Você está na página 1de 24

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

net/publication/263534830

Joint discomfort human performance measure for driver posture prediction:


some insights

Article  in  International Journal of Human Factors Modelling and Simulation · January 2012


DOI: 10.1504/IJHFMS.2012.051553

CITATION READS

1 41

4 authors, including:

Jared Gragg Aimee Cloutier


Virginia Polytechnic Institute and State University Rose Hulman Institute of Technology
21 PUBLICATIONS   77 CITATIONS    26 PUBLICATIONS   81 CITATIONS   

SEE PROFILE SEE PROFILE

Jingzhou (James) Yang


Texas Tech University
226 PUBLICATIONS   2,390 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

mechanical engineering View project

Effects of Volitional Spine Stabilization on Asymmetric Lifting Task in Recurrent Low Back Pain Population View project

All content following this page was uploaded by Jingzhou (James) Yang on 17 July 2017.

The user has requested enhancement of the downloaded file.


Int. J. Human Factors Modelling and Simulation, Vol. 3, Nos. 3/4, 2012 253

Joint discomfort human performance measure for


driver posture prediction: some insights

Jared Gragg, Brad Howard,


Aimee Cloutier and James Yang*
Human-Centric Design Research Laboratory,
Department of Mechanical Engineering,
Texas Tech University, Lubbock, TX 79409, USA
E-mail: j.gragg@ttu.edu
E-mail: bradley.howard@ttu.edu
E-mail: aimee.cloutier@ttu.edu
E-mail: james.yang@ttu.edu
*Corresponding author

Abstract: Digital human models have proven to be valuable tools for


understanding human reach envelope inside a vehicle. Typical digital human
posture prediction simulations employ optimisation techniques that find the
most likely posture that a human would realise to achieve a given task. Human
performance measures are included as an objective function in the optimisation
formulation. A previous study (Yang et al., 2004) defined a joint discomfort
human performance measure based on joint angles that incorporates three
distinct aspects of joint discomfort. However, the previously defined joint
discomfort function is poorly understood. This paper investigates the properties
of the joint discomfort function and how each parameter in the function affects
the predicted posture. An alternate formulation of the joint discomfort human
performance measure is proposed. The parameters of the new joint discomfort
function are investigated through graphical analysis, ANOVA analysis, and
sensitivity analysis. The joint discomfort function is then employed in several
posture prediction simulations that pertain to reach space inside a vehicle. The
postures are given to demonstrate the effect that the parameters of joint
discomfort have on predicted postures.

Keywords: posture prediction; joint discomfort; digital human models; reach


envelope.

Reference to this paper should be made as follows: Gragg, J., Howard, B.,
Cloutier, A. and Yang, J. (2012) ‘Joint discomfort human performance measure
for driver posture prediction: some insights’, Int. J. Human Factors Modelling
and Simulation, Vol. 3, Nos. 3/4, pp.253–275.

Biographical notes: Jared Gragg is a PhD student in the Human-Centric


Design Research Lab of the Mechanical Engineering Department at Texas Tech
University. He received his BS and MS from Texas Tech University in 2008
and 2010, respectively. His research interests include digital human modelling
and simulation.

Brad Howard is a PhD candidate in the Human-Centric Design Research Lab of


the Mechanical Engineering Department at Texas Tech University. He received
his Bachelor’s degree in Mechanical Engineering in 2008 from Michigan
Technological University. His research interests include robotics and digital
human modelling and simulation.

Copyright © 2012 Inderscience Enterprises Ltd.


254 J. Gragg et al.

Aimee Cloutier is a PhD student in the Human-Centric Design Research Lab of


the Mechanical Engineering Department at Texas Tech University. She
received her BS degree with honours from Texas Tech University in May 2012.
Her research interests include human modelling and simulation and human
prosthetic design, fabrication and control.

James Yang is currently Assistant Professor and Director of the Human-Centric


Design Research Lab, Department of Mechanical Engineering, Texas Tech
University, Lubbock, Texas, USA. He received his BS and MS degrees in
Automotive Engineering from Jilin University, and PhD in Mechanical
Engineering from the University of Iowa, Iowa City, Iowa, USA. He was a
faculty member at the Department of Automotive Engineering, Tsinghua
University, Beijing, and Research Engineer and Adjunct Professor at the
Center for Computer Aided Design with the Department of Mechanical
and Industrial Engineering at the University of Iowa. His research interests
include physics-based human modelling and simulation, bio-inspired systems,
ergonomics, biomechanics, healthcare engineering, and robotic and mechanical
systems.

1 Introduction

Vehicle design is challenging in part because vehicles must be designed for such a vast
range of users. Human body size, shape, stature, joint range of motion, joint strength and
other factors may vary significantly from person to person. Furthermore, even for a single
person factors such as weight, stature, joint range of motion, and joint strength can
change over time. Due to differences in their anthropometry, people adapt different
postures to perform the same reach task within a vehicle. One of the main benefits of
incorporating digital humans into the design process for vehicles is that the various
changes in anthropometry listed above are easily incorporated into the simulations. The
effects of changing anthropometry can be studied using a single simulation by varying
the input parameters to the simulation. When incorporated into the early stages of the
design process, digital human modelling and simulation provide significant advantages.
Specifically, they can reduce the need for experimental testing of products, thus reducing
costs and time to market (Gragg et al., 2011). Digital human modelling offers a valuable
tool for dealing with various anthropometric differences easily and quickly.
An important design aspect of vehicle seats is driver comfort. A feeling of discomfort
while operating a vehicle can lead to dissatisfaction with the product as well as driver
fatigue. There are numerous studies in the literature that study discomfort as it relates to
driving a vehicle. El Falou et al. (2003) studied the evolution of indices of fatigue,
discomfort, and performance for subjects seated in two types of car seats during
prolonged simulated driving, both with and without vibration. Two types of discomfort
assessment methods were employed, a mechanical method based on acceleration
measurements and a subjective method based on subjects’ responses. Porter et al. (2003)
investigated whether pressure distribution data was valuable for predicting reported
discomfort. No clear relationship was found between the interface pressure data and
reported discomfort. Hermann (2005) explored sitting postural control using non-linear
methods to investigate discomfort during long-term driving. Body behaviour, as
classified by analysing the change of dynamics in the centre of pressure time series data
Joint discomfort human performance measure for driver posture prediction 255

