalisation ability.
(2)
The third block, joint sensor feedback block, is the summation of golgi tendon organ
feedback and various modes of other feedback that are difficult to isolate.
(3)
Even though such a basic representation is practically efficient, there is a loss of fidelity,
more precisely when considering A/RPCs researches [Ref. 36]. So far, there has not been
any published study to emphasize the level of importance of neuromuscular model fidelity on
PIO/PAO related practices.
2.2.3 Control Theoretic Models
Several models have been developed with various approaches to model the pilot in the
aircraft/rotorcraft. During 1970s, McRuer [34] introduced the essence of the crossover model,
which became a baseline for several other approaches. Following sections are dedicated to
briefly present the pilot models developed by using classical and modern control theories.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 11 of 35
2.2.3.1.1 Crossover Model
McRuer stated that the human operator adjusts his/her control action in such a trend that
pilot-vehicle open loop dynamics can be described by the following simple transfer function:
(4)
where YP, YC, ωc and e represent the pilot transfer function, vehicle transfer function,
crossover frequency and effective time delay respectively. The crossover frequency depends
on the characteristics of the controlled vehicle and effective time delay used within the
model, which is the time taken for the perception and initial action of pilots body. The idea of
the model is that the human pilot will adjust to different aircraft/rotorcraft dynamics such that
final representation have the same human plus vehicle dynamics, namely in the predicted
area of the crossover. When the model is implemented in a closed loop system, the pilot
perception for vehicle output becomes important since there are the options of compensatory
or pursuit cases. This closed loop block diagrams and corresponding transfer function
representations are shown in Figure 8.
Figure 8: Closed loop Pilot-Vehicle system possibilities with compensatory and pursuit cases [34]
Some extensions of cross over models are written as [64].
(5)
where;
(6)
where H(s) is the equalization transfer function of initial crossover model with L and I are
lead time constant and lag time constant respectively, KP is the static pilot gain. P(s) is the
system transfer function with additional time delay and neuromuscular lag time constant
n. For example [Ref. 66] used the same pilot model for PIO susceptibility study with multiple
control effectors, combining with Neal-Smith time domain criterion.
Also, [Ref. 67] used the same crossover pilot model for developing a generic pilot-aircraft
model to be used in simulations.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 12 of 35
2.2.3.1.2 Precision Model
To capture the low frequency phase droop, Hess [Ref. 35] introduced a precision model,
which is an higher order transfer function with isomorphic characteristics .
(7)
Even though the resultant model is noticeably close to experiment data, see Figure 9, it must
be considered that tuning of the parameters and system representation highly depend on the
experiment conditions. Therefore, the resultant model represents the specific task and
configuration instead of a comprehensive pilot model.
Figure 9: A frequency response of the precision model compared with experiment data
obtained from a laboratory tracking task[43]
2.2.3.1.3 Quasi-Linear Model
After the introduction of McRuer‟s crossover model, previous manual control models of
similar studies [Ref. 44] are modified to contain the cross over principles and a new quasi-
linear pilot model is developed, as shown in Figure 10.
Figure 10: Quasi-linear pilot model. [34]
The equalisation parameters of the following quasi-linear model representation of McRuer
are adjusted such that open loop system behaves as an integrator around the crossover
frequency and the gains are tuned to capture a good feedback control system of the closed
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 13 of 35
loop characteristics. The important difference comes from the remnant function, which
represent the nonlinear component of pilot behaviour.
(8)
These models tend to be restricted into SISO (single input single output) analysis because of
the complexity of closed loop parameter tunings of MIMO (multi input multi output) systems.
Thus, recent applications of such models are mainly on matching and validation of
experiment data, instead of predictive purposes as introduced at the first development
stages.
2.2.3.1.4 Hess structure model
This model can be expounded as a combination of McRuer‟s crossover law philosophy and
at the same time some ideas of optimal control pilot model techniques. Initial model was
suggested by Smith [Ref. 45] with the theory that a measure of pilot opinion rating is the rate
at which nerve impulses arrive at a point within the central nervous system to be processed
[Ref. 38]. Initial model of Hess aimed to expand the optimum control human pilot model with
higher order dynamics with an additional time delay. This model was introduced to
accomplish handling qualities assessment for PIO studies [Ref. 51]. Hess improved this
model with a proprioceptive feedback and converted into a new structural model, which is
shown in Figure 11.
Figure 11: Compensatory structural model of the human pilot. Hess [35]
The additional part that Hess implemented to the initial model, which is the proprioceptive
sensory model, is presented as (YPF) [Ref. 38] ;
(9)
The proprioceptive sensory in the structural model is the part that reflects the optimal control
aspects, in means of pilot‟s internal representation of vehicle dynamics resembles the
Kalman filter estimator. Moreover, this sensory model depends on the adaptability of the pilot
to the altering flight dynamics around the crossover frequency. Even though overall structure
model is capable of modelling the pilot, which is able to maintain main compensation modes
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 14 of 35
that are adequate for the aircraft/rotorcraft, efficiency of the model relies on the
neighbourhood of the crossover frequency .
Vorst[68] reported the usage of a structural pilot model to perform helicopter manoeuvres
defined in ADS-33 [Ref. 17]. It was observed that even by using the gains of the „hover‟ pilot
for aggressive manoeuvres, overall inner-outer loop composition of the rotorcraft-pilot system
performed well. Though, feedforward control support for sharp manoeuvres thought to
present improvement for further researches [Ref. 68].
Also, Weber et al implemented the Hess structural model with a modified version into a
longitudinal precise tracking task scenario [Ref. 69]. In this study, it was concluded that
correlations between the pilot model and the controlled elements were not achieved (hence,
the developer needs the experiment data to adapt the model), but the pilot model was
capable of describing human pilot behaviour for even the nonlinear controlled element [Ref.
69].
2.2.3.1.5 Hossman model
The origin of this model is deemed from the research of Hossman [Ref. 46] on obtaining an
correlation between visual, vestibular stimulation and pilot perception and control. The results
of the research are mainly used for motion based simulator fidelity studies but in the
meantime, various stimulus response experiment results are obtained to relate the
contributions of individual senses towards pilot perception, practically from actual pilot
responses.
Overall Hess structure model is composed of individual sensory models passing through
central nervous system, see Figure 12. One of the important aspect of this model is the
present distinguish between central nervous system and peripheral visual system. Moreover,
transfer function modelling of otolith as an accelerometer with over damped mass-spring
system with a second differential form and modelling of semi-circular canals as over damped
torsion pendulum model with a second order transfer function make the overall model
independent of the scenario. Only the gains in the central nervous system model depend on
the task to be performed [Ref. 46].
Figure 12: Hossman’s proposed human pilot [46]
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 15 of 35
2.2.3.1.6 Modified Hess model
Cardullo combined the techniques used in Hess and Hossman model, and developed a
modified pilot model [Ref. 47]. Cardullo implemented Hossman‟s vestibular model with a
haptic sensory model within Hess structural model and proposed a time based parameter
varying model based on soft computing (e.g. neural network) that drove the structural model
parameters towards agreement with the crossover model“ [Ref. 38]. The overall block
diagram of the modified Hess model is presented in Figure 13.
