Content uploaded by Sumit Mandal
Author content
All content in this area was uploaded by Sumit Mandal on Oct 12, 2018
Content may be subject to copyright.
Original article
Modeling for predicting the thermal
protective and thermo-physiological
comfort performance of fabrics used
in firefighters’ clothing
Sumit Mandal , Simon Annaheim, Jemma Greve,
Martin Camenzind and Rene
´M Rossi
Abstract
Standardized test methods are available for measuring the thermal protective as well as thermo-physiological comfort
Performance of fabrics used in firefighters’ clothing. However, these tests are usually fabric destructive in nature, time
consuming, and/or expensive to carry out on a regular basis. Hence, the availability of empirical models could be useful
for conveniently predicting the thermal protective and thermo-physiological comfort performances from the fabric
properties. The aim of this study is to develop individual models for predicting thermal protective and thermo-
physiological comfort performances of fabrics. For this, different single- and multi-layered fabrics that are commercially
used to manufacture firefighters’ protective clothing were selected, and the fundamental properties of these fabrics
(weight, thickness, thermal resistance, air-permeability, evaporative resistance, and water spreading speed) were mea-
sured using the standard test methods developed by the International Organization for Standardization (ISO) or the
American Association of Textile Chemists and Colorists. The thermal protective performance of these fabrics was
measured by the ISO 9151:2016 test method under 80 kW/m
2
flame exposure. The thermo-physiological comfort
performance of fabrics was determined by the ISO 18640-1:2018 test method and a statistical model. Thereafter, the
key fabric properties affecting the thermal protective and thermo-physiological comfort performances of fabrics were
determined statistically. It has been found that thermal and evaporative resistances are the key fabric properties to affect
the thermal protective performance, whereas the fabric weight, evaporative resistance, and water spreading speed are
the key properties to affect the thermo-physiological comfort performance. By employing these key fabric properties,
Multiple Linear Regression and Artificial Neural Network (ANN) models were developed for predicting the thermal
protective and thermo-physiological comfort performances. Through a comparison of the predicting performance par-
ameters of these models, it has been found that ANN models can more accurately predict the performances of fabrics.
These models can be implemented in the textile industry and academia for effectively and conveniently predicting the
thermal protective and thermo-physiological comfort performances only by utilizing the key fabric properties.
Keywords
firefighters’ clothing, fabric properties, thermal protective performance, thermo-physiological comfort performance,
Multiple Linear Regression, Artificial Neural Network
Thermal protective clothing worn by firefighters pro-
vides them with protection from fire hazards while on
duty.
1,2
Fabrics used in this clothing are generally thick,
semi-air-permeable, and/or air-impermeable in order to
provide a high level of protection to firefighters from
different heat exposures.
2–4
As a consequence, these
fabrics impede the metabolic-heat, liquid-sweat,
Empa, Swiss Federal Laboratories for Materials Science and Technology,
Laboratory for Biomimetic Membranes and Textiles, St. Gallen,
Switzerland
Corresponding author:
Rene
´M Rossi, Empa, Swiss Federal Laboratories for Materials Science
and Technology, Lerchenfeldstrasse 5, St. Gallen, 9014, Switzerland.
Email: rene.rossi@empa.ch
Textile Research Journal
0(00) 1–14
!The Author(s) 2018
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/0040517518803779
journals.sagepub.com/home/trj
and/or sweat-vapor transfer from firefighters’ bodies to
their ambient environment, which ultimately lowers the
thermo-physiological comfort of firefighters and
increases their heat strain.
3
Hence, it is important to
experimentally measure both of these contradictory
aspects thermal protective and thermo-physiological
comfort performances of the fabrics – in order to
understand the interrelationship between them.
5,6
For this purpose, various test methods have been
developed by the International Organization for
Standardization (ISO).
7–11
In particular, the ISO 6942:2015 and ISO 9151:2016
test standards are used for measuring the thermal pro-
tective performance under radiant-heat and flame
exposures, respectively.
7,8
The ISO 6942:2015 standard
recommends to measure the protective performance
under low (5–10 kW/m
2
) and medium (10–40 kW/m
2
)
intensity radiant-heat exposures, whereas the ISO
9151:2016 standard recommends to measure the per-
formance under high-intensity (80 kW/m
2
) flame expos-
ure only. Moreover, the ISO 11092:2014 and ISO
18640-1:2018 test standards are widely used for measur-
ing the thermo-physiological comfort performance of
fabrics using a sweating guarded hot plate and torso
testers, respectively.
10,11
The ISO 11092:2014 standard
mainly measures the thermal and evaporative resist-
ances of the fabrics to understand their resistance
toward the metabolic-heat and sweat-vapor transfer
from firefighters’ bodies to the ambient environment.
However, the individual measurement of the thermal
and evaporative resistances is unrealistic because the
metabolic-heat, liquid-sweat, and/or sweat-vapor trans-
fer occur simultaneously from the human body to the
ambient environment and are coupled by phase change
mechanisms, such as liquid evaporation or vapor con-
densation.
12–14
Considering this, devices like a sweating
guarded torso test device that can holistically meas-
ure the thermo-physiological comfort performance
of the fabrics by considering the combined effect of
metabolic-heat and sweat-vapor transfer through the
fabric have been developed and documented in the
ISO 18640-1:2018 standard.
11
Although the above-mentioned test standards has
been used to measure the thermal protective and
thermo-physiological comfort performances of fabrics,
these tests are usually fabric destructive in nature, time
consuming, and/or expensive to carry out on a regular
basis.
15–19
Considering this, some researchers have
developed heat transfer models by studying the math-
ematical interactions between fabric properties and per-
formances.
20–22
However, the practical application of
these mathematical heat transfer models for predicting
the performances may be limited due to their complex-
ities. Therefore, a few researchers have developed
empirical Multiple Linear Regression (MLR) and/or
Artificial Neural Network (ANN) models to predict
the thermal protective performance from a set of
fabric properties (these fabric properties significantly
affected the performance) under flame, radiant-heat,
and/or hot surface contact exposures. In addition, it
has been found that ANN models can more accurately
predict the protective performance in comparison to
MLR models.
17,18,23
Although these researchers have
developed the individual models for predicting the pro-
tective performance under each exposure, the universal
application of these models together for different expos-
ures highly increases the complexity of the prediction,
comparability, and/or holistic understanding of the
thermal protective performance of fabrics. Hence, an
ANN model is required to measure and holistically
understand the thermal protective performance of fab-
rics under an exposure and intensity that simulates the
real working scenario of firefighters. Furthermore, no
models are available to date for conveniently predicting
the thermo-physiological comfort performance of
fabrics.