with non-linear techniques, was used as an indicator of discomfort over time. Hermann
and Bubb (2007) extended the work by correlating centre of pressure data with percent of
pressure under the ischial tuberosities and upper thighs. Zschocke and Albers (2008)
examined the relationships between subjective and objective evaluations of vehicle
handling. Kyung and Nussbaum (2008) investigated whether pressure at the driver-seat
interface could be used as an objective measure of discomfort. The study examined
associations between three subjective ratings (overall, comfort, and discomfort) and 36
measures describing driver-seat interface pressure, and identified pressure level, contact
area, and ratio variables. Yang et al. (2009) developed a workspace zone differentiation
tool that allows users to visualise reach envelopes and discomfort values for seated
applications. The mathematical rating for discomfort was the objective function in a
multi-objective optimisation (MOO) for seated posture prediction. Ostensvik et al. (2009)
evaluated whether surface electromyography could be used as an indicator of discomfort.
Surface electromyography of the right upper trapezius and right extensor digitorum
muscles was compared to subjective discomfort ratings to see if there was a correlation.
Jin et al. (2009) compared objective and subjective measures of discomfort. Pressure
distribution at the driver-seat interface was compared to subjective ratings of discomfort
for eight subjects during a simulated 100 minute driving task. Jin et al. (2011) extended
the study by examining associations between three subjective ratings (overall, comfort,
and discomfort) and over 200 measures describing driver-seat interface contact pressure,
force, and area and ratios of local to whole body. Junoh et al. (2011) developed a linear
programming model for optimising noise and vibration inside a vehicle cabin. Discomfort
due to vibration was related to magnitude, frequency, direction, and duration of the
vibration and quantified by the vibration dose value. Kang and Kaizu (2011) performed
vibration analysis on grass harvesting tractors. Discomfort was related to vibration
acceleration according to ISO vibration standards. Koizumi et al. (2011) evaluated ride
comfort performance during compound vibrations. The following conclusions were
obtained: discomfort depends on the increase and decrease of the acceleration vector
summation, amplitude ratio pattern influenced discomfort, and discomfort increased as
frequency increased. Majid et al. (2011) investigated the main causes for muscle fatigue
and discomfort during city driving. A full-body joint model was used to study the
interactions between the drivers and vehicle using an inverse dynamics approach. Wang
et al. (2011) developed a motion related discomfort modelling approach based on the
concept of ‘less constraint movement’. From questionnaire and motion analysis it was
observed that pedal resistance had a dominant effect on discomfort perception. Wang and
Trasbot (2011) investigated the effects of target location, stature, and hand grip type on
in-vehicle reach discomfort. Findings were based on the statistical analysis of subjective
ratings when reaching a target in a vehicle. Basri and Griffin (2011) studied the vibration
of backrests as it contributes to driver and passenger discomfort. The method of
magnitude estimation was used to obtain judgements on vibration discomfort for four
backrest angles. Gragg and Yang (2011) studied the effect of obesity on discomfort and
seated posture inside a vehicle. Gragg et al. (2012) developed a hybrid method for driver
accommodation using optimisation-based digital human models that minimised joint
discomfort for drivers of various anthropometries. Several additional studies have
incorporated subjective discomfort ratings in their findings (Massaccesi et al., 2003;
Jonsson and Johansson, 2005; Na et al., 2005; Durkin et al., 2006; Schust et al., 2006;
Smith et al., 2006; Robb and Mansfield, 2007; Szeto and Lam, 2007; Kyung et al., 2008;
Maeda et al., 2008; Carvalho and Callaghan, 2011).
256 J. Gragg et al.

Yang et al. (2004) reported a joint discomfort human performance measure for a
MOO-based posture prediction method. The discomfort human performance measure
incorporates three distinct facets of joint discomfort. The approach does not aim to
quantify discomfort but rather to model components that are proportional to discomfort.
While the discomfort human performance measure is a useful aid in digital human
posture prediction simulations, the function is not well understood. This study aims to
gain a better understanding of the discomfort human performance measure and its
formulation. Specifically, the original formulation’s parameters are investigated, and joint
discomfort values are plotted over the range of motion for a typical joint. Due to the high
sensitivity of the original discomfort function to its parameters, an alternative formulation
of joint discomfort is proposed and studied.
This paper investigates the properties of a new joint discomfort human performance
measure and how each parameter in the function affects the predicted posture in the
direct-optimisation formulation. First, an alternate formulation of the joint discomfort
human performance measure is given. Second, the parameters of the joint discomfort
function are investigated through graphical analysis, ANOVA analysis, and sensitivity
analysis. Finally, the joint discomfort function is employed in several posture prediction
simulations that pertain to reach space inside a vehicle. The postures are given to
demonstrate the effect that the parameters of joint discomfort have on predicted postures.

2 Joint discomfort human performance measure

This section investigates the joint discomfort human performance measure. First,
background on the joint discomfort function is given. Second, the joint discomfort
function is examined using originally selected values for function parameters. Third, a
new joint discomfort function is given with suggested values for parameters.

2.1 Original joint discomfort human performance measure


The joint discomfort human performance measure was introduced by Yang et al. (2004)
and Marler et al. (2005). The joint discomfort function was based on the lexicographic
method for MOO. In a typical MOO problem, several human performance measures, fk,
are combined in a single weighted objective function as follows:
1
⎧⎪ ⎡ ⎛ fk ⎞ ⎤
2
⎫⎪ 2
F(q) = ⎨
⎪⎩

k
⎢ωk ⎜ max
⎣ ⎝ fk
⎟ + 1⎥
⎠ ⎦

⎪⎭
(1)

where ωk is the weight and f kmax is the maximum value of the kth human performance
measure. Several human performance measures were used: joint displacement, visual
displacement, joint torque, joint discomfort, and delta potential energy (Yang et al.,
2004). Each human performance measure has desirable qualities but generally a single
human performance measure will not yield the most realistic posture, so several human
performance measures are combined using the MOO formulation.
The joint discomfort human performance measure incorporates three aspects of joint
discomfort. They are:
Joint discomfort human performance measure for driver posture prediction 257

1 the tendency to move toward a generally comfortable position or neutral position


2 the tendency to avoid postures in which joint angles are pushed to their limits, except
for joints where ligaments and/or tendons are not stretched as with the elbow and
clavicle
3 the idea that people strive to reach or contact a point using one set of body parts at a
time, i.e., one first tries to reach a point with one’s arm, then if necessary bends the
torso, then finally extends the clavicle (Yang et al., 2004).
Due to the computational expense for solving the lexicographic method, a weighted sum
method was introduced to approximate the results. For the weighted sum method, joint
weights were chosen as follows: for the joints composing the spine, the established joint
weight was 1 × 104; for the joints composing the clavicle, the established joint weight
was 1 × 108; and for the joints composing the right arm, the established joint weight was
1. It is mentioned in the reference that the values of the joint weights are not significant,
but rather that the weights have infinitely different orders of magnitude, so as to
approximate the results of the lexicographic method (Yang et al., 2004; Miettinen, 1999;
Romero, 2000).
The joint discomfort function was defined in Yang et al. (2004) as follows:

DOF
1
f dis (q) =
G ∑ ⎡⎢⎣γ ( Δq )
i =1
i
norm 2
i + G × QU i + G × QLi ⎤⎥

q − qiN
Δqinorm = Ui
qi − qiL
100
⎛ ⎛ 5.0 ( qiU − qi ) ⎞ ⎞ (2)
QU i = ⎜ 0.5sin ⎜ + 1.571⎟ + 1⎟
⎜ ⎜ ⎟ ⎟
⎝ qi − qi
U L
⎝ ⎠ ⎠
100
⎛ ⎛ 5.0 ( qi − qiL ) ⎞ ⎞
QLi = ⎜ 0.5sin ⎜ + 1.571 ⎟ + 1⎟
⎜ ⎜ qiU − qiL ⎟ ⎟
⎝ ⎝ ⎠ ⎠

where γi is the weight for the ith joint, G = 1 × 106 is a constant, qi is the ith joint angle,
qiN is the neutral angle for the ith joint, and qiU and qiL are the upper and lower limits for
the ith joint. The joint discomfort function is broken into three distinct parts. (Δqinorm ) 2
acts as an individual objective function similar to joint displacement. This term
approximates the human tendency to minimise movement away from a comfortable
neutral posture. Generally, this term is satisfactory in and of itself, but can result in
postures where multiple joints are extended to their limits, which is uncomfortable.
Thus, two additional terms, QUi and QLi, are added to the discomfort function as
penalties for joints that approach their upper and lower limits, respectively. Each term
varies between zero and one and is multiplied by the penalty constant, G. The penalty
terms, QUi and QLi, have a value of zero until the joint reaches the upper or lower 10% of
its range.
258 J. Gragg et al.

Table 1 Parameter values for equation (2)

G γ α β
6
10 1, 104, 108 100 5

2.2 Original joint discomfort human performance measure evaluation


The joint discomfort human performance measure has been analysed for an arbitrary joint
with upper and lower joint limits of π and 0, respectively. The neutral angle for this
arbitrary joint was given as half the difference between the upper and lower joint limits,
or π / 2. Figure 1 gives the discomfort function value over the full joint range of motion
for an arbitrary joint with weight values of γ = 1, γ = 1 × 104and γ = 1 × 108 respectively.
The joint discomfort function plotted is the discomfort value for a single joint, given by
equation (3).

DISi = γi ( Δqinorm ) + G × QU i + G × QLi


2
(3)

Figure 1 Joint discomfort values for an arbitrary joint with different γ values, (a) vertical axis
scale [0, 4 × 1023] (b) vertical axis scale [0, 50] (see online version for colours)

γ=1
γ=104
γ=108

(a)

γ=108
γ=104
γ=1

(b)
Joint discomfort human performance measure for driver posture prediction 259

As seen from Figure 1 there are problems with the selected values in Table 1 for the joint
discomfort function. From Figure 1(a) it is seen that with an expanded vertical axis scale
the joint discomfort function has the desired shape for weight values of γ = 1, 104, 108.
However, when the vertical axis scale is shrunk down, as seen in Figure 1(b), the joint
discomfort function exhibits unwanted behaviour when using the selected weight values
in Table 1. In Figure 1(b), for γ = 1, it is seen that there is essentially zero discomfort
until the joint angle becomes approximately 3π / 8 or 5π / 8, where the discomfort value
begins to increase exponentially. According to the claims in Yang et al. (2004), the
discomfort value should not experience the penalty until the joint angle approaches 10%
of the upper or lower limits, in this case π / 10 and 9π / 10. In Figure 1(b), for γ = 104, it is
seen that the discomfort value begins to increase exponentially if the joint angle changes
even slightly from the neutral angle, π / 2. For γ = 108, it is seen that any perturbation of
the joint angle from the neutral angle will cause an exponential increase in the
discomfort. It is obvious from the plots in Figure 1 that the joint discomfort function as
previously formulated does not accurately model human behaviour. For joints with a
weight of γ = 1, the joint discomfort function does not limit movement away from the
neutral posture at all until the joint reaches 3/8 of its upper or lower limit. This is in
contrast to the claim of the authors that the term (Δqinorm )2 acts as an individual objective
function similar to joint displacement. Further, due to the minimisation technique
employed in the posture prediction scheme, it is apparent that the joint will never reach a
joint angle other than one that lies between 3/8 and 5/8 of its joint range of motion. This
is clearly contradictory to human behaviour. For joints with a weight of γ = 1 × 104, the
discomfort function allows for only slight perturbation from the neutral angle. For joints
with a weight of γ = 1 × 108, the discomfort function essentially keeps the joint from
moving at all from the neutral angle. The next section details the improved joint
discomfort human performance measure.

2.3 New joint discomfort human performance measure


The joint discomfort human performance measure has been modified as follows:

MDi = γi ( Δqinorm ) + G × QU i + G × QLi


2

qi − qiN
Δqinorm =
qiU − qiL
α α
⎛ ⎛ ⎛ β ( qiU − qi ) ⎞ ⎞ ⎞ ⎛ ⎛ β ( qiU − qi ) ⎞ ⎞ (4)
QU i = ⎜ 0.5 ⎜ cos ⎜ U ⎟ + 1 ⎟ ⎟ = 0.5α ⎜ cos ⎜ ⎟ + 1⎟
⎜ ⎜ ⎜ qi − qiL ⎟ ⎟ ⎟ ⎜ ⎜ qiU − qiL ⎟ ⎟
⎝ ⎝ ⎝ ⎠ ⎠⎠ ⎝ ⎝ ⎠ ⎠
α α
⎛ ⎛ ⎛ β ( qi − qiL ) ⎞ ⎞ ⎞ ⎛ ⎛ β ( qi − qiL ) ⎞ ⎞
QLi = ⎜ 0.5 ⎜ cos ⎜ U ⎟ + 1 ⎟ ⎟ = 0.5α ⎜ cos ⎜ ⎟ + 1⎟
⎜ ⎜ ⎜ qi − qiL ⎟ ⎟ ⎟ ⎜ ⎜ qiU − qiL ⎟ ⎟
⎝ ⎝ ⎝ ⎠ ⎠⎠ ⎝ ⎝ ⎠ ⎠

where α and β are scaling coefficients and MDi is the modified joint discomfort for the ith
joint. Note the similarities of the new discomfort function. The sin (χ + 1.571) term has
been changed to cos χ, due to the trigonometric identity sin(χ + π/2) = cos χ. Second,
there is an additional parenthesis after the 0.5 term in the penalty terms. Other than these
260 J. Gragg et al.

two changes, equations (3) and (4) are identical when α = 100 and β = 5.0. The addition
of the parenthesis in the penalty term may seem minor, but the results are drastic.
The new joint discomfort function has been analysed for an arbitrary joint with upper
and lower joint limits of π and 0, respectively. The neutral angle for this arbitrary joint
was given as half the difference between the upper and lower joint limits, or π / 2.
Figure 2 gives the joint discomfort function value over the full joint range of motion for
an arbitrary joint with weight values of γ = 1, γ = 1 × 104, and γ = 1 × 108, respectively.
The joint discomfort function plotted is the discomfort value for a single joint, given by
equation (4). The scaling parameters are given as α = 100 and β = 5.0. The original
discomfort function is also shown for comparison.

