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

net/publication/324845182

Prediction of dimensional change in finished fabric through artificial neural

networks

Article  in  Tekstil ve Konfeksiyon · April 2018

CITATIONS READS

4 653

3 authors:

Mihriban Kalkancı Mahmut Sinecen

Pamukkale University Aydın Adnan Menderes University
11 PUBLICATIONS   26 CITATIONS    36 PUBLICATIONS   82 CITATIONS   

SEE PROFILE SEE PROFILE

Gülseren Kurumer
Dokuz Eylul University
13 PUBLICATIONS   25 CITATIONS   

SEE PROFILE

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

Design of Ergonomic Office Chair Which Textile Comfort Properties Optimised View project

A RESEARCH ON DETERMINATION FOR SUITABILITY OF THE OFFICE CHAIRS TO COMFORT PROPERTIES AND ANTHROPOMETRIC MEASUREMENTS View project

All content following this page was uploaded by Mahmut Sinecen on 30 April 2018.

The user has requested enhancement of the downloaded file.

 (REFEREED RESEARCH)

PREDICTION OF DIMENSIONAL CHANGE IN FINISHED

FABRIC THROUGH ARTIFICIAL NEURAL NETWORKS
Mihriban KALKANCI1, Mahmut SİNECEN2, Gülseren KURUMER3 *

1
Pamukkale University, Buldan Vocational College, Denizli, Turkey, e-mail: mkalkanci@pau.edu.tr
2
Adnan Menderes University, Department of Computer Engineering, Aydın, Turkey.
3
Dokuz Eylül University, Department of Textile Engineering, Izmir, Turkey.

Received: 25.03.2017 Accepted: 14.03.2018

ABSTRACT
When anti-shrinkage precaution is taken for finishing processes, shrinkage could be observed with cotton and viscose fabrics by 8-
15% and 20%, respectively. Therefore, capability of estimation of shrinkage rate for fabrics at the end of finishing would be a significant
advantage. This study tried to estimate the shrinkage of single jersey and interlock fabrics at the end of relaxation processes by means of
the Artificial Neural Networks (ANN). To that end totally 72 varieties of fabric were manufactured in two groups of the elastane and the
non-elastane fabrics. Then, in each of two groups included 36 different varieties on the basis of single jersey and interlock weaving types
using six different raw materials in three different densities. The processes were applied to fabrics during finishing process are thermo-
fixing, washing, drying and sanforizing process. ANN model was used to predict dimensional change at the end of the sanforizing. For
ANN, the two-layer feed-forward perceptron, also called single hidden layer feed-forward neural network was used to estimate
dimensional change of width and length. Finally, the ANN exhibited successful performance in prediction of dimensional change in
fabrics. The prediction of the dimensional properties produced by the neural network model was proved to be highly reliable (R2> 0.98).
Keywords: Dimensional change, ANN, estimation, finished fabric, finishing

Corresponding Author: Mihriban Kalkancı, mkalkanci@pau.edu.tr.

1. Introduction measures are taken to avoid fabric tension; operating

machines causing minimum tension during finishing as
Dimensional stability is considered as one of the most
much as possible; giving rest breaks; checking tension
important properties of fabrics. Especially, dimensional
during process; using systems such as conveyor bands for
change problem encountered with jersey fabric emerges as
transfer of fabrics; applying non-tension mercerization in
a significant quality issue.
other words causticization; working with advance based on
Due to the tensions applied to the fabric during knitting, the feed-forward principle during stenter drying; applying anti-
shape of the loop changes. When the knitting process is wrinkle finishing to fabrics; and machines such as
finished and the forces are removed, the loops try to return sanforizing machines which ensure mechanical resistance
their natural shape. This change in the shape of the loop is against shrinkage must be employed. The most effective
also reflected in the knitted fabric and the shape of the fabric finishing process in gaining dimensional stability is applied
also changes. This change is called fabric relaxation. Dry, the drying operation with heating [2].
wet, wash, full and industrial relaxations are defined by
Controlling dimensional changes since raw fabric production
researchers [1].
is essential to prevent the problem of dimensional change.
In the event that no any measure is taken against shrinkage Estimation of dimensional change before manufacturing
during finishing, cotton and viscose fabrics could be seen could allow us to maintain control over dimensional change.
shrinkage up to 8-15% and 20%, respectively. Application of the Artificial Neural Networks (ANN) in
predictive studies in textile industry has increased recently.
Fabrics are required to be relieved from internal tension and ANN has been harnessed in numerous fields such as
to be processed to prevent swelling in unit section while classification of errors in textile manufacturing; prediction of
they are taken in process in finishing facilities. Following yarn quality parameters; classification of weaving and