Another recent approach to modify the Hess Model is presented by study of Efremov et
al.[Ref. 60]. In the study, two pilot behaviour models are presented; a modified structural
model and a composite model with the neural network approach. In this study[60], the
following remarks are highlighted to mention the drawbacks of Hess‟s structural pilot [35] :
The resonant peak of modelled closed loop pilot-aircraft system is lower considerably
in high frequency range
It is assumed that crossover frequency is constant for any dynamics configuration
during the adjustment rules for the choice of parameters. However, experiments
showed that crossover frequency alters for different dynamic configurations [Ref. 61]
The visual block of the Hess‟s model is drastically simplified, as being a gain
coefficient.
Figure 14 illustrates the modified version of the Hess model
Figure 14: Modified Hess model of Efremov et al. [60]
Modifications applied to the original Hess models can be listed as; a more complex model of
visual block, ultimate changes in inner loop and adjustment rules, and including nonlinear
Figure 13: Cardullo’s modified Hess structural model block diagram [47]
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 16 of 35
remnant of the pilot into the model [Ref. 60]. The final modified model introduced a better
match than initial Hess model with experiment data, especially at high frequencies.
2.2.3.1.7 Optimal Control Models
Initial optimal control model (OCM) was developed by Kleinman et al, during 1970s [Ref. 48].
The basis of the model is that a well-trained and motivated pilot behaves in an optimal
manner while remaining properties are subjected to inherent physiological and physical
limitations, which can be summarised as time delay, motor and observation noises. As inherit
in all attempts of modelling the pilot behaviour, the actual internal functional sequence of
brain activity within a pilots head cannot be modelled. Therefore, this model uses a “black
box” model, which refers to explicit representation of the real human pilot outputs with
respect to model outputs.
One of the most crucial aspect of the OCM is the defined cost function within the solution of
optimal pilot gains, as indicated by the following equation;
(10)
On the other hand, the weightings (Q, R and G), which stand for internal model, pilots
control action and control rate, define the validity of the OCM, due to the fact that the
accurate specification refers to pilots control objectives.
Instead of the nonlinear remnant component in quasi-linear model, OCM introduces
observation and motor noises, see Figure 15.
Figure 15: The optimal control pilot model [47]
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 17 of 35
The OCM is based upon the assumption that the well-trained and motivated human controller
behaves optimally in some sense, adjusting the pilot's compensation for a given vehicle and
task, subject to human limitations. The OCM has been widely used and has been validated in
a number of tasks. It has been used to model task performance and to assess flying
qualities, to model human-controller-describing functions, and for both the analysis and
synthesis of manual control loops. In the OCM, the pilot's compensation is modelled by
linearquadratic regulator gains, a Kalman-Bucy filter and a linear predictor [Ref. 19].
3 Pilot Models based on soft-computing techniques
3.1 Hybrid Neural and Fuzzy Pilot Models Neural network techniques are well known for their use to approximate arbitrary nonlinear function as well as their capability for online learning [Ref. 18,19]. Neural network models of human pilot behaviour rely upon the power of neural nets to accurately describe the nonlinear signal processing behaviour of the human pilot. These models are particularly useful in mapping pilot cues into control in tasks for which extensive experimental data is available. However, the neural network approach suffers from the difficulty in interpreting results because of their „black box‟ nature. This limitation may be overcome through a hybrid control structure that combines other control algorithm such as fuzzy logic [Ref. 20]. Fuzzy models are based on fuzzy-set theory that leads to a description of cause and effect
relationships that differ considerably from the control-theoretic constructs that have been
described to this point [Ref. 20]. These models have been used to describe such diverse
human control activity as helicopter piloting and automobile driving [Ref. 22,23] One of the
main advantages of fuzzy-logic is that inference rules that model the rule-based stage of
human behaviour can be developed based on previous experience.
An example of applying a hybrid control structure using both a neural network and fuzzy logic
to build a pilot model is illustrated in Figure 16.
Figure 16: Fuzzy logic adapted to McRuer Crossover and Hess structural model [47]
Another application of Neural Network implementation was performed by Efremov et al. [Ref.
60] and the corresponding pilot model is called “composite pilot model”, as shown in Figure
17.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 18 of 35
(a) Structure of Neural Network Model (b) Final version of pilot Neural Network System
Figure 17: Composite pilot model (a) and (b) of Efremov et al.[60]
The idea behind the composite model[60] is based on the consideration that pilot control
response characteristics related to the specific configuration are generated by use of
experience gained from previous system responses with similar dynamic characteristics.
Therefore, training of the Neural Network with the similar configurations to the examined one
is an important step to model the pilot. In other words, prediction of the composite pilot model
depends on the predefined (“trained by Neural Network”) dynamic configurations, which are
similar to the interested dynamic configuration. Even though the interested configuration is
not included in the set of database, the composite model is capable of predicting the aircraft-
pilot closed loop system characteristics. As shown in Figure 17(a), the time delay neural
network (TDNN) was used while developing the model. Specific effort was given on selection
of TDNN architecture, definitions of model input-outputs, definitions of layers, number of
neurons and type of neuron actuation functions. Experiments were conducted [Ref. 61] to
obtain these parameters. Moreover, a first order filter is introduced to input channel of Neural
Network Model (NNM) , see Figure 17(b), to achieve a better correlation with the experiment
data. In order to train the NNM, a configuration was chosen such that there were close
configurations in HAVE PIO and Neal-Smith databases. The results showed a good
agreement with the simulator experiments. An application of this composite pilot model into a
predictive criteria development for flying qualities and PIOs is presented further sections.
Another application of Fuzzy pilot model was performed in DLR (German Aerospace Centre)
[Ref. 65].A fuzzy adaptive pilot behavioural model is used to perform ILS approaches, which
is known to be one of the high pilot workload demanding phase of a flight envelope. Figure
18 shows the fuzzy pilot model used in the study .
Figure 18: Fuzzy pilot model for ILS Tracking Task [65]
Fuzzy pilot model, which is a cognitive pilot, showed similar controls with the experiment data
in terms of magnitude and trend. But it must be noted that the fuzzy logic parameters are set
according to the data obtained from the “real” human pilots performing an ILS approach.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 19 of 35
Gestwa[65] stated that using neural networks would automate the learning process, whereas
it would also bring the` black box` property of the neural network, making it hard to interpret.
Nevertheless, a combination of fuzzy logic with neural network models; so called neural
fuzzy, was proposed to be an interest of further research , including helicopter pilot modelling
development [Ref. 65].