Figure 2 Joint discomfort value, DIS, and modified joint discomfort value, MD, for arbitrary joint
with different γ values (see online version for colours)

DIS, MD: γ=108


DIS, MD: γ=104
DIS: γ=1
MD: γ=1

As seen from Figure 2, the joint discomfort values for the new joint discomfort function
are similar for the previously selected weight values of γ = 1, 104, 108. Only the joints
with weight γ = 1 show significant improvement. Through analysis of the new discomfort
function it was determined that more suitable parameter values are: G = 50, γ = 50,
α = 100, β = 5. When the new joint discomfort function is plotted with these parameter
values, the result is an appropriate discomfort relationship for an arbitrary joint with
upper and lower joint limits of π and 0, respectively, and a neutral angle of π / 2. This plot
is shown in Figure 3.
Note the new discomfort function, as seen in Figure 3, demonstrates the desired
characteristics: The function is minimal at the neutral angle, with an increasing
discomfort value as the joint angle moves away from the neutral angle. Also, the penalty
begins to affect the discomfort at approximately 10% of the joint’s upper and lower
limits.
Joint discomfort human performance measure for driver posture prediction 261

Figure 3 New joint discomfort function with appropriate parameters (see online version
for colours)

G=50
γ=50
α=100
β=5

3 Joint discomfort human performance measure parameter analysis

Three analyses were performed on the new joint discomfort human performance measure
in order to study the effect of the parameter values ( qiN , G , γi , αi , βi ) on the discomfort
function value. The three analyses were graphical analysis, ANOVA analysis, and
sensitivity analysis. The graphical analysis was used to demonstrate the effect that
changing a single parameter had on the discomfort value. The ANOVA analysis was used
to study the effect that changing multiple parameters had on the discomfort value. The
sensitivity analysis was used to determine the discomfort function’s sensitivity to the
input parameters.

3.1 Graphical analysis


As seen from Figure 3, acceptable parameter values for the new joint discomfort
function, equation (4) are: G = 50, γ = 50, α = 100, β = 5. For the graphical analysis, the
parameter values of the new joint discomfort function, equation (4), were set as:
qiU = π , qiL = 0, qiN = π / 2, G = 50, γ = 50, α = 100, β = 5 (5)

The following, Figures 4 to 8, detail the effect that holding all of the parameter values of
the joint discomfort function constant except for one, which is varied and plotted, has on
the discomfort value. The arrow shows the trend for each parameter as it increases.
Figure 4 shows the effect that the parameter G has on the joint discomfort function, with
values ranging from G = 0, 10, …, 100.
262 J. Gragg et al.

Figure 4 Effect of parameter G on joint discomfort (see online version for colours)

G=100

G=50

G=0

As seen in Figure 4, the penalty parameter G affects how much discomfort is added for a
joint that is near its limits. A higher value of the penalty parameter causes a steeper rise in
discomfort. A penalty parameter of zero has no effect on the discomfort function.
Figure 5 shows the effect that the weighting parameter γ has on the discomfort function,
with values ranging from γ = 0, 40, …, 400.

Figure 5 Effect of parameter γ on joint discomfort (see online version for colours)

γ=400

γ=200

γ=0

As seen in Figure 5, the weighting parameter γ affects how much discomfort is added for
a joint in the middle portion of its range of motion. A higher value of the weighting
parameter causes a steeper rise in discomfort. This effect is similar to the joint
displacement human performance measure (Yang et al., 2004). A weighting parameter of
zero only takes into consideration the penalty when the joint is near its limits. Figure 6
shows the effect that the scaling parameter α has on the discomfort function, with values
ranging from α = 10, 20, …, 100.
Joint discomfort human performance measure for driver posture prediction 263

Figure 6 Effect of parameter α on joint discomfort (see online version for colours)

α=10 α=100

As seen in Figure 6, the scaling parameter α affects when the penalty for a joint being
near its limits begins to dominate the function. A lower value of the scaling parameter
causes the penalty to dominate closer to the neutral angle. This phenomenon can be
controlled through either scaling parameter, α or β. Figure 7 shows the effect that the
scaling parameter β has on the discomfort function, with values ranging from
β = 0.5, 0.75, …, 5.

Figure 7 Effect of parameter β on joint discomfort (see online version for colours)

β=0.5

β=5

As shown in Figure 7, the scaling parameter β affects when the penalty for a joint being
near its limits begins to dominate the function. A lower value of the scaling parameter
causes the penalty to dominate closer to the neutral angle. This phenomenon can be
controlled through either scaling parameter, α or β. Figure 8 shows the effect that the
264 J. Gragg et al.

neutral angle qiN has on the discomfort function, with values ranging from
qiN = π / 4, 7π / 20,… , 3π / 4.

Figure 8 Effect of parameter qiN on discomfort (see online version for colours)

Figure 9 Joint discomfort curves for the human upper body (see online version for colours)

γ=200

γ=100

γ=1

As seen in Figure 8, the neutral angle qiN affects where the joint discomfort function is
minimal, or zero. The parameter has no effect on the shape of the function. By varying
the parameters, G, γ, α, β, it is possible to construct curves for a wide range of desired
shapes. For human posture prediction, it is generally acceptable to employ parameter
values of G = 50, α = 100, β = 5 and choose appropriate weighting parameters that
demonstrate the three aspects of joint discomfort mentioned previously in the text
(Yang et al., 2004). For the new discomfort function, equation (4), appropriate values are:
Joint discomfort human performance measure for driver posture prediction 265

γ = 1 for the arm joints, γ = 100 for the spine joints, and γ = 200 for the clavicle joints.
Figure 9 shows the discomfort curves for joints with these joint weights of γ = 1, 100,
200.

3.2 The analysis of variance and 24 factorial design


A 24 full factorial design was created and run in order to test the statistical significance of
the four main input effects (previously referred to as parameters): α, β, γ, and G. The
major interactions of the effects were also tested. The ‘high-low’ levels of the parameters
in the 24 design correspond to the values used in the plots of Figures 4 to 7. These values
are shown in Table 2.
Table 2 Values used for the 24 factorial design

G γ α β
High 100 400 100 5
Low 0 0 10 0.5

The 24 design was analysed using the built-in standard least squares script in the JMP®8
Statistical Discovery Software. The results from the analysis are summarised. The
analysis of variance (ANOVA) is shown in Table 3. A test on the lack of fit of the model
estimation was also completed. The results are shown in Table 4. No significance was
found in the lack of fit so the parameter estimates can be trusted.
Table 3 ANOVA table detailing the results of the ANOVA

Analysis of variance
Source DF Sum of squares Mean square F-ratio
Model 10 535,339.7 53,534 75.0636
Error 165 117,674.9 713.2 Prob. > F
Total 175 653,014.7 < .0001*

Table 4 Results detailing lack of fit

Lack of fit
Source DF Sum of squares Mean square F-ratio
Lack of fit 5 7,803.35 1,560.67 2.2727
Pure error 160 109,872 686.7 Prob. > F
Total error 165 117,675 0.0498

The parameter estimates are shown in Table 5. All of the main effects are significant, as
they are all inputs into a mathematical function. G is the most significant, followed by β,
α, and γ, as shown in the table. The interactions of the main effects G and β, G and α are
also significant. The significance here deals with the significance of the contribution to
the discomfort value. The more significant the effect is, the more sensitive the discomfort
value is to the effect’s change.
266 J. Gragg et al.