In this study, the thermal protective performance of
a set of fabrics was measured under flame and radiant-
heat exposures at different intensities using standar-
dized test methods; in addition, the thermo-physiologi-
cal comfort performance of the fabrics was measured
using the standard sweating guarded torso test and a
statistical model. As a first objective of this study, it was
necessary to identify a standard representative test
method with a particular exposure and intensity that
can holistically measure and classify the protective per-
formance of the fabrics for further modeling. Next, the
key fabric properties that affect the thermal protective
and thermo-physiological comfort performances were
identified. By employing these key fabric properties,
MLR and ANN models were developed to predict the
thermal protective and thermo-physiological comfort
performances. Finally, two better-fitting high-
performance models were identified for predicting the
thermal protective as well as the thermo-physiological
comfort performances.
Methods
Fabric selection and property measurement
For this study, six single- and 13 multi-layered fabrics
which were developed by conventional (weaving, finish-
ing) and/or latest technology (nano nonwoven) were
selected (Table 1). These fabrics are commercially used
in firefighters’ station uniforms and turnout gear cloth-
ing. The fundamental properties of these fabrics
(weight, thickness, thermal resistance, air-permeability,
evaporative resistance, and water spreading speed) were
measured using standard test methods developed by the
2Textile Research Journal 0(00)
ISO or the AATCC (American Association of Textile
Chemists and Colorists) (Table 2).
Selection of a representative standard test
method for measuring the thermal protective
performance of fabrics
The thermal protective performance of the fabrics was
measured under radiant-heat exposures at 10 and
40 kW/m
2
using the ISO 6942:2015 standard.
7
In add-
ition, the protective performance was measured under
the flame exposure at 80 kW/m
2
using the ISO
9151:2016 standard.
8
For both the methods, a copper
sensor was directly placed behind the back side of a
fabric specimen (i.e., toward the side of the specimen
that is in contact with wearers’ skin) and the front
side of the specimen (i.e., toward the side of the speci-
men that faces the fire hazard) was exposed to the
Table 1. Selected fabrics [Outer Layer (OL) faces thermal exposures; Middle Layer (ML) is sandwiched between OL and IL; and
Inner Layer (IL) is in contact with wearers’ skin]
Fabrics Composition and/or construction
Single-layered 1 50% Meta-aramid/50% Fire Retardant (FR) viscose woven fabric
2 34% Meta-aramid and Para-aramid/33% Lyocell/31% Modacrylic/2% Antistatic fibers
3 93% Meta-aramid/5% Para-aramid/2% Antistatic Fibers
4 93% Meta-aramid/5% Para-aramid/2% Antistatic Fibers
5 55% FR modacrylic/45% FR cotton woven fabric
6 FR cotton woven fabric
Multi-layered 7 Double woven fabric with meta-aramid face and para-aramid back (OL) + Polytetrafluorethylene (PTFE)
coated membrane on a aramid nonwoven fabric with dots toward IL (ML) + 50% meta-aramid/50% FR
viscose (IL) woven fabric (IL)
8 Double woven fabric with meta-aramid face and para-aramid back (OL) + PTFE coated membrane on two
aramid fabrics with dots toward OL (IL)
9 Double woven fabric with meta-aramid face and para-aramid back (OL) + PTFE coated membrane on a
aramid nonwoven fabric (ML) + Aramid regenerated nonwoven felt sewn with 50% meta-aramid/50%
FR viscose woven fabric (IL)
10 99% aramid/1% beltron woven fabric (OL) + PTFE coated on meta-aramid nonwoven fabric
(ML
layer1
) + Aramid spunlace fleece (ML
layer2
) + 50% meta-aramid/50% viscose woven fabric (IL)
11 Meta-aramid woven fabric with thread on backside (OL) + Polyurethane (PU) liner on meta-aramid non-
woven fabric (ML) + Aramid woven fabric (IL)
12 64% FR viscose/35% meta-aramid/1% antistatic woven fabric (OL) + PU coated on 50% meta-aramid/50%
FR viscose nonwoven fabric (ML) + 65% FR viscose/35% meta-aramid fabric (IL)
13 64% FR viscose/35% meta-aramid/1% antistatic woven fabric (OL) + PTFE coated on 50% meta-aramid/
50% FR viscose nonwoven fabric (ML) + 65% FR viscose/35% meta-aramid woven fabric (IL)
14 Meta-aramid woven fabric (OL) + PTFE coated on 25% meta-aramid/25% para-aramid/50% basofil non-
woven fabric (ML) + Nonwoven meta-aramid quilted with 50% meta-aramid/50 FR viscose fabric (IL)
15 75% meta-aramid/23% para-aramid/2% antistatic woven fabric (OL) + PTFE coated membrane on a aramid
nonwoven fabric (ML
layer1
) + Meta-aramid nonwoven fabric (ML
layer2
) + 93% meta-aramid/5% para-
aramid/2% antistatic woven fabric (IL)
16 75% meta-aramid/23% para-aramid/2% antistatic woven fabric (OL) + PTFE coated membrane on a aramid
nonwoven fabric (ML
layer1
) + Meta-aramid nonwoven fabric (ML
layer2
) + Meta-aramid nano nonwoven
fabric (ML
layer3
) + 93% meta-aramid/5% para-aramid/2% antistatic woven fabric (IL)
17 Filament Twill Technology-based meta-aramid spun yarn and para-aramid filament yarn woven fabric
(OL) + PTFE coated membrane on two aramid nonwoven fabrics (ML) + PTFE coated membrane on a
aramid woven fabric (IL)
18 Filament Twill Technology based meta-aramid spun yarn and para-aramid filament yarn woven fabric
(OL) + PTFE coated membrane on a aramid nonwoven fabric (ML
layer1
) + Meta-aramid nonwoven fabric
(ML
layer2
) + Meta-aramid nano nonwoven fabric (ML
layer3
) + 93% meta-aramid/5% para-aramid/2% anti-
static woven fabric (IL)
19 Filament Twill Technology-based meta-aramid spun yarn and para-aramid filament yarn woven fabric
(OL) + PTFE membrane on a aramid nonwoven fabric (ML
layer1
) + Meta-aramid nonwoven fabric
(ML
layer2
) + 93% meta-aramid/5% para-aramid/2% antistatic woven fabric (IL)
Mandal et al. 3
radiant-heat (generated from a radiant-heat panel) or
flame (generated from a Meker burner). During the
exposure, the time (in seconds) was required to increase
(from the starting temperature of the sensor at
25–28C) the temperature of the sensor by 24C
(T24) was measured through software. Notably, T24
values of three specimens of each fabric were measured
and then averaged. This averaged T24 value was
interpreted as the thermal protective performance of
the fabric. Through statistical comparison of the
T24 values under different exposures and intensities, a
representative standard test method (at a particular
exposure and intensity that can simulate the real
working scenario of firefighters) was identified for the
holistic measurement and further modeling of the
protective performance of fabrics. As described in
earlier research,
24,25
a fabric with a high T24 value
was inferred as the high thermal protective perfor-
mance-based fabric in comparison to a fabric with
a low T24.