3.2 OCM Extensions
3.2.1 Modified Optimum Control Model (MOCM)
Figure 19: The conceptual block diagram of modified optimum control model MOCM [52]
A second-order Pade approximation was chosen as approximation to a pure delay over the
pilot‟s frequency range of interest, the delay was placed at the operator‟s output and was
considered a part of the plant dynamics, as shown in Figure 19 This modified version of
optimum control model, which was developed by Davidson [Ref. 52] could be used for
interactive modification of pilot-plant parameters, direct calculation of system and pilot
transfer functions, system transfer function manipulation and determination of system
frequency responses. “The major difference between the OCM and the MOCM is the
replacement of the linear predictor of the OCM by the augmentation of the system dynamics
with the pilot's effective time delay before calculation of pilot control and estimation gains.
This difference allows for the direct calculation of the pilot and system transfer functions in
pole-zero form in the MOCM.” [Ref. 52].
An application of MOCM into fixed wing aircraft, especially for wake vortex interaction
research is presented in [Ref.63]. In order to reduce cost, time and risk during development
and flight test phase and to enlighten pilot-aircraft coupled system behaviour, MCOM was
applied to investigate WVE (Wake Vortex Encounter).
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 20 of 35
Figure 20: MOCM pilot model used in [Ref. 63]
The application of the MCOM pilot model and ILS (Instrumental Landıng System) tracking of
large aircraft experiments revealed the conclusion that “MOCM has PIO tendency in
nonlinear case for low encounter angles” [Ref. 63].
3.2.2 Revised Optimum Control Model (ROCM)
Figure 21: The conceptual block diagram of revised optimum control model ROCM[53]
Although the OCM [48]. and the MOCM[52] had been proved to be a successful and satisfactory model of pilot performance, these models also have problems. Since the MOCM places the delay after the neuromuscular dynamics, the overall pilot structure doesn‟t refer to appropriate physical essence to the real pilot. MCOM implies that the pilot‟s brain sends a signal to the muscles, then the neuromuscular systems limits that signals. Hence, the output is delayed [Ref. 55]. In ROCM the delay is placed after the estimator and gains, it implies a cortical processing delay prior to muscular command determination and delayed output signal is subject to the neuromuscular lag [Ref. 53]. In OCM and MCOM, control rate is used in the optimal control cost function. Although the limitation of the neuromuscular dynamics is not mentioned directly, the control rate is corresponded with limiting the neuromuscular dynamics. The control rate weighting is chosen to achieve desired neuromuscular lag in the OCM and MOCM. Shulltz[55] presented that the proposed pilot structure of OCM and MCOM include the neuromuscular lag to model the pilot‟s desire to limit input control rates. According to main characteristic of the human neuromuscular system [Ref. 56], the neuromuscular dynamics can be defied by the physical limitations of motor neural fiber and skeletal muscle cell. In ROCM, the optimal control cost function adopts the standard optimal Linear Quadratic Gaussian formulation. The neuromuscular lag is defined by the physical limitation of the human rather than cost function ROCM aims to solve the problems of OCM and MCOM about representing the appropriate physical essence of the real human pilot, while still benefiting from coherent advantages of these models. [Ref. 53].
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 21 of 35
3.2.3 Variable Strategy Pilot Model
McRuer‟s crossover model[34] and the optimal control model[48]. can reflect the pursuit tracking
characteristics. Further, both models represent a continuous operator control strategy. On
the other hand, discontinuous operator responses are also observed, particularly when the
control amplitudes are large when compared to nominal flight operations. Therefore, in order
to fill the gap of discontinuous control strategies, “bang-bang” (or relay control) variable
strategy models are described [Ref. 57]. Besides, it must be noted that general strategy of
the manual pilot control is composed of both discontinuous and continuous stages, e.g.
pursuit tracking. The discontinuous model of Xiangju[57].presented a variable strategy for both
control and pilot models by using switching function for varying magnitude of applied control
amplitudes, as presented in Figure 22(b).
Variable strategy model uses the pilot function by adopting the pre-perceive system model
and introduces two loops, namely inner and outer loops, as shown in Figure 22(a).
(a) (b)
Figure 22: Inner/Outer loop of pilot model (a) and pilot model structure for discontinuous
behaviour (b) [54]
In Figure 22, ycmd is the guidance and navigation information, yobs is the observed state, yest is
the estimated and predicted flight state, and y is the actual dynamic response of the aircraft.
The transform function Ypo includes the operator‟s neuromuscular dynamics and the reaction
time delay, which are similar with the McRuer‟s crossover model [34] and the optimal control
model[48].Moreover, the outer loop is the control strategy model, and the switching surface is
the representation of the pilot‟s acquisition strategy [Ref. 54]. Zhenhai[58] proposed the
preview follower theory, which refers outer loop as „the reviewer‟ and the inner loop as „the
follower‟. Practically, during an approach of a large aircraft, the transform function Ypo in
outer loop is used to translate the guidance and navigation commands into aircraft attitude
and power commands. The acquisition strategy of the pilot is implemented at this stage [Ref.
54]. Similarly, the transform function Ypi in inner loop is introduced to translate the attitude
and power commands into stick, throttle, or pedal commands. After arranging inner loop as a
block, Figure 23 represents the behaviour model for small deviations.
Figure 23: Behaviour Model for small deviation [54]
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 22 of 35
The sliding control parameters, which are used in behaviour model (Figure 23 ) are σθ and
σT, pitch angle and throttle function variables, respectively. SX is the switching surface setting
for front side landing strategy [Ref. 59]. It is concluded that, for an application of the variable
strategy model into a landing aircraft scenario, integral character is present on the cut off
frequency, the gain is optimal and the dynamic response of the variable strategy is very
similar with the desired sliding control model [Ref. 54]. The varying control technique for the
continuous and discontinuous patterns of this pilot model makes it innovative in means of
control strategy adaptation. Besides, it must be kept in mind that the approach to the pilot
inner loop model is derived from McRuer‟s crossover methodology.
3.2.4 Kalman Filter Models
Figure 24: Kalman filter adaptation with optimal controller within an augmented plant [49]
Rodney[49] combined the Kalman Filtering and Optimal Control aspects in a pilot model, as
shown in Figure 24 and the results of single-axis compensatory tracking task were compared
with experimental database of McRuer [Ref. 50]. A good correlation was achieved, even with
the existence of nonlinear “remnant” part. During the modelling process, the neuromuscular
characteristics were modelled as a first order lag with a time constant of 0.3 seconds, and
the time delay was 0.1 second. A Gaussian white noise was applied to the neuromuscular
system and measurements. Parameters of this noise were selected to provide a consistent
response, which correlates with experiment data. One of the important aspect of the study
was the “closed eye” tracking performance, which refers to tracking without error signal
provided back to the pilot model. The proposed pilot model with Kalman Filter showed that it
can operate from a “built in” source of information “learned” during previous operations [Ref.
49].Hence, the proposed model indicated that it had a precognitive ability, which is the
estimation of the pilot according to experienced characteristics of the forcing function and
adapting the input according to gained information.