Table 5 Parameter estimates

Sorted parameter estimates


Term Estimate Std. error t-ratio Prob. > |t|
G 37.69146 2.012999 18.72 < .0001*
β –22.78911 2.012999 –11.32 < .0001*
G *β –22.78911 2.012999 –11.32 < .0001*
α –13.85302 2.012999 –6.88 < .0001*
G *α –13.85302 2.012999 –6.88 < .0001*
γ 12.41855 2.012999 6.17 < .0001*
α *β 6.658622 2.012999 3.31 0.0012*
γ *α –7.90E-10 2.012999 0 1
G*γ –6.19E-10 2.012999 0 1
γ *β –3.92E-10 2.012999 0 1

3.3 Sensitivity analysis


A probabilistic analysis was performed using G, γ, α, and β as random variables in the
discomfort function to find the sensitivity factors of the function to each input parameter.
The sensitivities are found based on the function’s probability of failure, or the likelihood
that the value of the function will exceed a certain level of discomfort. In this case, the
threshold defining the probability of failure is the point at which 30% of the values of qi
will cause the discomfort to exceed the threshold; i.e., the threshold is the value of the
discomfort function at qi = π / 15 or qi = 14π / 15. The safety index, defined according to
equation (6), is the ratio of the mean of the discomfort function to the standard deviation.
The safety index is used to calculate the sensitivity factors according to equation (7).
μz
SI = = −Φ −1 ( Pf ) (6)
σz

∂SI ∂SI
SFi = or (7)
∂μi ∂σ i

Based on the acceptable values obtained for the new discomfort function, the random
variables were normally distributed with a coefficient of variation of 5% (the ratio of the
standard deviation to the mean) and mean values of G = 50, α = 100, and β = 5. The
sensitivity factors were found for three different cases: using mean values of γ = 1, 100,
and 200. In each case, the joint angle, qi, was varied with a mean of π / 2 and a standard
deviation of π / 4. This means that the probabilistic analysis focuses on the sensitivities of
the input parameters within the range of π / 4 ≤ qi ≤ 3π / 4. qiU , qiL , and qiN were given
deterministic values of π, 0, and π / 2, respectively. The probabilistic analysis was
performed using NESSUS software, and the sensitivity factors are shown in Figure 10.
Joint discomfort human performance measure for driver posture prediction 267

Figure 10 (a) Sensitivities with respect to the mean for three different mean values of γ
(b) Sensitivities with respect to the standard deviation for three different mean values
of γ (see online version for colours)

(a)

(b)
268 J. Gragg et al.

In Figure 10, each group of bars corresponds to the sensitivity of the discomfort function
to the input parameters G, γ, α, and β. The sensitivity factors show whether the amount of
discomfort would increase or decrease if the mean or standard deviation were to change.
A positive sensitivity factor indicates that if the mean or standard deviation were raised,
the amount of discomfort would likely increase; if the sensitivity factor is negative,
raising the mean or standard deviation would likely lower the amount of discomfort.
Based on the results shown in Figure 10(a), the sensitivity of γ with respect to the
mean is much higher than the sensitivities of the other three parameters. Consistent with
the graphical analysis in Figure 5, the sensitivity analysis shows that raising γ causes a
steep rise in discomfort. Figure 5 also shows that raising the value of γ can significantly
increase the amount of discomfort. Based on the results of the graphical analysis, high
sensitivity factors for γ can be expected within the range of π / 4 ≤ qi ≤ 3π / 4.
Figure 10(b) shows that the discomfort function is most sensitive to changing the
standard deviation of the parameter β. Because the sensitivity factors are negative, raising
the standard deviation of β would lower the likelihood of increased discomfort.
Sensitivity factors for G are generally low. This result can be predicted by examining
the graphical analysis in Figure 4. Raising G does significantly affect the amount of
discomfort, but only at the outer edges of the joint range of motion. Until the penalty
parameter takes effect, there is no change in discomfort. Therefore, the sensitivity of the
discomfort function to the parameter G is expected to be lower, for the sensitivity
analysis, because G only affects the outer edges of the joint range of motion.
The results from the ANOVA and sensitivity analysis differ because they are studying
different things. The ANOVA looks at the relationship between the input parameters and
the discomfort function value over the whole joint range of motion, while the
probabilistic analysis focuses on a specific range of values, π / 4 ≤ qi ≤ 3π / 4. With the
sensitivity analysis, it is difficult to formulate the problem for the entire joint range of
motion, since probabilistic design focuses the analysis around the mean value of a
parameter, which in this case is the neutral angle.

4 Posture prediction with new joint discomfort function

The posture prediction algorithm was developed as part of previous work (Yang et al.,
2004). The posture prediction algorithm is a direct optimisation-based problem where
human performance measures serve as a single objective function or are combined in a
MOO problem as described in Section 2. For this study, the new joint discomfort human
performance measure, equation (4), was used as the objective function in the posture
prediction algorithm. The constraints were that the distance between the right hand
end-effector (index finger of the right hand) and the target point (specified by vehicle
geometry) was zero and all joint angles fell within upper and lower joint limits based on
human anatomy. The human model was adapted from previous work (Cloutier et al.,
2012). The model has 21 degrees of freedom (DOF) consisting of: the spine, q1, …, q12;
the clavicle joints, q13, q14; and the arm joints, q15, …, q21. Parameter values of
G = 50, α = 100, β = 5 were chosen for all joints and weight values were: γ = 1 for the
arm joints, q15, …, q21; γ = 100 for the spine joints, q1, …, q12; and γ = 200 for the clavicle
joints, q13, q14. The human model is detailed in Figure 11.
Joint discomfort human performance measure for driver posture prediction 269

Figure 11 21-DOF human upper-body model (see online version for colours)

Clavicle
z13

z15 z11
z10
z12
z16 z9
z8 z7
z18 z14 Spine
z20 z17 Shoulder z6
z5 z4
Elbow z3
z19 z2 z1
z0
Wrist Y Hip

Global
X
Reference
Frame
Z

Seven reach tasks were defined for a driver inside a vehicle. The seven target points were
as follows: parking brake (PB), gearshift (GS), radio tuner (RT), glove box (GB),
rearview mirror (RM), visor (V), and seatbelt (SB). All target points were measured from
the driver’s seat of a 2006 Toyota Corolla. The Cartesian coordinates of each of the target
points can be found in Table 6. The coordinate system was affixed to the centre of the hip
joints as defined in Figure 12. The target points were ordered by their distance from the
right hand end-effector when the human model was at the neutral posture. The neutral
posture was adapted from Yang et al. (2004). The human model’s link lengths were
representative of a 50th percentile (stature) female.