Thermo-physiological comfort performance
measurement of fabrics
The thermo-physiological comfort performance of the
fabrics was measured using the ISO 18640-1:2018
standard test method
11
and a statistical model provided
by Annaheim et al. in 2016
26
(i.e., Equation (1)). This
method includes an upright standing heated cylinder
with a surface area of 0.43 m
2
, representing a human
trunk/torso. The upper and lower sides of the torso are
guarded in order to prevent any heat loss from it.
Furthermore, perspiration is imitated by the release of
deionized water through 54 evenly distributed nozzles
on the cylinder surface. A fabric specimen was clipped
around the sweating guarded torso (the surface tem-
perature of the torso was kept at 35C) that was
placed in a climatic chamber with controlled air tem-
perature (20 0.5C), relative humidity (50 5%), and
air velocity (1 0.1 m/s). During the test, the thermal
resistance of the fabric (R
ct
in m
2
.K/W) was measured
Table 2. Fundamental properties of the selected fabrics
Fabrics
Properties
Weight
a
(g/m
2
)
Thickness
b
(mm)
Thermal
resistance – R
ct c
(Km
2
/W.10
3
)
Air-permeability
d
(cm
3
/cm
2
/s)
Evaporative
resistance – R
ete
(m
2
Pa/W)
Water (sweat)
spreading speed
f
(mm/s)
1 197.0 0.4 12.9 148.6 3.0 2.8
2 243.7 0.6 8.9 90.0 2.5 4.3
3 154.7 0.3 9.7 502.2 2.4 6.7
4 229.8 0.4 9.4 91.4 3.3 6.8
5 367.3 0.7 13.0 44.3 3.6 0.5
6 366.8 0.8 12.9 43.0 4.7 0.8
7 609.9 3.9 79.7 0 20.2 0.8
8 547.5 3.9 80.0 0 15.8 0.5
9 673.8 4.9 127.7 0 25.4 0.3
10 635.2 3.2 82.7 0 15.6 3.3
11 592.1 3.9 95.4 0 23.9 1.9
12 587.8 2.1 46.6 0 9.4 0.6
13 599.8 2.3 49.3 0 10.0 0.5
14 493.0 2.3 71.0 0 16.8 4.3
15 520.5 2.1 61.8 1.0 12.2 2.6
16 514.8 2.2 65.6 0.8 13.0 3.4
17 568.1 3.0 83.3 0 12.8 3.2
18 496.7 2.1 64.2 0.8 14.2 3.0
19 490.4 1.9 60.0 1.1 13.2 2.3
a
Measured by ISO 3801:1977.
b
Measured by ISO 5084:1996.
c
Measured by ISO 11092:2014.
d
Measured by ISO 9237:1995.
e
Measured by ISO 11092:2014.
f
Water spreading speed of the fabric layer in contact with wearers’ skin was measured by AATCC 195.
4Textile Research Journal 0(00)
during a dry phase (without sweating) and the initial
cooling rate (IC in C/h) was obtained during an activ-
ity phase consisting of a metabolic-heat production of
125 W and a sweat rate of 100 g/h [corresponding to a
human physical activity of 5 Mets (290 W/m
2
), and a
sweat rate of 230 g/m
2
/h]. Based on R
ct
and IC, com-
parative Time to Heat Stress (cTHEST in minutes) was
calculated for fabrics (Equation (1)). Notably, cTHEST
values of three specimens of each fabric were calculated
and then averaged. This averaged cTHEST value was
interpreted as an indicator for wearing comfort and
thermo-physiological impact of the fabric. Here, a
high value of cTHEST indicates a high thermo-
physiological comfort performance for the fabric.
26
Furthermore, cTHEST values have been validated
based on human subject trials including similar physical
activity as applied during the activity phase of the
sweating guarded torso methodology (induced by walk-
ing on a treadmill) in environmental conditions of 40C
air temperature and 30% relative humidity.
26
Time to
reach a core body temperature of 38.5C, starting from
thermo-neutral status (i.e., core body temperature of
37.0 0.5C), was compared to the cTHEST of four
fabrics. Contextually, a significant linear relationship
has been detected [the coefficient of determination
(R
2
) is 0.84]; thus, Annaheim et al.
26
concluded a
powerful relevance of the cTHEST for the assessment
of thermo-physiological comfort performance of
fabrics
Time to exceed core body temperature of 38:5C
or cTHEST ¼0:68 IC 0:067 Rct þ119:39
ð1Þ
Modeling approaches for predicting the thermal
protective and thermo-physiological comfort
performances of fabrics
Empirical modeling techniques were used to predict the
thermal protective and thermo-physiological comfort
performances of fabrics. Different fabric properties
were used as input parameters for predicting the ther-
mal protective or thermo-physiological comfort per-
formance. Contextually, it is notable that too many
and/or mutually dependent input variables (e.g.,
fabric properties) could disturb the empirical models
and lead to the error in predicting the output variable
(e.g., thermal protective performance, thermo-
physiological comfort performance). Therefore, key
fabric properties for protective and comfort perform-
ances were identified and used as input variables in the
empirical models. In this study, MLR and ANN mod-
eling methodologies were used for predicting the
performance.
17,23
These models were compared based
on their predicting performance parameters [coefficient
of determination (R
2
), Root Mean Square Error
(RMSE), and p-value] to identify the better-fitting
high-performance models to predict the protective and
comfort performances. A model with high R
2
,low
RMSE, and p-values of <0.05 was inferred as the
better-fitting high-performance model. The following
sections present the procedure to identify the key
fabric properties affecting the performances and the
modeling methodologies of the MLR and ANN models.
Procedure to identify the key fabric properties affecting
the thermal protective and thermo-physiological comfort
performances of fabrics. To identify the key fabric proper-
ties, linear regression t-tests were conducted between
the individual fabric properties (weight, thickness, ther-
mal resistance, air-permeability, evaporative resistance,
or water spreading speed) and performance (thermal
protective or thermo-physiological comfort perform-
ance) using the SPSS Statistics 23 Data Editor devel-
oped by IBM Corporation, USA. The p-values
obtained from the t-tests were analyzed to identify the
fabric properties that significantly affected the perform-
ance. Significance tests were carried out at the signifi-
cance level of 0.05. Contextually, it is notable that the
set of significant fabric properties identified could also
be mutually dependent. To identify the most significant
property, the correlation coefficient (r) was calculated
between each mutually dependent fabric property and
the performance. The fabric property with the highest
rvalue was identified as the most significant fabric
property among all the mutually dependent fabric
properties. The significant fabric properties identified
based on the p-values and/or rvalues were inferred as
the key fabric properties affecting the performance. The
‘+’ or ‘–’ sign of the T-stat value obtained from the
t-test indicated the positive or negative association
between a key fabric property and the performance,
respectively. This association was further justified
based on the theories of textile science (e.g., fabric ther-
mal insulation, liquid spreading speed) as well as heat
and mass transfer (e.g., convective or radiative heat
transfer, sweat-vapor/liquid transfer) through porous
fabrics.