A crucial drawback of this model is that it depends on the experimental data to be used to
match the total operator-man system response and the heuristically variance of forcing
function and noise parameters. Moreover, a level of loss in fidelity exists due to merging the
complicated human sensor system into a first order Pade approximated time delay and a first
order lag representation of the neuromuscular system.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 23 of 35
3.2.5 Sensitivity Pilot Model (H∞ and H2)
Anderson[64] developed a pilot model which compromises sensitivity function shaping control
synthesis formulation for a compensatory tracking task. The sensitivity model benefits from
the simple representation of McRuer‟s crossover frequency model, but in multi-loop control
tasks by using H∞ and H2 control solutions to compute model parameters, and adapting
weighting filters to shape the sensitivity functions of the closed-loop operator controlled
feedback system based upon established characteristics of manual operator systems [Ref.
64]. The core idea behind the development of this model was to compensate the need of
„sequential manual loop closure‟ of crossover pilot model in a state-space solution (like
OCM), by implementing H∞ and H2 synthesis methods to directly control feedback loop
shapes. Hence, the loop shaping formulation has the ability of frequency domain analysis of
the pilot‟s control objectives and limitation, like crossover model.
Manual compensatory tracking and pilot model synthesis block diagrams are shown in
Figure 25.
(a) (b)
Figure 25: Manual compensatory tracking (a) and operator model synthesis diagram (b) [64]
One of the remarkable conclusions of the sensitivity model application was that “suitable
matches to experimental data can be obtained without a specific model of neuromuscular
dynamics as long as first-order roll-off characteristics are retained in the operator model.”
[Ref. 64]. Moreover, it was observed that resonant peaking behaviour experienced during the
compensatory tracking task stems from the pilot time delay, since there was no
neuromuscular dynamics involved in the whole model structure.
3.3 Gray’s BAT Model
Gray developed the BAT model, shown in Figure 26, and provided analysis techniques for predicting the associated boundary-avoidance model parameters [Ref. 5].
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 24 of 35
AIRCRAFT
SATURATION
& DELAY
FEEDBACK
SELECTOR
SATURATION
& DELAYPRIORITY
LOWER
BOUNDARY
GAIN
UPPER
BOUNDARY
GAIN
RATE DISPLACEMENTPOINT TRACKING FEEDBACK
LOOP
BOUNDARY TRACKING FEEDBACK
LOOP
ERROR OUTPUTDESIRED
POSITION
Figure 26 The boundary-avoidance tracking model [5]
The switching between PT and BAT is assumed to be discontinuous. Moreover, the variation of the BAT feedback control gain given by Gray and Warren is hypothesized to increase linearly as the boundary is approached. They recognized that this process is likely to be non-linear in practice, influenced by the complexity of the pilot‟s prospective control, the channels used to sense information, the flight control system and the aircraft aerodynamic characteristics. The pilot may not always apply the maximum input for different BAT events, except perhaps when reaching control saturation. When the pilot perceives that the hazard posed by the impending boundary is reducing, the control input will gradually be reduced to avoid other problems, such as reaching rate limits. Therefore, in reality, the parameters used to configure the BAT model are likely to be „adaptive‟ parameters.
4 Application of Pilot Models into PIO/PAO studies
4.1. New Optical Tau Criteria and Pilot Model for RPC Prediction A planned UoL contribution ARISTOTEL will be the further investigation of both BAT events and more severe BAT PIOs. To begin, some new criteria for BAT PIO prediction, extending recent research at the USAF Test Pilot School (TPS)[Ref. 5], will be and modified. The reason for including these new introduced criteria within this report is to show how operator input information closely connects to BAT pilot model development. The results from the development of pilot models will access these criteria and help to partly explore the fidelity of the new pilot models, through comparison with real-time piloted simulation. Furthermore, other potential project partners may benefit from this early dissemination. In addition, these criteria will be applied during data analysis of tests for both rigid-body and aero-servo-elastic rotorcraft and will be implemented for real-time prediction. Second, the research will use optical tau (or the optical tau BAT pilot model) to try to detect BAT events and furthermore to establish if optical tau provided clues to the incipience of a PIO. Finally, the previous two stages aim to help to build new pilot models that combine the knowledge of Gray's BAT model [Ref. 24-26], the control field, and optical tau [Ref. 5,6,7,27,28]. Summarised proposals are contained in the following sections.
4.1.1 Categorisation of APCs Based on Different Theories
A/RPC events have been categorised into three groups in Table 1, with regard to different viewpoints [Ref. 24]. The table is included to provide theoretical support for the ongoing research, from which any new pilot models in the current research project will partly depend
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 25 of 35
on the perception information. The first column, traditionally and widely used, is the one given by the National Academy of Science [Ref. 1,7,13,14].The second column interprets the APC events in terms of the knowledge of cognitive science [Ref. 4]. The pilot, as a subsystem, plays a vital role in the entire closed-loop system for evaluating Handling Qualities (HQ). Traditional models describing the pilot are only informative analogies, such as a simple gain-plus-delay model [Ref. 29] or complex structure dynamics based on feedback loops [Ref. 17]. The knowledge in cognitive science can bring much to the HQ field, e.g. an aircraft can be considered as a temporarily part of the pilot‟s body [Ref. 35,30]. The third column defines the APC events, borrowing categorizing ideas from the concept of spatial disorientation (SD) [Ref. 31].
Table 1 Categorisation of APCs based on different theories
Category
of APC Classical Cognitive Science Analogy of SD
Cat I
Governed by linear
behaviour of the
pilot and system.
The couplings are
often associated
with high gain and
increased time or
phase delay effects.
APC in the learning
stage with cognitive
control. The immature
development of the
cerebella model makes a
pilot suffering from more
hesitation, higher PIO
susceptibility and more
possible handling faults
Unrecognized APC
(dangerous). The pilot is
unaware of the developing
of an APC by
misinterpreting as a control
problem. The continuous
control efforts may result in
the situation worse.
Cat II
Typically involves
limit cycle
oscillations of the
pilot-vehicle system
due to nonlinear
control elements,
e.g. rate and
position saturation.
APC in the autonomous
control stage with the
fully developed internal
cerebella model. The
limitations of human
motor control apply on
the maximum
performance of
autonomous control.
Recognized APC (not
hazardous). The pilot can
normally recover the
problem by reducing control
effort or back out-of-loop at
the cost of lower handling
performance.
Cat III
Covers severe
APCs, which are
inherently nonlinear
and characterised
by a model or flight
control system(FCS)
transition
APC in the stage of the
internal cerebella model
insufficient for a highly
demanding task or a
nonlinear (rapidly and
unexpectedly) system
transition.
Incapacitating APC. An
APC is recognized but the
pilot may fail to take
appropriate control strategy
to recover back, suffering
from the disconnection
between conscious
perception and reality.
4.1.2 Criteria for Workload Build-Up Flight Test Technique
In the period of Boundary-Avoidance-Tracking (BAT) research, the workload build-up Flight
Test Technique (FTT) was developed at USAF Test Pilot school (TPS) [Ref. 29,32]. The
technique is based on tracking a series of continuous manoeuvres within user-defined
boundaries to characterize both pilot tracking performance and pilot inceptor workload. A
modified version of this FTT has been proposed here for ARISTOTEL.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 26 of 35
4.1.2.1. Pilot BAT Tracking Performance
It is hypothesized that the relationship described in Figure 27 exists, relating desired and
achieved tracking performance to the influence of a defined boundary [Ref. 24,26] .