Figure 12 Seven reach tasks for a driver (see online version for colours)

Visor
Rearview Mirror
Seatbelt

Radio Tuner
Y
X Z Glove Box

Gearshift

Parking Brake
270 J. Gragg et al.

Table 6 Coordinates of reach tasks (cm)

Target point Abbrev. x y z


Parking brake PB –30.0 –6.0 16.0
Gearshift GS –37.5 –1.5 33.0
Radio tuner RT –46.5 19.0 48.0
Glove box GB –77.0 2.0 50.0
Rearview mirror RM –37.5 60.0 28.0
Visor V 0.0 64.5 19.0
Seatbelt SB 28.5 46.0 –22.0

Table 7 details the results from the seven posture prediction simulations. The table
provides the following information: joint weight, γi, for each joint angle, qi; the percent
change of each joint angle from the neutral posture (posture at the start of the simulation)
to the final posture (completion of the reach task), calculated according to equation (8),
for all reach tasks; and the absolute change in each joint angle from the neutral posture to
the final posture for all reach tasks. Table 8 details the resulting joint discomfort, MD,
values for each of the reach tasks.
⎧ qiF − qiN F N⎫
⎪ qU − q N , qi ≥ qi ⎪
⎪ i i ⎪
% change = ⎨ F ⎬, (8)
⎪ qi − qi , q F ≤ q N ⎪
N

⎪⎩ qiN − qiL i i
⎪⎭

where qiF is the value of the joint angle in the final posture.
As seen in Table 7, most reach tasks can be completed by movement of the arm, with
limited movement in the spine and clavicle regions. The exceptions are the glove box and
seatbelt. This concurs with the first and third aspects of joint discomfort: the first aspect
being that humans tend to limit movement away from a comfortable posture, or neutral
posture; and the third aspect being that humans tend to move their arms to reach targets
before the spine and clavicle. Note that the change in joint angles of the spine and
clavicle regions for the reach tasks PB, GS, RT, RM, and V are small. Thus, movement
from the neutral posture is limited. Also note that joint angles of the arm regions for these
reach tasks change significantly. Thus, the human model is reaching the target point with
its arms alone and not moving the spine and clavicle regions. For the other two reach
tasks, GB and SB, it is necessary for the human model to reach first with its arm and then
its spine and finally the clavicles. This is evidenced by the fact that the absolute change,
Δq, in joint angles for the arm region is large, followed by a lesser movement in the spine
regions and lesser still in the clavicle regions. These results are intuitive. The only two
reach tasks where it is necessary to use the spine and clavicles to reach the point are the
glove box, GB, and seatbelt, SB. The results also demonstrate the second aspect of joint
discomfort, which is that people tend to avoid postures in which joints are pushed to their
limits. As seen from Table 7 the only reach task that results in a posture where a joint
angle reaches its joint limits is the seatbelt, SB. All other joint angles for the other reach
tasks result in a posture where each angle lies within 10% of its upper and lower joint
limits.
Table 7

PB GS RT GB RM V SB
qi γi % % % %
% change Δq (º) Δq (º) Δq (º) % change Δq (º) % change Δq (º) Δq (º) Δq (º)
change change change change
Spine1 1 100 –0.24% –0.04 –0.16% –0.02 –0.14% –0.02 49.59% 7.44 49.59% 7.44 –0.24% –0.04 25.17% 3.78
2 100 –0.25% –0.03 –0.29% –0.03 –0.21% –0.02 26.56% 6.64 26.56% 6.64 –0.25% –0.03 14.59% 3.65
3 100 –0.02% 0.00 0.06% 0.01 0.10% 0.01 5.35% 0.80 5.35% 0.80 –0.02% 0.00 1.41% 0.21
Spine2 4 100 –0.19% –0.03 –0.13% –0.02 –0.12% –0.02 41.11% 6.17 41.11% 6.17 –0.19% –0.03 54.72% 8.21
5 100 –0.18% –0.02 –0.27% –0.03 –0.21% –0.02 21.60% 5.40 21.60% 5.40 –0.18% –0.02 33.41% 8.35
6 100 –0.02% 0.00 0.06% 0.01 0.10% 0.01 6.39% 0.96 6.39% 0.96 –0.02% 0.00 3.52% 0.53
Spine3 7 100 –0.13% –0.02 –0.11% –0.02 –0.11% –0.02 30.88% 4.63 30.88% 4.63 –0.13% –0.02 80.54% 12.08
8 100 –0.11% –0.01 –0.24% –0.02 –0.22% –0.02 16.02% 4.01 16.02% 4.01 –0.11% –0.01 59.06% 14.76
9 100 –0.02% 0.00 0.06% 0.01 0.10% 0.01 6.96% 1.04 6.96% 1.04 –0.02% 0.00 13.05% 1.96
Spine4 10 100 –0.07% –0.01 –0.08% –0.01 –0.10% –0.01 20.39% 3.06 20.39% 3.06 –0.07% –0.01 83.30% 12.49
11 100 –0.03% 0.00 –0.22% –0.02 –0.22% –0.02 10.61% 2.65 10.61% 2.65 –0.03% 0.00 84.87% 21.22
12 100 –0.02% 0.00 0.06% 0.01 0.10% 0.01 7.22% 1.08 7.22% 1.08 –0.02% 0.00 44.12% 6.62
Posture prediction results for seven reach tasks

Clavicle 13 200 0.14% 0.03 –0.01% –0.01 –0.05% –0.03 0.76% 0.17 0.76% 0.17 0.14% 0.03 82.79% 18.61
14 200 –0.05% –0.02 0.05% 0.02 0.10% 0.05 17.70% 8.33 17.70% 8.33 –0.05% –0.02 –22.05% –9.47
Shoulder 15 1 41.06% 15.97 37.64% 14.64 17.72% 6.89 –49.01% –74.06 –49.01% –74.06 41.06% 15.97 78.07% 30.36
16 1 –25.26% –6.37 15.56% 20.19 37.04% 48.06 34.40% 44.64 34.40% 44.64 –25.26% –6.37 –100.00% –25.23
17 1 –11.72% –8.20 –14.91% –10.44 –15.04% –10.53 –1.39% –0.97 –1.39% –0.97 –11.72% –8.20 –86.32% –60.43
Elbow 18 1 –28.11% –33.82 –17.38% –20.92 –11.12% –13.38 46.29% 11.42 46.29% 11.42 –28.11% –33.82 70.33% 17.34
19 1 –25.39% 19.09 –12.01% 9.03 –7.78% 5.85 –67.09% –50.45 –67.09% –50.45 –25.39% 19.09 –65.51% –49.26
Wrist 20 1 –3.60% –2.04 –3.22% –1.82 –2.65% –1.50 9.62% 0.81 9.62% 0.81 –3.60% –2.04 –0.07% –0.04
21 1 65.24% 34.16 31.02% 16.24 20.95% 10.97 –10.59% –8.22 –10.59% –8.22 65.24% 34.16 62.54% 32.75
Joint discomfort human performance measure for driver posture prediction
271
272 J. Gragg et al.