MLR and ANN modeling methodologies
MLR modeling. MLR modeling was used to predict
the thermal protective or thermo-physiological comfort
performances from the key fabric properties (obtained
from the previous section) using the IBM SPSS
Statistics 23 software. The generic form of MLR
models used is shown in Equation (2), where, Cis the
identically distributed constant normal error,
1
...
n
are the regression coefficients that determine relative
Mandal et al. 5
strength of the respective key fabric properties, and
(KFP)
1
...(KFP)
n
are key fabric properties. Notable
inherent limitations of the MLR model are that, firstly,
it assumes the linear relationship between all input (key
fabric properties) and output variables (thermal pro-
tective or thermo-physiological performance), which
may not always be the case; and, secondly, this model
should not be used to predict an output variable
beyond the range of the input variables employed in
the model
17,19,27–30
Performance ¼Cþ1ðKFPÞ1þ2ðKFPÞ2
þ...nðKFPÞn
ð2Þ
ANN modeling. The ANN is a powerful data model-
ing tool that could capture and represent any kind of
relationship between input (key fabric properties) and
output (thermal protective or thermo-physiological
comfort performance) variables.
23,27–32
Two ANN
models were developed for predicting the thermal pro-
tective and thermo-physiological comfort performances
of fabrics using the MATLABÕR2015b software.
In this study, Multi Layer Perceptron (MLP) archi-
tecture was employed, as MLPs are universal function
approximators and commonly used to create mathem-
atical models by regression analysis. After setting dif-
ferent values for hyper-parameters (e.g., the number of
hidden layers and neurons, choice of activation func-
tions) and different training algorithms (e.g., gradient
descent, Levenberg–Marquardt), a three-layered (input
layer, one hidden layer, and output layer) feed-forward
(one of the specialized versions of feed-forward net-
work, i.e., fitnet) back-propagation (Levenberg–
Marquardt back-propagation method) ANN model
was employed (Figure 1). Generally, a challenge in
using the feed-forward back-propagation ANN model
is to decide the number of neurons in the hidden layer.
If the neurons are too few in the hidden layer, the
model is usually unable to differentiate between com-
plex patterns, and it might lead to a linear estimate of
the actual relationship between the input and output
variables, whereas if the neurons are too many, the
model follows a noise in the data set, and it might lead
to an inaccurate output.
29
To choose the optimum
number of neurons in the hidden layer, the ANN
models were trained with different numbers of neurons,
and the best predictive ANN models were found with 10
hidden neurons. Different measures and criteria such
as performance plot, error histogram, and regression
plot were consulted for deciding that the trained
model is not over fitted or the trained model has general-
ized well. After a model was trained, we looked at the
performance plot, which showed a decrease in Mean
Square Error (MSE) as the model became trained.
In this three-layered feed-forward model, each layer
of the neural network contained connections to the next
layer (e.g., from the input to the hidden layer), but there
were no connections back. All the neurons in a particu-
lar layer received a signal from the neurons of the pre-
vious layer. The signal received was then multiplied by
a weight factor known as a synaptic weight (w). Next,
the weighted inputs were summed up and passed to a
transfer function to generate the output in a fixed range
of values. This output was then transferred to the neu-
rons of the next layer. As the model used back-propa-
gation supervised training, the final outputs predicted
were always compared with the actual output. Through
this comparison, the back-propagation training algo-
rithm calculated the prediction error and adjusted the
synaptic weight of layers backward from the output to
the input layer. Eventually, the error signal decreased
iteratively and the model got closer to producing the
desired final output. The hyperbolic tangent sigmoid
transfer function (Equation (3)) was assigned as an acti-
vation function in the hidden layer, and the linear func-
tion (Equation (4)) was used in the output layer. These
specific functions can easily be applied with all types of
data and provide the best performance for an ANN
model.
27
In Equations (3) and (4), xis the weighted
sum of inputs to a neuron and f(x) is the transformed
output from that neuron.
In this study, MATLAB randomly assigned 70% of
data (i.e., significant fabric properties and performance)
for the training, 15% of data for the validation, and the
remaining 15% of data to test the predicting perform-
ance of the ANN models. Contextually, it is notable
that these ANN models were trained on a small
data set. Thus, these models might be unstable and
may not be generalized for predicting the thermal
Input Layer
Hidden Layer
Output Layer
Input Neurons
(Key Fabric Properties)
Hidden Neurons
Output Neuron
(Thermal Protective or Thermo-
physiological Comfort Performance)
w1,1
w1,2
w4,10
w4,9
w1,1
w10,1
Figure 1. Schematic diagram of the three-layered feed-forward
back-propagation Artificial Neural Network model with 10
hidden neurons (as an example: four input neurons, 10 hidden
neurons, and one output neuron are used).
6Textile Research Journal 0(00)
protective or thermo-physiological comfort perform-
ance of all types of fabrics.
fxðÞ¼
sinhx
coshx ¼exex
exþex¼e2x1
e2xþ1ð3Þ
fxðÞ¼xð4Þ
Results and discussion
Thermal protective performances (T24) of the fabrics
under flame and radiant-heat exposures at different
intensities are shown in Table 3, and the relationships
between the thermal protective performances for these
exposures and intensities are shown in Figure 2. The
coefficient of determination (R
2
)>0.9 indicated a high
interrelation between the thermal protective perform-
ances predicted at these exposures and intensities.
This linear correlation suggests that the heat absorption
and degradation energy generated by the samples are
comparable for all of these exposures and intensities.
The values obtained at 80 kW/m
2
flame exposure test
were selected as a representative standard test method
for measuring, classifying, and modeling the thermal
protective performance of fabrics in this study. This is
because the flame test can holistically simulate the com-
bined effect of flame, radiant-heat, and hot gasses
instead of considering the radiant-heat exposure only;
also, firefighters could be at a greater risk when exposed
to high-intensity heat.
33–35
In addition, Table 3 presents
the thermo-physiological comfort performances
(cTHEST) of the fabrics. In this study, an inverse rela-
tionship was detected for T24 and cTHEST (Table 3).
This outcome was expected as fabrics with high T24
values or high protective performance generally allow
limited metabolic-heat exchange, which results in low
cTHEST values or low comfort performance.