Figure 27 Illustration of tracking performance variation by imposing
reducing boundary size [24,26]
Figure 27 shows that the achieved tracking performance is theoretically expected to increase
as the desired tracking performance becomes larger. Moreover, as illustrated in Figure 27,
when the desired performance boundaries are relaxed, the pilot can spend more effort on
other tasks (whilst maintaining performance). When the desired performance criteria tighten
(represented by the narrower boundary size), the pilot may meet a critical situation and must
focus entirely on the primary tracking task (single focus in Figure 27). After this situation, the
performance will deteriorate and may produce a BAT PIO event. Within ARISTOTEL, a
number of piloted simulations will be investigated to test the theory presented in Figure 27.
4.1.2.2. Pilot inceptor Workload - Pilot Duty Cycle and
Aggressiveness
The available numerical approaches for measurement of pilot workload are immature [Ref.
25,26]. Among these methods, the most common relies upon the use of subjective rating
scales evaluated by test pilots. The approach illustrated in Figure 28 is used in USAFTPS for
measuring pilot inceptor workload because of its feasibility and simplicity for use with
students.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 27 of 35
Figure 28 Illustration of inceptor workload (duty cycle vs aggressiveness)
Figure 28 describes the workload through the use of two new variables: Duty Cycle and
Aggressiveness. Their definitions and measurements are outlined in the following context.
Moreover, the measurement approaches provided in this report are slightly different from the
original version [Ref. 1,24] and will be explained later.
Duty Cycle is measured as the ratio of pilot action time applying controls to the total manoeuvre time [Ref. 24,29,33]. Sheppard‟s version to formulation this term is cited as follows [Ref. 24,29,33] :
0
1100% ( )
fT
f
DC f t dtT
where
0, ( ) 0
( )
1, ( ) 0
dt
dtf t
dt
dt
(11)
The proposed version for use in ARISTOTEL is defined in the form
2
12 1
1100% ( )
t
tDC f t dt
t t
where 1 20, ( ) and ( )( )
1,
dt K t K
f t dt
(12)
where the term is control input, Tf is the total flight time and t1 and t2 are the start and end time for a certain manoeuvre period. K1 and K2 are user-defined threshold values for determining the time the pilot holds the inceptor nearly motionless. The Duty Cycle definition represented in Eq.12 has two refined features compared with Eq. 11. The first is the addition of an inceptor amplitude constraint. This is to account for the situation where the pilot holds the stick at a large-amplitude position while moving the stick at a slow rate. In the case of the worst possible square-like stop-to-stop BAT PIO [Ref. 33], the pilot applies the stick at maximum effort with a period of inactivity. Without introducing this additional constraint condition, inactivity would be ignored despite the pilot being in a high workload situation. Secondly, the definition in the arbitrary period [t1, t2] provides increased functionality for the
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 28 of 35
criteria (e.g. beneficial to capture the abnormal period in which there is high BAT PIO susceptibility). The refined form can allow Duty Cycle to be used as one of possible method of analysing BAT PIO tendency in real-time. Finally, it is evident that as the value of Duty Cycle increases, pilot inceptor workload is likely to increase.
Gray[24] defined Aggressiveness (a newly proposed title for inceptor power) as the rate of the inceptor movement during a piloted task. It suggests that the faster the pilot is moving the stick, the harder he is working. Aggressiveness is calculated as the root-mean squared per-second average of the inceptor measure (position or force) as given in Eq.13. The extension to this criteria, that will be investigated during ARISTOTEL is presented in Eq.14.
0
( )1100%
max
fT
G
f
tA dt
T
(13)
2
1max max 2 1
1 1100% ( ) ( )
( ) ( )
t
Gt
A t t dtt t
(14)
where the term is the rate of control input. The values of max andmax , included for
normalisation, represent the displacement and rate limits for a given actuator respectively. These values are determined by the performance of an actuator. As a consequence, their product can serve as a good normalisation base for comparing workload between different pilots and tasks across different types of aircraft. In addition, the definition of Aggressiveness given in Eq.14 is different from that presented in Ref. [24] in which only the rate of inceptor is contained. In Eq. 13 purely stick deflection is considered as an „aggressiveness‟ measure. The new proposed criteria accounts for both stick position and moving rate simultaneously in terms of their product (inceptor power). Therefore, it may give a truer representation of pilot workload. The larger percentage indicates the higher workload that is reflected through either strength of applied force (high gain possibly), rate of stick movement (rate limiter triggered possibly) or a combination. Therefore, this situation consists with the region close to the right-top corner in Figure 28 representing the area with maximum possibility of PIO. If the situation is at its worst ‒ both position and rate at the critical level, the extreme point (1, 1) in Figure 28, may be reached with the boundary in Eq. 14 narrow enough.
4.1.3 Application of Optical Tau for APC Prediction
The BAT parameter predictions from tau-theory provide a glimpse of the power of using the
optical variables, rather than trajectory or control variables, to define the propensity to PIOs.
The hypothesised criteria from that [Ref. 6] are summarised in Table 2, where C-PIO refers
to conventional PIOs.
Table 2 and conditions for BAT event and PIOs at the target (edge) crossing
b Values b values BAT and PIO
0.5b
0b BAT event or BAT PIO possible
0b but b + BAT event possible
0b but b − BAT event/PIO unlikely
0.5 1b
0b BAT or C-PIO likely
0b but b + BAT event/PIO likely
0b but b − BAT event likely
1b 0b BAT or C-PIO very likely
0b BAT or C-PIO likely
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 29 of 35
These criteria have brought together optical tau theory and boundary-avoidance tracking
developments in flight control.
4.2. A new criteria development by using the Composite Pilot
model for predicting PIO tendency
The original criteria for prediction of flying qualities and PIO tendency levels was developed
by Efremow et al. [Ref. 62]. Even though the referred Russian document is not found on
open literature, a brief descriptive summary is cited in [Ref. 60] ,such that the proposed
criteria use two fundamental parameters; resonant peak of closed loop system and pilot
workload parameter, which were both measured by experiments with more than 80 HAVE
PIO, Neal Smith and LAHOS configurations. First and second levels of flying qualities and
PIO tendency levels were introduced by comparison of the results with pilot ratings with
comprehensive experiment measurements [Ref. 62].Further, a new aircraft-pilot system was
defined by using “composite pilot model”, which was described in section 3.1. The reference
states that there arises the necessity of changing the boundary levels of the initial criteria. On
the other hand, the new modified criteria demonstrated that they have high probability rates
of predicting PIO and flying qualities levels (0.8 for first level,0.75 for second level and 0.65
for third level), according to pilot rating levels [Ref. 60]. Figure 29 demonstrates the resultant
flying qualities level graph.