TEKSTİL ve KONFEKSİYON 28(1), 2018 43

knitted fabrics; prediction of fabric behavior characteristics; 4. Drawing of all fabrics with the dimensional pattern
prediction of garment comfort; air permeability of features of
5. Finishing treatments
woven and knitted fabrics; prediction of drape of fabric;
prediction of concentration of dyestuff; and prediction of 6. Finished fabric measurements (Dimensional Change,
color recipe [3-16]. ANN method was also harnessed in Areal Density, Fabric Roll Width)
prediction of the dimensional characteristics of rib fabrics. In
the relevant study, it was emphasized that prediction 7. Prediction of dimensional change with ANN.
capability of the model established with the ANN method
towards the rib fabric was satisfactory [17-18].
2.2. Fabric Manufacturing and Raw Fabric
With the effect of tension and wet processes on the finishing Measurements
process, the fabric will undergo a shape change and will try
to return to its natural state at the first time it gets rid of the Within the scope of the study, 72 different fabrics were
fabric tension. This is a great inconvenience as it may result manufactured whose characteristics were showed in Table
in becoming smaller and shrinkage of the garment during 2. In order to determine the effect of addition of elastane into
use. During the production of ready-made garments, in fabric on dimensional change; fabrics were manufactured in
order to avoid problems in the fabrics coming from the two distinct groups. Whereas elastane was not included in
finishing treatments, it is necessary to perform the right the first fabric group, 5% elastane was added in the second
shrinkage process from the beginning of the fabric and to fabric group. Due to their common application in the
control all the processes from the raw fabric. All mechanical industry, cotton, cotton/viscose, viscose, cotton/polyester,
stresses and feeds should be kept under control during polyester and polyester/viscose fabrics were preferred.
finishing operations. Polyester fiber and polyester blends which exhibit
In this study, it is aimed to prediciton the dimensional dimensional stability were also included in the scope of the
changes of various knitted fabrics by ANN at the end of study. For the single jersey fabrics of Gauge is (E) 28, and
finishing operations. Firstly, ANN input parameters were machine diameter of is 32 inches. For the interlock fabrics of
determined, then single jersey and interlock fabrics were Gauge is (E) 24, and machine diameter of is 32 inches.
produced and measurements were taken on raw fabric. Yarn count of all fabrics is 20 tex.
After the raw fabric measurement procedures, finishing
All fabrics were manufactured by means of the three
treatments were applied to the fabrics. Dimensional
different pile yarn length in dense, medium and loose
changes were measured at the end of finishing treatments
patterns. The fabrics were coded in a specific order. For
and dimensional changes were predicted with ANN.
instance, the fabric with LS 5 code was no 5. single jersey
th
fabric with lycra form including 5 fiber 100% polyester.
2. Material and Method Specimens from 72 different fabrics manufactured were
2.1 Test and Measurement Plan taken from the knitting machines. After that raw fabric
measurements were taken as shown in Table 1. Finally,
Table 1 exhibits test plan, followed standards and they were applied dry relaxation process by remaining them
measurement points within the scope of the study. The on a smooth plane surface under standard atmosphere
study is planned as follows. conditions for a week.
1. Determination of ANN input parameters 2.2.1. Course Density Measurement
2. Production of raw fabrics The loop stitches, which are 1 cm long of the fabric, were
3. Raw fabric measurements (after initial and dry relaxation counted by means of a lupe. This measurement was
from the knitting machine-Course Density, Areal Density, repeated five times at different areas of the fabric for each
Fabric Roll Width) specimen; and their average value was taken into
consideration (TS EN 14971).