Models for predicting the thermal protective and
thermo-physiological comfort performances of fabrics
In order to develop the models, firstly, the key fabric
properties affecting the thermal protective and thermo-
physiological comfort performances are identified and
tabulated (see the Key fabric properties affecting the
thermal protective and thermo-physiological comfort
Table 3. Thermal protective and thermo-physiological comfort performances of fabrics
Fabrics
Thermal protective performance
Thermo-physiological
comfort performance
Radiant-heat test Flame test
sweating guarded torso test and
statistical model (Equation (1))
T24 at
10 kW/m
2
(s)
T24 at
40 kW/m
2
(s)
T24 at
80 kW/m
2
(s)
cTHEST
(min)
1 24.4 5.5 4.3 124.4
2 23.6 7.1 4.5 123.6
3 22.1 6.3 3.6 128.0
4 24.4 6.9 4.0 128.0
5 27.1 7.8 5.9 119.2
6 26.6 7.1 5.7 121.0
7 64.2 23.2 17.7 118.5
8 46.1 19.2 14.9 115.5
9 81.8 39.1 24.6 114.0
10 54.9 18.8 16.0 116.2
11 60.0 21.4 19.7 114.8
12 52.9 18.8 14.5 116.7
13 47.7 16.1 14.1 115.6
14 45.9 16.6 14.6 119.3
15 44.4 15.0 14.2 122.2
16 45.5 15.8 15.2 118.8
17 66.1 22.1 16.0 116.7
18 52.4 17.3 13.4 118.0
19 50.6 15.8 12.7 121.7
Mandal et al. 7
performances of fabrics section). By using these key
fabric properties, the MLR and ANN models are devel-
oped for predicting the thermal protective and thermo-
physiological comfort performances (see the MLR and
ANN models for predicting the thermal protective and
thermo-physiological comfort performances of fabrics
section). By comparing these MLR and ANN models,
two better-fitting high-performance models are inferred
for predicting the thermal protective and thermo-phy-
siological comfort performances.
Key fabric properties affecting the thermal protective and
thermo-physiological comfort performances of fabrics. The
results obtained from statistical analysis to select
key fabric properties are presented in Table 4. As per
Table 4, weight, thickness, and thermal resistances were
found to significantly affect both the thermal protective
R² = 0.9315
0
10
20
30
40
50
0 102030405060708090
Radiant-heat
(40 kW/m2)
Radiant-heat (10 kW/m2)
R² = 0.9011
0
10
20
30
40
0 5 10 15 20 25 30 35 40 45
Flame
(80 kW/m2)
Radiant-heat (40 kW/m2)
R² = 0.9248
0
10
20
30
0 102030405060708090
Flame
(80 kW/m2)
Radiant-heat (10 kW/m2)
y = 0.47x – 5.28
y = 0.70x + 1.34
y = 0.34x – 3.09
Figure 2. Relationships between the thermal protective performances under flame and radiant-heat exposures at different
intensities.
Table 4. Results of the t-tests
Fabric
properties
Performance
Thermal protective
Thermo-physiological
comfort
p-value T-stat p-value T-stat
Weight 0.0001 10.96 0.0001 8.50
Thickness 0.0001 6.42 0.001 4.16
Thermal
resistance
0.0001 18.14 0.0001 5.12
Air-permeability 0.006 3.11 0.001 3.94
Evaporative
resistance
0.0001 12.83 0.0001 4.62
Water spreading
speed
0.05 2.31 0.0001 4.49
8Textile Research Journal 0(00)
and thermo-physiological comfort performances,
because their p-values are lower than 0.05. As these
three fabric properties are mutually dependent,
36,37
the correlation coefficient (r) between each of these fab-
ric properties and the performances was calculated in
order to identify the most significant parameter (see
Set 1 in Table 5). It was found that thermal resistance
(r¼0.98) and weight (r¼0.90) were the most
significant properties for thermal protective and
thermo-physiological comfort performances, respect-
ively. Here, the positive (+) T-stat value revealed a
positive/direct correlation between thermal resistance
and thermal protective performance, whereas the nega-
tive ()T-stat value revealed a negative/indirect rela-
tionship between weight and thermo-physiological
comfort performance (Table 4). Actually, a fabric
with high thermal resistance traps a lot of dead air
within its structure and/or possesses high thickness.
This situation lowers the convective, radiative, and/or
conductive heat transfer through the fabrics and
enhances the thermal protective performances of the
fabrics.
6,38
In this context, Schmid et al.
39
also found
that thermal resistance is a key fabric property to affect
the thermal protective performance, especially under
flash fire exposure. In addition, a fabric with high
weight could exert a physical burden on firefighters. A
high-weight fabric might be thicker as well, and that
impedes the metabolic-heat and sweat-vapor transfer
from firefighters’ bodies to the ambient environment.
This situation could lower the thermo-physiological
comfort performance of fabrics.
40
From Table 4, it is also notable that two mutually
dependent fabric properties, such as air-permeability
and evaporative resistance (generally a fabric with
high air-permeability results in low evaporative resist-
ance) significantly (p-value <0.05) affect the thermal
protective and thermo-physiological comfort perform-
ances. However, the evaporative resistance can most
significantly affect the performances, based on the high-
est rvalues (rfor thermal protective perform-
ance ¼0.95; rfor thermo-physiological comfort
performance ¼0.75) in Set 2 of Table 5. Furthermore,
it is clear from Table 4 that evaporative resistance is
directly (‘+’ T-stat value) and indirectly (‘’T-stat
value) related to the thermal protective and thermo-
physiological comfort performances of the fabrics,
respectively. Actually, a fabric with high evaporative
resistance usually possesses fewer pores within its struc-
ture. This situation lowers the transmission of convect-
ive heat through the pores of the fabric and enhances
the thermal protective performance of the fabrics.
Contradictorily, a fabric with fewer pores might have
more conductive heat transfer, depending upon the
properties of the fibers used in the fabric, and this
could result in low protective performance.
38,41
Furthermore, evaporative resistance could restrict the
evaporative heat transfer from firefighters’ bodies to the
ambient environment. Eventually, a fabric with high
evaporative resistance could restrict the metabolic-
heat and sweat-vapor transfer from firefighters’ bodies
to the ambient environment. This situation ultimately
lowers the comfort performance of the fabrics.
As this study focused on the thermal protective per-
formance of only dry fabrics, it is clear from Table 4
that water spreading speed is not a significant property
in this case. However, this property is significant for the
thermo-physiological comfort performance of fabrics.
Basically, a fabric (i.e., applied in the sweating guarded
torso methodology) with high water spreading speed
could quickly distribute the liquid-sweat within the
fabric layer next to wearers’ skin or other fabric
layers, and then it could easily evaporate. This situation
results in the enhanced thermo-physiological comfort
performance of the fabrics.
Overall, Table 6 summaries the key fabric properties
identified for the thermal protective and thermo-phy-
siological comfort performances based on the p-values
(<0.05) and/or rvalues (the highest rvalues for a par-
ticular set of fabrics in Table 5).