Figure 29: Modified criteria for flying qualities and PIO prediction [60]
The development process of the original criteria should be reviewed to understand the
intensive methodology and interpretations of the new modified criteria. During the
ARISTOTEL project, such innovative and specific techniques will be reviewed and updated
via pilot modelling documents.
5 Discussion and future plan
Conventional and alternative pilot modelling techniques are revised in previous sections.
Comparison of the pilot models and their A/RPC related aspects are listed below:
Perception organs contribute to involuntary or false pilot controls, thus amplifying the
A/RPC; e.g. “graveyard spiral” which is the false level flight sensation of pilot while
actually performing a descending turn.
Three common fundamental modelling techniques; namely crossover, optimal and
structural, include a representation of the neuromuscular settings according to their
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 30 of 35
methodologies. Effects of the fidelity of modelling the neuromuscular processing on
A/RPCs are not exactly identified yet.
Quasi linear models have the disadvantage of being incapable of parameter variation
with respect to changes in the task required. The model has no ability to initiate an
A/RPC. Sinusoidal forcing functions of varying frequencies are used to drive the
system towards instability and the phase difference between the inputs and outputs
are used to predict any A/RPC signature.
Hess structure model has ability to model an A/RPC event by modelling the pilot as
regressing into a tracking behaviour where the effective error rate is controlled via no
proprioceptive feedback. Besides, the model is incapable of triggering the A/RPC by
itself. Moreover, the modified Hess has the ability of proprioceptive feedback signal,
but restricted to the existence crossover law.
One of the widest area of Optimal Control Model and its variations are the analysis of
the time delay effects on handling qualities and A/RPC prone identifications.
Hess model, like most of the pilot models, has the restriction of requiring experiments
to adjust inherent parameters, such as the control sensitivity. Any PIO or A/RPC
related boundaries derived by the model reflects the particular setting of the pilot with
the proposed task and gathered settings from the experiment. Hence, any change in
parameters of the pilot will lead to different boundary.
The application of the MCOM pilot model and ILS tracking of large aircraft
experiments revealed the conclusion that “MOCM has PIO-Tendency in nonlinear
case for low encounter angles”[63]. McRuers crossover pilot model provides an easy to
interpret pilot model with minimum parameters to represent a pilot, but yet sufficiently
effective around the crossover frequency and letting user to customize the
parameters by classical control notions[70].
Pilot optimum control model presents a state-space modelling approach with
computational advantages and tracking error minimization, which reflects the natural
connection to operators primary task objective during compensatory tracking tasks.
However, the representation of the system parameters include many components like
quadratic cost function weights, signal-to-noise ratios and noise intensities[64].
In general, software aided pilot models methodize the shaping functions and
parameter sets of the pilot model by various methodologies like fuzzy logic, neural
network, etc. Even though new computing techniques are remarkable tools to modify
the pilot model techniques (structural and behavioural), they all require detailed
selection of parameters of the method chosen. This leads to adjustments according to
experiments.
Most of the pilot modelling approaches and defined pilot models are validated and perfected
according to the conducted „model specific‟ experiments. Therefore, a comprehensive,
generic pilot model has not been introduced, but methodologies prove themselves according
to corresponding experiment data, which provide the pilot model parameter adjustments. To
conclude, the chosen pilot modelling techniques should be coupled with specific
experiments, which should be designed to provide the required parameters defined by the
chosen pilot modelling techniques.
It is observed that McRuer‟s crossover frequency pilot model is commonly adopted and
integrated into several pilot modelling techniques, including state-of-the-art soft computing
models. Therefore, it is planned to use MCruer‟s crossover frequency model for the initial
pilot modelling activity in TU Delft. Further adaptations of the pilot models for RPCs will be
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 31 of 35
decided according to collectively gathered literature examples and corresponding
adaptabilities to ARISTOTEL project.
Tau theory has provided an effective and feasible approach to determining the start and end BAT parameters. In addition to investigating the nonlinearities integral to the BAT model, and its characteristic parameters, the work done in that paper has developed an effective methodology to predict the occurrence of a BAT event. Within ARISTOTEL, research will be continued by UoL as follows:
The first objective of the research is to use optical tau to detect a BAT event by
determining associated BAT timings, and furthermore to establish whether or not
optical tau provides clues to the incipience of a PIO. An extension to this objective is
to explore how the pilot works directly with the available optical information, and to
establishing a relationship between the aircraft motion, control activity and the optical
flow variables in a new model of BAT PIOs.
The second objective is to develop further experiments to investigate and validate the
approach used in [Ref. 6]: to extend it to other aircraft types and manoeuvres, such
as the newly-built Bo105 and Puma rotorcraft models. For example, aircraft more
prone to experience fully developed PIO cases should be investigated, where the
efficiency of early warning systems based on the direct measurement of tau and its
derivatives can be explored.
Thirdly, research will explore the even more attractive prospect of predicting
situations of incipient PIOs ahead of the boundary crossing, based on the time
difference between target and boundary, providing the information needed to create a
PIO alert system.
Finally, Gray‟s BAT model will be revised based on the optical information and more
advanced pilot model will be developed based on the above hybrid control structure.
6 References
1 Padfield, G. D., Helicopter Flight Dynamics, 2nd ed., Blackwell Science, Oxford, 2007.
2 Pavel, M. D., and Padfield, G. D., "Understanding the Peculiarities of Rotorcraft-Pilot
Couplings,” 64th Annual Forum of the American Helicopter Society, Montreal, Canada,
May 2008.
3 Lu, L., Padfield, G. D., and Jump, M., “The Strongly Controlled Helicopter,” 34th
European Rotorcraft Forum, Liverpool, UK, Sept. 16th – 19th, 2008.
4 McRuer, D. T., et al., Aviation Safety and Pilot Control Understanding and Preventing
Unfavourable Pilot-Vehicle Interactions, National Academy Press, ASEB National
Research Council, Washington D.C., 1997.
5 Gray, III, W. R., "Boundary-Avoidance Tracking: A New Pilot Tracking Model,” AIAA
Atmospheric Flight Mechanics Conference and Exhibit, San Francisco, California, pp.
86-97, Aug.15th - 18th, 2005.
6 Lu, L., Padfield G.D., and Jump, M., "Optical tau in boundary-avoidance tracking - a
new perspective on pilot-induced oscillations,” 36th European Rotorcraft Forum, Sept.
6th - 9th, Paris, France, 2010.
7 Lee, D. N., "Guiding Movement by Coupling Taus," Ecological Psychology, Vol. 10, No.
3, 1998, pp. 221-250.
8 Gibson, J. J., "The Perception of the Visual World," AJP, Vol. 64, 1951, pp. 622-625.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 32 of 35
9 Lee, D. N., Craig, C. M., and Grealy, M. A., "Sensory and Intrinsic Coordination of
Movement," Proc. of Biological sciences, Vol. 266, No. 1432, Jul.1999, .pp. 2029-2035.
10 Lee, D. N., "Tau in Action in Development," Action as an Organizer of Learning and
Development, edited by. J. J. Rieser, J. J. Lockman, and C. A. Nelson, 33rd Volume in
the Minnesota Symposium on Child Psychology, Minnesota, 2005, pp. 3-50.