Table 8 Joint discomfort, MD, value for seven reach tasks

Target point Abbrev. MD


Parking brake PB 0.160
Gearshift GS 0.074
Radio tuner RT 0.124
Glove box GB 24.389
Rearview mirror RM 0.377
Visor V 0.813
Seatbelt SB 209.005

As seen in Table 8, the discomfort value indicates a posture’s discomfort. The reach tasks
that require movement of the spine and clavicles, GB and SB, exhibit significantly more
discomfort than the other reach tasks. These results are intuitive. It can also be seen from
Table 8 that reaching for the visor, V, is less comfortable than reaching for the gear shift,
GS, which is intuitive as well. The benefit of employing the joint discomfort function in
the posture prediction is that it is not necessarily intuitive that reaching for the glove box
is more comfortable than reaching for the seatbelt with the right hand or that reaching for
the parking brake is less comfortable than reaching for the gear shift. By employing the
discomfort function and posture prediction in a virtual environment, a designer can gain
additional insight into the placement of driving controls so as to achieve an optimal cab
layout. The simulation can easily be applied to humans of various anthropometries by
simply changing the input link lengths in the posture prediction simulation.

5 Conclusions

This study examined the effectiveness of the joint discomfort human performance
measure as defined in Yang et al. (2004) and determined that the function as originally
proposed did not accurately demonstrate the desired characteristics of joint discomfort.
Changes to the joint discomfort human performance measure were proposed, and through
analysis of the new joint discomfort function, it was shown that parameter values of
G = 50, α = 100, β = 5, and γ = 1, 100, 200 demonstrated the desired characteristics of
joint discomfort for digital human models and posture prediction. Furthermore, a series of
analyses examined the effect of the parameter values on the value of the joint discomfort
function. The effectiveness of the new joint discomfort human performance measure was
shown through a series of seven reach tasks inside a vehicle. The new joint discomfort
human performance measure exhibited each of the three aspects of joint discomfort as
defined in Yang et al. (2004).
Discomfort can be categorised in many ways, including objective and subjective
methods. While useful in an objective quantification of discomfort, the proposed joint
discomfort human performance measure provides nothing in the way of a subjective
quantification of discomfort or discomfort associated with fatigue. The joint discomfort
function, as proposed, is best employed in a MOO-based simulation in coordination with
other human performance measures such as joint torque, visual displacement, and delta
potential energy. In addition, the numerical value of joint discomfort has no specific
Joint discomfort human performance measure for driver posture prediction 273

meaning, rather the relative value of joint discomfort between multiple postures suggests
that the posture with the higher numerical value has more discomfort.
Another aspect of the joint discomfort human performance measure is the effect that
anthropometric differences in subjects have on the numerical value of joint discomfort.
Since the joint discomfort function is based on joint angles, link lengths and other
anthropometric statistics have no influence on the numerical value of discomfort. This is
both an advantage and a disadvantage. It is possible that anthropometrics have an
influence on discomfort and this joint discomfort function does not account for that.
Further research is necessary to determine how much of an influence anthropometry has
on discomfort. Along the same lines, it is unclear whether persons of various
anthropometries have a different set of preferred, or neutral, joint angles. For this study, it
was assumed that all subjects had the same neutral posture and therefore the same neutral
angles. Furthermore, research must be conducted on the effect that anthropometry has on
the joint weights in the joint discomfort human performance measure. For this study, it
was assumed that all subjects had the same joint weights regardless of anthropometry.
The new joint discomfort human performance measure can provide vehicle designers
with valuable insights into the placement of controls inside a vehicle. Through a series of
posture prediction simulations, an optimal cab layout can be determined that allows
drivers of various anthropometry to safely and comfortably reach all necessary controls.

Acknowledgements

This research work was partly supported by National Science Foundation (Award
#0926549), AT&T Chancellor’s Fellowship, Whitacre College of Engineering Dean’s
Fellowship, and the Honours College Undergraduate Research Fellowship, Texas Tech
University.

References
Basri, B. and Griffin, M. (2011) ‘The vibration of inclined backrests: perception and discomfort of
vibration applied normal to the back in the x-axis of the body’, Journal of Sound and
Vibration, Vol. 330, Nos. 18–19, pp.4646–4659.
Carvalho, D. and Callaghan, J. (2011) ‘Passive stiffness changes in the lumbar spine and effect of
gender during prolonged simulated driving’, International Journal of Industrial Ergonomics,
Vol. 41, No. 6, pp.617–624.
Cloutier, A., Gragg, J., Yang, J. and Alemayehu, F. (2012) ‘Sensitivity analysis of achieving reach
tasks within a vehicle considering driver joint angle variability’, SAE 2012 World Congress
and Exhibition, 24–26 April, Detroit, MI, USA.
Durkin, J., Harvey, A., Hughson, R. and Callaghan, J. (2006) ‘The effects of lumbar massage on
muscle fatigue, muscle oxygenation, low back discomfort, and driver performance during
prolonged driving’, Ergonomics, Vol. 49, No. 1, pp.28–44.
El Falou, W., Duchene, J., Grabisch, M., Hewson, D., Langeron, Y. and Lino, F. (2003)
‘Evaluation of driver discomfort during long-duration car driving’, Applied Ergonomics,
Vol. 34, No. 3, pp.249–255.
Gragg, J. and Yang, J. (2011) ‘Effect of obesity on seated posture inside a vehicle based on digital
human models’, SAE 2011 World Congress and Exhibition, 12–14 April, Detroit, MI, USA.
Gragg, J., Yang, J. and Howard, B. (2012) ‘Hybrid method for driver accommodation using
optimization-based digital human models’, Computer Aided Design, Vol. 44, No. 1, pp.29–39.
274 J. Gragg et al.