MLR and ANN models for predicting the thermal protective and
thermo-physiological comfort performances of fabrics
MLR models. By implementing the key fabric proper-
ties (Table 6) in the MLR modeling methodology
described in the MLR and ANN modeling methodologies
section, the models developed for predicting the ther-
mal protective and thermo-physiological comfort per-
formances are presented in Equations (5) and (6),
Table 5. Correlation coefficient of mutually dependent fabric
properties
Mutually dependent fabric
properties
Correlation coefficient (r)
Thermal
protective
Thermo-
physiological
comfort
Set 1
Weight 0.94 –0.90
Thickness 0.84 –0.71
Thermal resistance 0.98 –0.78
Set 2
Air-permeability –0.61 0.70
Evaporative resistance 0.95 –0.75
Mandal et al. 9
respectively, where R
ct
is thermal resistance, R
et
is evap-
orative resistance, Wis weight, and WSS is water
spreading speed
Thermal Protective Performance
¼3:13 þ0:14 Rct þ0:14 Ret
ð5Þ
Thermo physiological Comfort Performance
¼126:32 0:02 Wþ0:64 WSS 0:05 Ret
ð6Þ
ANN models. For both thermal protective and com-
fort performance, ANN models were developed based
on the key fabric properties (Table 6). The coding of
these software programs for ANN models is presented
in Appendices 1 and 2.
Comparison between MLR and ANN models. The
prediction performance of the MLR and ANN
models developed is presented in Table 7. Both predic-
tion model approaches were found to provide statistic-
ally significant results. However, the ANN models
reached higher coefficient of determination (R
2
) values
than the MLR models for both the prediction of ther-
mal protective and thermo-physiological comfort per-
formances. Moreover, the prediction errors (RMSE) of
the ANN models are lower than that of the MLR
models. In summary, the ANN models performed
better than the MLR models for predicting the thermal
protective and thermo-physiological comfort perform-
ances in terms of the precision and accuracy. As ANN
models can capture and represent any kind of relation-
ship between the key fabric properties and perform-
ances, they outperformed the MLR models.
Nevertheless, the precision and accuracy of the
ANN model for predicting the thermal protective
performance is higher than that of the ANN model
for predicting the thermo-physiological comfort per-
formance (by comparing the R
2
and RMSE values of
the ANN models of thermal protective and thermo-
physiological comfort performances). A possible
reason for this difference in precision and accuracy
could be related to heat and/or mass (sweat or mois-
ture) transfer phenomena occurring through the fabrics
while evaluating the thermal protective and thermo-
physiological comfort performances of fabrics (using
the methods described in the Thermo-physiological com-
fort performance measurement of fabrics and Selection
of a representative standard test method for measuring
the thermal protective performance of fabrics sections).
While combined and complex heat and mass transfer
(liquid-sweat as well as sweat-moisture-vapor) phenom-
enon occurs through the fabrics in the case of thermo-
physiological comfort performance evaluation a more
simplified phenomenon of heat transfer only is applied
in the case of thermal protective performance evalu-
ation. Due to these differences in heat and/mass trans-
fer phenomena, different numbers and types of key
fabric properties affect these two performances
(Table 6). This difference in the numbers and types of
key fabric properties leads to the difference in precision
and accuracy of the ANN models developed for the
protective and comfort performances. As compara-
tively fewer key fabric properties were involved in the
development of ANN model for thermal protective per-
formance, this model could generate fewer errors while
predicting the protective performance. Finally, the pre-
cision and accuracy of the ANN model for thermal
protective performance is higher than the ANN
model for thermo-physiological comfort performance.
Summary and conclusion
The standardized test methods available for measuring
the thermal protective and thermo-physiological com-
fort performances are usually fabric destructive in
nature, time consuming, and/or expensive to carry out
Table 6. Key fabric properties for the thermal protective and
thermo-physiological comfort performances (3¼key fabric
properties, ¼non-key fabric properties)
Fabric properties
Key fabric properties for the
performance
Thermal
protective
Thermo-physiological
comfort
Weight 3
Thickness
Air-permeability
Water spreading speed 3
Thermal resistance 3
Evaporative resistance 33
Table 7. The R
2
, Root Mean Square Error (RMSE), and p-values
of the Multiple Linear Regression (MLR) and Artificial Neural
Network (ANN) models
Predicting
performance
parameters of
models
Models
Thermal protective
performance
Thermo-physiological
comfort performance
MLR ANN MLR ANN
R
2
0.95 0.99 0.86 0.94
RMSE 1.38 0.71 1.70 1.53
p-value 0.0001 0.0001
10 Textile Research Journal 0(00)
on a regular basis. Considering this, the present study
aimed at developing empirical models for conveniently
predicting the protective and comfort performances
from fabric properties. For this, properties and per-
formances of a set of fabrics used in firefighters’ cloth-
ing were measured using the standardized test methods.
Thermal protective performance was measured in terms
of time to increase the wearers’ skin temperature by
24C under different heat exposures (radiant-heat and
flame) and intensities faced by firefighters in a fire
hazard. In addition, the thermo-physiological comfort
performance of fabrics was measured in terms of com-
parative time to generate heat stress on wearers.
It has been found that the thermal protective per-
formances of fabrics under exposures of 10 kW/m
2
radi-
ant-heat, 40 kW/m
2
radiant-heat, and 80 kW/m
2
flame
are linearly correlated. This suggests that heat absorp-
tion and degradation energy generated by the fabrics are
comparable under these exposures and intensities. It can
also be concluded from this study that the performance
values obtained in the 80 kW/m
2
flame exposure test
could be a representative to holistically measure, classify,
and model the protective performance of fabrics.
It has been found that thermal and evaporative
resistances are the key fabric properties to directly
affect the thermal protective performance. Also,
weight and evaporative resistance were identified as
the key fabric properties to indirectly affect the
thermo-physiological comfort performance, whereas
the water spreading speed of the fabric that is in contact
with the wearers’ skin was identified as the key fabric
property to directly affect the comfort performance.
Also, this study clearly found that thermal and evap-
orative resistances are positively related to the protect-
ive performance, while these two properties are
negatively related to the comfort performance of the
fabrics. This confirms a need for optimizing the pro-
tective and comfort performances for a particular ther-
mal exposure by controlling the fabric properties.
By implementing the key fabric properties and
performance values, the MLR and ANN models are
developed for predicting the thermal protective and
thermo-physiological comfort performances of fabrics.
As per the finding from this study, ANN models are
recommended to use for more accurately predicting the
protective and comfort performances of fabrics.
Contextually, it is notable that the ANN model
developed for predicting the thermal protective per-
formance was based on the experimental performance
values obtained for the dry fabrics only. As firefighters
sweat a lot while working, this sweat/moisture could
also affect the thermal protective performance of fab-
rics.
41,42
In addition, existing microclimate air gaps
between the clothing and firefighters’ bodies can sub-
stantially influence the protective and comfort
performances of fabrics.
3,43
In the future, it is recom-
mended to develop ANN models considering the sweat/
moisture and microclimate air gaps for more accurately
predicting the protective and comfort performances of
fabrics as well as whole clothing in consideration with
clothing features, such as fit, size, and closures.