11 Lee, D. N., "How movement is guided," Psychology Department, School of Philosophy,
Psychology and Language Sciences, University of Edinburgh, 2006.
12 Padfield, G. D., Lee, D. N., and Bradley, R., "How Do Helicopter Pilots Know When to
Stop, Turn or Pull Up?," AHS Journal, Vol. 48, No. 2, 2003, pp. 108-119.
13 Padfield, G. D., Clark, G., and Taghizad, A., "How Long Do Pilots Look Forward?
Prospective Visual Guidance in Terrain-Hugging Flight," AHS Journal, Vol. 52, No. 2,
2007, pp. 134-145.
14 Jump, M., and Padfield, G. D., "Investigation of the Flare Manoeuvre Using Optical
Tau," J. Guide. Control Dynamics,, Vol. 29, No. 5, pp. 1189-1200, Sept.2006.
15 Jump, M., and Padfield, G. D., "Progress in the Development of Guidance Strategies
For the Landing Flare Manoeuvre Using Tau-based Parameters," Aircraft Eng Aero
Tech, Vol. 78, No. 1, 2006, pp. 4-12.
16 Meyer, M., and Padfield, G. D., "First Steps in the. Development of Handling Qualities
Criteria for a Civil Tilt Rotor,” AHS Journal, Vol. 50, No. 1, pp.33-46, 2005.
17 Anon., Aeronautical Design Standard Performance Specification Handling Qualities
Requirements for Military Rotorcraft ADS-33E-PRF, US Army Aviation and Missile
Command, Aviation Engineering Directorate, Redstone Arsenal, Alabama, Mar. 2000.
18 Nguyen, N., Krishnakumar, K., and Kaneshige, J., " Flight Dynamics and Hybrid
Adaptive Control of Damaged Aircraft,” J. Guide. Control Dynamics,, Vol. 29, No. 5, pp.
751-764, Jun. 2008.
19 Kamalasadan, S., and Ghandakly, A. A., “A Neural Network Parallel Adaptive
Controller for Fighter Aircraft Pitch-Rate Tracking,” IEEE Transactions of
Instrumentation and Measurements, Vol. 60, No. 1, pp. 258-267, Jan. 2011.
20 Pasquiera, M., Quek, C., and Tohb, M., "Fuzzylot: A Novel Self-Organising Fuzzy-
Neural Rule-Based Pilot System for Automated Vehicles,” Neural Networks, Vol. 14,
No. 8, pp. 1099-1112, Oct. 2001.
21 Kamalasadan, S., and Ghandakly, A. A., “Multiple Fuzzy Reference Model Adaptive
Controller Design for Pitch-Rate Tracking,” IEEE Transactions of Instrumentation and
Measurements, Vol. 56, No. 5, pp. 1797-1808, Oct. 2007.
22 Innocenti, M., Pollini, L., and Turra, D., “A Fuzzy Approach to the Guidance of
Unmanned Air Vehicles Tracking Moving Targets,” IEEE Transactions on Control
Systems Technology, Vol. 16, No. 6, pp. 1125-1137, Nov. 2008.
23 Anderson, M. R., Clark, C., and Dungan, G., “Flight Test Maneuver Design Using a
Skill- and Rule-Based Pilot Model,” IEEE International Conference on Intelligent
Systems for the 21st Century, Systems, Man and Cybernetics, Vancouver, BC ,
Canada, 22th -25th Oct. 1995.
24 Gray III, W. R., “A Boundary Avoidance Tracking Flight Test Technique for
Performance and Workload Assessment”, Society of Experimental Test Pilots, 2007.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 33 of 35
25 Blake, R.D., “Boundary Avoidance Tracking: Consequences (and Uses) of Imposed
Boundaries on Pilot-Aircraft Performance,” MSc Thesis, Department of the Air Force,
United States Air Force Institute of Technology, Wright-Patterson Air Force Base, Ohio,
2009.
26 Dotter, J.D., “An Analysis of Aircraft Handling Quality Data Obtained From Boundary
Avoidance Tracking Flight Test Techniques,” MSc Thesis, Department of the Air Force,
United States Air Force Institute of Technology, Wright-Patterson Air Force Base, Ohio,
2009.
27 Brinkerink, N., and Pavel, M. D., "Capturing he Switch between Point Tracking and
Boundary Avoiding Pilot Behaviour in a PIO Event,” Proc. of the 4th Rotorcraft
Handling Qualities, Liverpool, 4th – 6th, UK, Nov. 2008.
28 Pavel, M.D., et al., "Adverse Rotorcraft-Pilot Couplings – Prediction and Suppression of
Rigid Body RPCs,” 34th European Rotorcraft Forum, Liverpool, England, Sept. 16th -
18th, 2008.
29 Gray III, W.R., “Handling Qualities Evaluation at the USAF Test Pilot School,” AIAA
Atmospheric Flight Mechanics Conference, Chicago, US, 10th - 13th Aug. 2009.
30 Hess, R. A., and Marchesi, F. “Analytical Assessment of Flight Simulator Fidelity Using
Pilot Models,” J. Guide. Control Dynamics, Vol. 32, No. 3, pp. 760-770, Jun. 2009.
31 Wilson, R. A., and Keil, F. C., The MIT Encyclopedia of the Cognitive Sciences
(MITECS), The MIT Press, 2001.
32 U.S.Air Force, Instrument Flight Procedures11-217, V1, Air Force Manual, 2010.
33 Shepherd, M.J., MacDonald, A., Gray III, W.R., and Cobb, R.G., “Limited Simulator
Aircraft Handling Qualities. Evaluation of an Adaptive Controller,” 2009 IEEE
Aerospace conference, Big Sky, Montana, Mar. 7th – 14th, 2009.
34 McRuer, D.T., 1995, Pilot-induced-oscillations and human dynamic behaviour. NASA
contractor report 4683.
35 Hess, R.A., “Unified Theory for Aircraft Handling Qualities and Adverse Aircraft-Pilot
Coupling”. Journal of Guidance, Control and Dynamics, Vol. 20 No 6 1997, pp 1141-
1148
36 Mohammad, M.L., Alastair, K.C. “Review of pilot modeling techniques”, 48th AIAA
Aerospace Sciences Meeting, AIAA 2010-297, 4-7 January 2010,Orlando, US
37 Schmidt, D. and Silk, A., “Modeling Human Perception and Estimation of Kinematic
Responses during Aircraft Landing”, AIAA Guidance, Navigation and Control
Conference, Minneapolis, MN, Aug 15-17, 1988, No. AIAA-1988-4186, 1988
38 Mohammad, M.L., Alastair, K.C. “Review of pilot modelling techniques”, 48th AIAA
Aerospace Sciences Meeting, AIAA 2010-297, 4-7 January 2010,Orlando, US
39 Levison, W. and Harrah, C., “Biomechanical and performance response of man in six
different directional axis vibration environments,” Tech. Rep. AMRL-TR-77-71,
Aerospace Medical Research Laboratory, 1977.