Gragg, J., Yang, J. and Long, J. (2011) ‘Optimisation-based approach for determining driver seat
adjustment range for vehicles’, International Journal of Vehicle Design, Vol. 57, Nos. 2/3,
pp.148–161.
Hermann, S. (2005) ‘Exploring sitting posture and discomfort using nonlinear analysis methods’,
IEEE Transactions on Information Technology in Biomedicine, Vol. 9, No. 3, pp.392–401.
Hermann, S. and Bubb, H. (2007) ‘Development of an objective measure to quantify automotive
discomfort over time’, IEEE International Symposium on Industrial Electronics, Art
No. 4375059, pp.2824–2830.
Jin, X., Cheng, B., Wang, B. and Shen, B. (2009) ‘Assessment of driver’s seating discomfort using
interface pressure distribution’, IEEE Intelligent Vehicles Symposium Proceedings, Art
No. 5164493, pp.1419–1424.
Jin, X., Cheng, B., Wang, B., Tao, X. and Shen, B. (2011) ‘Assessment of driver’s seating
discomfort using interface pressure distribution’, Proceedings – 3rd International Conference
on Measuring Technology and Mechatronics Automation, ICMTMA 2011, Art No. 5721019,
pp.72–75.
Jonsson, P. and Johansson, O. (2005) ‘Prediction of vehicle discomfort from transient vibrations’,
Journal of Sound and Vibration, Vol. 282, Nos. 3–5, pp.1043–1064.
Junoh, A., Nopiah, Z., Muhamad, W., Ghazali, N., Shakir, N., Nor, M. and Fouladi, M. (2011)
‘Linear programming model for optimizing of noise and vibration in passenger car cabin’,
ICCRD2011 – 3rd International Conference on Computer Research and Development, Art
No. 5763848, pp.35–40.
Kang, T. and Kaizu, Y. (2011) ‘Vibration analysis during grass harvesting according to ISO
vibration standards’, Computers and Electronics in Agriculture, Vol. 79, No. 2, pp.226–235.
Koizumi, T., Tsujiuchi, N., Okumura, S., Yamada, S., Ninomiya, J. and Tajima, T. (2011)
‘Suggestion for evaluation methods of ride comfort at low frequencies’, Conference
Proceedings of the Society for Experimental Mechanics Series, Vol. 3, No. 1, pp.135–142.
Kyung, G. and Nussbaum, M. (2008) ‘Driver sitting comfort and discomfort (part II): relationships
with and prediction from interface pressure’, International Journal of Industrial Ergonomics,
Vol. 38, Nos. 5–6, pp.526–538.
Kyung, G., Nussbaum, M. and Babski-Reeves, K. (2008) ‘Driver sitting comfort and discomfort
(part I): use of subjective ratings in discriminating car seats and correspondence among
ratings’, International Journal of Industrial Ergonomics, Vol. 38, Nos. 5–6, pp.516–525.
Maeda, S., Mansfield, N. and Shibata, N. (2008) ‘Evaluation of subjective responses to whole-body
vibration exposure: effect of frequency content’, International Journal of Industrial
Ergonomics, Vol. 38, Nos. 5–6, pp.509–515.
Majid, N., Notomi, M. and Rasmussen, J. (2011) ‘Musculoskeletal computational analysis of the
influence of car-seat design/adjustment on fatigue-induced driving’, 4th International
Conference on Modeling, Simulation, and Applied Optimization, ICMSAO 2011, 19–21 April,
2011, Kuala Lumpur, Malaysia.
Marler, R.T., Rahmatalla, S., Shanahan, M. and Abdel-Malek, K. (2005) ‘A new discomfort
function for optimization-based posture prediction’, SAE Human Modeling for Design and
Engineering Conference, June, Society of Automotive Engineers, Iowa City, IA, Warrendale,
PA.
Massaccesi, M., Pagnotta, A., Soccetti, A., Masali, M., Masiero, C. and Greco, F. (2003)
‘Investigation of work-related disorders in truck drivers using RULA method’, Applied
Ergonomics, Vol. 34, No. 4, pp.303–307.
Miettinen, K. (1999) Nonlinear Multiobjective Optimization, Kluwer Academic Publishers, Boston.
Na, S., Lim, S., Choi, H. and Chung, M. (2005) ‘Evaluation of driver’s discomfort and postural
change using dynamic body pressure distribution’, International Journal of Industrial
Ergonomics, Vol. 35, No. 12, pp.1085–1096.
Joint discomfort human performance measure for driver posture prediction 275

Ostensvik, T., Veiersted, K. and Nilsen, P. (2009) ‘A method to quantify frequency and duration of
sustained low-level muscle activity as a risk factor for musculoskeletal discomfort’, Journal of
Electromyography and Kinesiology, Vol. 19, No. 2, pp.283–294.
Porter, J., Gyi, D. and Tait, H. (2003) ‘Interface pressure data and the prediction of driver
discomfort in road trials’, Applied Ergonomics, Vol. 34, No. 3, pp.207–214.
Robb, M. and Mansfield, N. (2007) ‘Self-reported musculoskeletal problems amongst professional
truck drivers’, Ergonomics, Vol. 50, No. 6, pp.814–827.
Romero, C. (2000) ‘Bi-criteria utility functions: analytical considerations and implications in the
short-run labor market’, European Journal of Operations Research, Vol. 122, No. 1,
pp.91–100.
Schust, M., Bluthner, R. and Seidel, H. (2006) ‘Examination of perceptions (intensity, seat comfort,
effort) and reaction times (brake and accelerator) during low-frequency vibration in x- or
y-direction and biaxial (xy-) vibration of driver seats with activated and deactivated
suspension’, Journal of Sound and Vibration, Vol. 298, No. 3, pp.606–626.
Smith, D., Andrews, D. and Wawrow, P. (2006) ‘Development and evaluation of the automotive
seating discomfort questionnaire (ASDQ)’, International Journal of Industrial Ergonomics,
Vol. 36, No. 2, pp.141–149.
Szeto, G. and Lam, P. (2007) ‘Work-related musculoskeletal disorders in urban bus drivers of
Hong Kong’, Journal of Occupational Rehabilitation, Vol. 17, No. 2, pp.181–198.
Wang, X. and Trasbot, J. (2011) ‘Effects of target location, stature, and hand grip type on
in-vehicle reach discomfort’, Ergonomics, Vol. 54, No. 5, pp.466–476.
Wang, X., Pannetier, R., Burra, N. and Numa, J. (2011) ‘A biomechanical approach for evaluating
motion related discomfort: illustration by an application to pedal clutching movement’, 3rd
International Conference on Digital Human Modeling, ICDHM 2011, 9–14 July, Orlando, FL,
USA.
Yang, J., Marler, R., Kim, H., Arora, J. and Abdel-Malek, K. (2004) ‘Multi-objective optimization
for upper body posture prediction’, 10th AIAA/ISSMO Multidisciplinary Analysis and
Optimization Conference, 30 August–1 September 2004, Albany, New York, USA.
Yang, J., Verma, U., Marler, T., Beck, S., Rahmatalla, S. and Harrison, C. (2009) ‘Workspace zone
differentiation tool for visualization of seated postural comfort’, International Journal of
Industrial Ergonomics, Vol. 39, No. 1, pp.267–276.
Zschocke, A. and Albers, A. (2008) ‘Links between subjective and objective evaluations regarding
the steering character of automobiles’, International Journal of Automotive Technology,
Vol. 9, No. 4, pp.473–481.

View publication stats

Você também pode gostar