Furthermore, a fabric with very high thermal protective
performance results in very low thermo-physiological
comfort performance. As these performances are inver-
sely related, the development of a categorization tool in
the future based on the protective and comfort per-
formances could help in finding the best balance
between these performances for the necessary protec-
tion. This type of tool could guide clothing manufac-
turers’ and/or fire stations’ clothing procurement
managers to select an appropriate fabric for the cloth-
ing based on their end-uses namely to select a fabric
having high thermal protective performance with the
maximum possible thermo-physiological comfort per-
formance for a particular thermal exposure.
Acknowledgements
The authors would like to thank DuPont, Switzerland and
Trans-Textil GmbH, Germany for supplying the fabrics for
this study. The authors appreciate the technical support from
Mr Max Aeberhard, Ms Shelley Kemp, and Mr Thomas Pitts
during the laboratory tests.
As an authors’ note, the thermal protective and thermo-
physiological comfort performances were obtained by testing
the fabrics in simulated, controlled environments to improve
reproducibility. These environments represent accepted simu-
lations of emergency conditions or scenarios to assess the
physiological impact (standardized test methods). As real
heat and fire exposure conditions are uncontrolled and
random regarding intensities, the presented results may not
be applicable to all real exposure conditions.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with
respect to the research, authorship, and/or publication of this
article.
Funding
The authors received no financial support for the research,
authorship, and/or publication of this article.
ORCID iD
Sumit Mandal http://orcid.org/0000-0001-8970-9902
References
1. Rossi R. Firefighting and its influence on the body.
Ergonomics 2003; 46: 1017–1033.
2. Kahn SA, Patel JH, Lentz CW, et al. Firefighter burn
injuries: predictable patterns influenced by turnout gear.
J Burn Care Res 2012; 33: 152–156.
Mandal et al. 11
3. Song G, Mandal S and Rossi R. Thermal protective cloth-
ing for firefighters. Kidlington, UK: Woodhead
Publishing, 2016.
4. Song G, Paskaluk S, Sati R, et al. Thermal protective
performance of protective clothing used for low radiant
heat protection. Text Res J 2011; 81: 311–323.
5. Shalev I and Barker RL. Protective fabrics: a comparison
of laboratory methods for evaluating thermal protective
performance in convective/radiant exposures. Text Res J
1984; 54: 648–654.
6. Song G and Mandal S. Testing and evaluating thermal
comfort of clothing ensembles. In: Wang L (ed.)
Performance testing of textiles: Methods, technology, and
applications. Kidlington, UK: Woodhead Publishing,
2016, pp.39–64.
7. ISO 6942:2015. Protective clothing protection against heat
and fire method of test: evaluation of materials and mater-
ial assemblies when exposed to a source of radiant heat.
8. ISO 9151:2016. Protective clothing against heat and flame
determination of heat transmission on exposure to flame.
9. ISO 12127-1:2015. Clothing for protection against heat
and flame determination of contact heat transmission
through protective clothing or constituent materials.
10. ISO 11092:2014. Textiles - physiological effects -
measurement of thermal and water-vapour resistance
under steady-state conditions (sweating guarded-hotplate
test).
11. ISO 18640-1:2018. Protective clothing for firefighters -
physiological impact - part 1: measurement of coupled
heat and moisture transfer with the sweating TORSO.
12. Zimmerli T and Weder MS. Protection and comfort - a
sweating torso for the simultaneous measurement of pro-
tective and comfort properties of PPE. In: Stull J and
Schwope A (eds) Performance of protective clothing:
STP 19909S. West Conshohocken, PA: American
Society for Testing Materials, 1997, pp.271–280.
13. Keiser C, Becker C and Rossi RM. Moisture transport
and absorption in multilayer protective clothing fabrics.
Text Res J 2008; 78: 604–613.
14. Annaheim S, Wang L, Psikuta A, et al. A new method to
assess the influence of textiles properties on human ther-
mophysiology. Part I: thermal resistance. Int J Cloth Sci
Tech 2015; 27: 272–282.
15. Benisek L and Phillips WA. Evaluation of flame retard-
ant clothing assemblies for protection against convective
heat flames. Cloth Text Res J 1979; 7: 2–20.
16. Stull JO. Comparison of conductive heat resistance and
radiant heat resistance with thermal protective perform-
ance of fire fighter protective clothing. In: Jeffrey OS and
Arthur DS (eds) Performance of protective clothing,
ASTM STP 1273. West Conshohocken, PA: American
Society for Testing Materials, 1997, pp.248–269.
17. Mandal S and Song G. An empirical analysis of thermal
protective performance of fabrics used in protective
clothing. Ann Occup Hyg 2014; 58: 1065–1077.
18. Mandal S and Song G. Modeling of thermal protective
performance of commercial woven fabric using artificial
neural network. In: 9th international meeting for manikins
and modeling, Tokyo, Japan, 21–24 August 2012.
19. Mandal S. Studies of the thermal protective performance of
textile fabrics used in firefighters’ clothing under various
thermal exposures. Edmonton: University of Alberta, 2016.
20. Song G, Barker RL, Hamouda H, et al. Modeling the
thermal protective performance of heat resistant gar-
ments in flash fire exposures. Text Res J 2014; 74:
1033–1040.
21. Song G, Chitrphiromsri P and Ding D. Numerical simu-
lations of heat and moisture transport in thermal protect-
ive clothing under flash fire conditions. Int J Occup Saf
Ergon 2008; 14: 89–106.
22. Torvi DA and Threlfall TG. Heat transfer model of flame
resistant fabrics during cooling after exposure to fire. Fire
Technol 2006; 42: 27–48.
23. Udayraj, Talukdar P, Das A, et al. Development of cor-
relations and artificial neural network models to predict
second degree burn time for thermal-protective fabrics.
J Text I. Epub ahead of print 25 March 2016. DOI:
10.1080/00405000.2016.1163917.
24. Mandal S, Camenzind M, Annaheim S, et al.
Characterization and modelling of thermal protective
performance of fabrics under radiant-heat exposure.
J Ind Text. Epub ahead of print 28 February 2018.
DOI: 10.1177/1528083718760801.
25. Mandal S, Annaheim S, Pitts T, et al. Studies of the ther-
mal protective performance of fabrics used in firefighters’
clothing under fire exposures: from small-scale to hexa-
gon tests. Text Res J. Epub ahead of print 2 August 2017.
DOI: https://doi.org/10.1177/0040517517723020.
26. Annaheim S, Pitts T, Morrissey M, et al. Prediction of
thermo-physiological impact of fire fighter protective
clothing. In: Wang F (ed.) 11th international meeting for
manikins and modeling, Suzhou, China, 12–14 October
2016, pp.1–54. Hong Kong: The Hong Kong
Polytechnic University.
27. Hui CL and Ng SF. Predicting seam performance of com-
mercial woven fabrics using multiple logarithm regression
and artificial neural networks. Text Res J 2009; 79:
1649–1657.
28. Majumdar PK and Majumdar A. Predicting the breaking
elongation of ring spun cotton yarns using mathematical,
statistical and artificial neural networks models. Text Res
J2004; 74: 652–655.