40 Griffin, M., “The validation of biodynamic models,” Clinical Biomechanics, Vol. 16,
2001, pp. 81–92.
41 Hohne, G., “A biomechanical pilot model for prediction of roll ratcheting,” AIAA
Atmospheric Flight Mechanics Conference, 1999.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 34 of 35
42 Magdaleno, R. , McRuer, D., “Experimental validation and analytical elaboration for
models of the pilot‟s neuromuscular subsystem in tracking tasks,” Tech. Rep. NASA
CR 1757, National Aeronautics and Space Administrations, 1971.
43 Tsang P.S., Vidulich M.A., “Principles and Practice of Aviation Psychology”, Aug 1,
2002
44 Tustin, A., “An Investigation of the Operator‟s Response in Manual Control of a Power
Driven Gun,” C.S. Memorandum 169, Metropolitan-Vickers Electrical Co., August 1944.
45 Smith, R., “A Unifying Theory for Pilot Opinion Rating,” Proceedings of the 12th Annual
Conference on Manual Control, No. NASA TM X-73, NASA, May 1976
46 Hosman, R., Pilot's Perception & Control of Aircraft Motions, No. 90-407-1384-7, Delft
University Press, November 1996.
47 Cardullo, F., Zaychik, K., and George, G., “A Conspectus on Operator Modeling: Past,
Present and Future,” AIAA Modeling and Simulation Conference and Exhibit, No. AIAA-
2006-6625, AIAA, Keystone, Colorado, August 2006.
48 Kleinman, D., Baron, S., and Levison, W., “An Optimal Control Model of Human
Response Part 1: Theory and Validation,” Automatica, Vol. 6, 1970, pp. 357–369.
49 Wierange R.D., “An Evaluation of a Pilot Model Based on Kalman Filtering and Optimal
Control”, IEEE Transactions on Man-Machine Systems, Vol. MMS-10, No.4, 1969
50 MeRuer, D. Graham, E. Krendel, and W. Reisener, Jr.,"Human pilot dynamics in
compensatory systems: Theory, models, and experiments with controlled element and
forcing function variations," AFFDL TR-65-15, July 1965.
51 Hess, A. R.,” Prediction of Aircraft Handling Qualities Using Analytical Models of the
Human Pilot”, NASA Technical Memorandum 84233, April 1982
52 John B,Davidson, David K.Schmidt. “Modified optimal control pilot model for computer-
aided design and analysis”, NASA TM-4384,1992
53 Chunguang W., Feng L., Junwei H.; Guixian L.,” A Revised Optimal Control Pilot
Model for Computer Simulation”, The 2nd International Conference on Bioinformatics
and Biomedical Engineering, 16-18 May 2008.
54 Ge Z., Xu H., Liu L., “A Variable Strategy Pilot Modeling and Application”, Proceedings
of the 2007 IEEE International Conference on Mechatronics and Automation August 5 -
8, 2007, Harbin, China.
55 Karl Ulrich Schultz, “Development of a nonlinear estimator based model of pilot
performance during brownout conditions”, Ph. D. dissertation, Old Dominion University,
2006 ,pp.90-120
56 Hu Zhaofen, “Man machine system and flying qualities”, Beihang University Press,
1994, pp.51-54
57 Qu Xiangju, Cui Hailian. “Variable strategy pilot model of carrier landing approach,”
Journal of Beijing University of Aeronautics and Astronautics, vol. 29, no. 11, pp. 993-
997, November 2003.
58 Zhenhai G., Xin G., Konghui G. “Application of preview follower theory and driver
model in the research of vehicle intelligent handling,” Journal of Traffic and Transport
Engineering, vol. 2, no. 2, pp. 993-997,June 2002.
59 Phillips J M, Anderson M R. “A variable strategy pilot model,” AIAA 2000-3983, 2000.
266073_ARISTOTEL_D2-3_ State_of_the_art _pilot _model _for _RPC_prediction_v1.0.doc
Page 35 of 35
60 Efremov A.V., Alexandrov V.V., Koshelenko A.V., Tjaglik M.S. “ Development of Pilot
Modeling and Its Application to Manual Control Tasks”, 27th International Congress of
the Aeronautical Sciences, ICAS 2010
61 Efremov A.V., Ogloblin A.V. “Progress in pilot-in the loop investigations for flying
qualities prediction and evaluation”, 25th International congress of the Aeronautical
Sciences, ICAS 2006
62 Efremov A.V. et al Pilot as a dynamic system (in Russian), Mashinostroenie. Russia.
1993.
63 Schönfeld A.,” Modified Optimal Control Model and Wake Vortex Encounter” ,
WakeNet3 - Europe Specific Workshop: Models and Methods for WVE Simulations,
Berlin, 2 June 2010
64 Anderson M R., “A Model of the Human Operator Using Sensitivity Function Shaping”,
Proceedings of the American Control Conference, TA-13- 11.40, Balimore, Maryland,
June 1994
65 Gestwa M., “Using Fuzzy Control for Modelling the Control Behaviour of a Human
Pilot”, German Aerospace Centre (DLR) Institute of Flight Systems, Germany
66 Yan.L., Shi-wei Y., Zheng-hong G. “Effects Of Control Allocation On Pilot-Induced
Oscillation (Pio) Susceptibility Of Aircrafts With Multiple Control Effectors”, 27th
International Congress of the Aeronautical Sciences, ICAS 2010
67 Johnson E.N., Pritchett A.R.,” Generic Pilot And Flight Control Model For Use In
Simulation Studies” , AIAA 2002-4694, AIAA Modeling and Simulation Technologies
Conference and Exhibit, Monterey, California, 5-8 August 2002
68 Vorst J.,” A pilot model for helicopter manoeuvres”, NLR-TP-98448, National
Aerospace Laboratory NLR, January 2001
69 Weber G., Efremov A.V., Oglobin A.V., “Development Of Criteria For Flying Qualities
Prediction Using Structural Modelling Of Human Pilot Behaviour In The Longitudinal
Precise Tracking Task”, European Conference For Aerospace Sciences (EUCASS),
France, 2005
70 Thompson, P.M., and McRuer, D.T., "Comparison of the Human Optimal Control and
Crossover Models," AIAA Paper No. 88-4183, Guidance, Navigation, and Control
Conference, Minneapolis, MN, Aug. 1988
7 List of Abbreviations
A/RPC Aircraft/Rotorcraft Pilot Coupling
PT Point Tracking
PAO Pilot-Assisted Oscillations
PIO Pilot-Induced Oscillations
SD Spatial Disorientation
SISO Single Input Single Output
MIMO Multi Input Multi Output
OCM Optimum Control Model
TDNN Time Delay Neural Network
NNM Neural Network Model
ROCM Revised Optimum Control Model
MOCM Modified Optimum Control Model
WVE Wake Vortex Encounter
BAT Boundary Avoidance Tracking
FTT Flight Test Technique
TPS Test Pilot school
C-PIO Conventional PIO
ILS Instrument Landing System