29. Murrells CM, Tao XM, Xu BG, et al. An artificial neural
network model for the prediction of spirality of fully
relaxed single jersey fabrics. Text Res J 2009; 79: 227–234.
30. Pynckels F, Kiekens P, Sette S, et al. Use of neural nets
for determining the spinnability of fibres. J Text I 1995;
86: 425–437.
31. Arupjoyti S and Iragavarapu S. New electrotopological
descriptor for prediction of boiling points of alkanes and
aliphatic alcohols through artificial neural network and
multiple linear regression analysis. Comp Chem 1998; 22:
512–522.
32. Zaefizadeh M, Khayatnezhad M and Gholamin R.
Comparison of multiple linear regression (MLR) and
artificial neural networks (ANN) in predicting the yield
using its components in hulless barley. Am Eur J Agric
Environ Sci 2011; 10: 60–64.
12 Textile Research Journal 0(00)
33. Shalev I and Barker RL. Analysis of heat transfer char-
acteristics of fabrics in an open flame exposure. Text Res
J1983; 53: 475–482.
34. Torvi DA and Dale JD. Effect of variation in thermal
properties on the performance of flame resistant fabrics
for flash fires. Text Res J 1998; 68: 787–796.
35. Torvi DA and Hadjisophocleous GV. Research in pro-
tective clothing for firefighters: state of the art and future
directions. Fire Technol 1999; 35: 111–130.
36. Booth JE. Principles of textile testing. New York, USA:
Butterworths, 1986.
37. Collier BJ and Epps HH. Textile testing and analysis.
New York, USA: Prentice Hall, 1998.
38. Mandal S, Song G, Ackerman M, et al. Characterization
of textile fabrics under various thermal exposures. Text
Res J 2013; 83: 1005–1019.
39. Schmid M, Annaheim S, Camenzind M, et al.
Determination of critical heat transfer for the prediction
of materials damages during a flame engulfment test. Fire
Mater 2016; 40: 1036–1046.
40. Mandal S, Annaheim S, Camenzind M, et al. Evaluation
of thermo-physiological comfort of clothing using mani-
kins. In: Nayak R and Padhye R (eds) Manikins for tex-
tile evaluation. Kidlington, UK: Woodhead Publishing,
2017, pp.115–140.
41. Rossi R and Zimmerli T. Influence of humidity on the
radiant, convective and contact heat transmission
through protective clothing materials. In: Johnson JS
and Mansdor SZ (eds) fifth international symposium on
performance of protective clothing: improvement through
innovation, San Francisco, USA, 25–27 January 1994,
pp.269–280. West Conshohocken, PA: American
Society for Testing and Materials.
42. Lawson LK, Crown EM, Ackerman MY, et al. Moisture
effects in heat transfer through clothing systems for wildland
firefighters. Int J of Occup Saf Ergon 2004; 10: 227–238.
43. Song G. Clothing air gap layers and thermal protective
performance in single layer garment. J Ind Text 2007; 36:
193–205.
Appendix 1: MATLAB program for the
Artificial Neural Network model to
predict thermal protective performance
% input training data i.e. Key Fabric Properties
(KFP) for thermal protective performance.
input ¼[12.9 8.9 9.7 9.4 13 12.9 79.7 80 127.7
82.7 95.4 46.6 49.3 71 61.8 65.6 83.3 64.2 60; 3
2.5 2.4 3.3 3.6 4.7 20.2 15.8 25.4 15.6 23.9 9.4 10
16.8 12.2 13 12.8 14.2 13.2];
% target training data for thermal protective
performance.
target ¼[4.3 4.5 3.6 4 5.9 5.7 17.7 14.9 24.6
16 19.7 14.5 14.1 14.6 14.2 15.2 16 13.4 12.7];
% choosing a training function
trainFcn ¼‘trainlm’; % Levenberg-Marquardt
backpropagation.
% creating a fitting network
hiddenLayerSize ¼10;
netp ¼fitnet(hiddenLayerSize,trainFcn);
% setting up the division of data for training,
validation, testing
netp.divideParam.trainRatio ¼70/100;
netp.divideParam.valRatio ¼15/100;
netp.divideParam.testRatio ¼15/100;
% training the network
[netp,tr] ¼train(netp,input,target);
% testing the network
y¼netp(input);
% assessing the performance of the trained
network. The default performance function is
mean squared error.
performance ¼perform(netp,target,y)
% saving the trained network
save netp;
% loading the trained network
load netp;
% calculating the root mean square error
rmse¼sqrt(performance);
% viewing the network
view(netp);
% using the regression analysis to judge the
network performance
[m,b,r]¼postreg(y,target);
% entering the new input
newinput ¼[64.2; 14.2];
% predicting the output corresponding to new
input
newoutput ¼netp(newinput)
Appendix 2: MATLAB program for
the Artificial Neural Network model
to predict thermo-physiological
comfort performance
% input training data i.e. Key Fabric Properties
(KFP) for thermo-physiological comfort
performance.
input¼[197 243.7 154.7 229.8 367.3 366.8-
609.9 547.5 673.8 635.2 592.1 587.8 599.8 493-
520.5 514.8 568.1 496.7 490.4; 2.8 4.3 6.7 6.8 0.5
0.8 0.8 0.5 0.3 3.3 1.9 0.6 0.5 4.3 2.6 3.4 3.2 3
2.3; 3 2.5 2.4 3.3 3.6 4.7 20.2 15.8 25.4 15.6 23.9
9.4 10 16.8 12.2 13 12.8 14.2 13.2];
% target training data for thermo-physiologi-
cal comfort performance.
target¼[124.4 123.6 128 128 119.2 121 118.5
115.5 114116.2 114.8 116.7 115.6
119.3 122.2 118.8 116.7 118 121.7];
% choosing a training function
Mandal et al. 13
trainFcn ¼‘trainlm’; % Levenberg-Marquardt
backpropagation.
% creating a fitting network
hiddenLayerSize ¼10;
netcp ¼fitnet(hiddenLayerSize,trainFcn);
% setting up the division of data for training,
validation, testing
netcp.divideParam.trainRatio ¼70/100;
netcp.divideParam.valRatio ¼15/100;
netcp.divideParam.testRatio ¼15/100;
% training the network
[netcp,tr] ¼train(netcp,input,target);
% testing the network
y¼netcp(input);
% assessing the performance of the trained
network. The default performance function is
mean squared error.
performance ¼perform(netcp,target,y)
% saving the trained network
save netcp;
% loading the trained network
load netcp;
% calculating the root mean square error
rmse¼sqrt(performance);
% viewing the network
view(netcp);
% using the regression analysis to judge the
network performance
[m,b,r]¼postreg(y,target);
% entering the new input
newinput ¼[496.7; 3;14.2];
% predicting the output corresponding to new
input
newoutput ¼netcp(newinput)
14 Textile Research Journal 0(00)