Table 1. Test and Measurement Plan

Activity / Tested
Measurement Point No Test Standard Measurement Point
Characteristics
Course Density Right After Process in the Machine
1 TS EN 14971
Measurement After Dry Relaxation
Wale Density Right After Process in the Machine
2 TS EN 14971
1.Raw Fabric (Knitted Measurement After Dry Relaxation
Fabric) Areal Density Right After Process in the Machine
3 TS 251
Measurement After Dry Relaxation
Right After Process in the Machine
4 Fabric Roll Width -
After Dry Relaxation
After Drying
1 Dimensional Change TS EN ISO 3759- 2009
After Sanforizing
2.Finished Fabric Areal Density After Drying
2 TS 251
(Finishing) Measurement After Sanforizing
After Drying
3 Fabric Roll Width -
After Sanforizing

44 TEKSTİL ve KONFEKSİYON 28(1), 2018

 Table 2. Fabric Characteristics
Stitch
length Stitch length
Type of Fabric Type of Fabric
Type of yarn (100 Type of yarn (100 needle
Fabric Code Fabric Code
needle /cm)
/cm)
S.1 100%Cotton 27 İNT.1 100%Cotton 32
S.2 %50 Cotton - %50 Viscose 27 İNT.2 %50 Cotton - %50 Viscose 32
S.3 100%Viscose 27 İNT.3 100%Viscose 32
S.4 %50 Cotton -%50 Polyester 27 İNT.4 %50 Cotton -%50 Polyester 32
S.5 %100 polyester 27 İNT.5 %100 polyester 32
S.6 %50 Polyester -%50 Viscose 27 İNT.6 %50 Polyester -%50 Viscose 32
S.7 100%Cotton 29 İNT.7 100%Cotton 34
S.8 %50 Cotton - %50 Viscose 29 İNT.8 %50 Cotton - %50 Viscose 34
Single S.9 100%Viscose 29 İNT.9 100%Viscose 34
Jersey S.10 %50 Cotton -%50 Polyester 29
Interlock İNT.10 %50 Cotton -%50 Polyester 34
S.11 %100 polyester 29 İNT.11 %100 polyester 34
S.12 %50 Polyester -%50 Viscose 29 İNT.12 %50 Polyester -%50 Viscose 34
S.13 100%Cotton 32 İNT.13 100%Cotton 36
S.14 %50 Cotton - %50 Viscose 32 İNT.14 %50 Cotton - %50 Viscose 36
S.15 100%Viscose 32 İNT.15 100%Viscose 36
S.16 %50 Cotton -%50 Polyester 32 İNT.16 %50 Cotton -%50 Polyester 36
S.17 %100 polyester 32 İNT.17 %100 polyester 36
S.18 %50 Polyester -%50 Viscose 32 İNT.18 %50 Polyester -%50 Viscose 36
LS.1 100%Cotton 27 LİNT.1 100%Cotton 32
LS.2 %50 Cotton - %50 Viscose 27 LİNT.2 %50 Cotton - %50 Viscose 32
LS.3 100%Viscose 27 LİNT.3 100%Viscose 32
LS.4 %50 Cotton -%50 Polyester 27 LİNT.4 %50 Cotton -%50 Polyester 32
LS.5 %100 polyester 27 LİNT.5 %100 polyester 32
LS.6 %50 Polyester -%50 Viscose 27 LİNT.6 %50 Polyester -%50 Viscose 32
Single LS.7 100%Cotton 29 LİNT.7 100%Cotton 34
Jersey LS.8 %50 Cotton - %50 Viscose 29 Interlock LİNT.8 %50 Cotton - %50 Viscose 34
LS.9 100%Viscose 29 LİNT.9 100%Viscose 34
/ LS.10 %50 Cotton -%50 Polyester 29
/ Lycra LİNT.10 %50 Cotton -%50 Polyester 34
Lycra LS.11 %100 polyester 29 (95/5) LİNT.11 %100 polyester 34
(95/5) LS.12 %50 Polyester -%50 Viscose 29 LİNT.12 %50 Polyester -%50 Viscose 34
LS.13 100%Cotton 32 LİNT.13 100%Cotton 36
LS.14 %50 Cotton - %50 Viscose 32 LİNT.14 %50 Cotton - %50 Viscose 36
LS.15 100%Viscose 32 LİNT.15 100%Viscose 36
LS.16 %50 Cotton -%50 Polyester 32 LİNT.16 %50 Cotton -%50 Polyester 36
LS.17 %100 polyester 32 LİNT.17 %100 polyester 36
LS.18 %50 Polyester -%50 Viscose 32 LİNT.18 %50 Polyester -%50 Viscose 36

2.2.2. Wale Density Measurement as to minimize variation caused by the finishing process
(Table 3). The processes were applied to fabrics during
The loop stitches, which are 1 cm wide of the fabric, were
finishing process are thermo-fixing, washing, drying and
counted by means of a lupe. This measurement was
sanforizing process.
repeated five times at different areas of the fabric for each
specimen; and their average value was taken into Washing process was carried out with 1/6 flotte ratio. The
consideration (TS EN 14971). fabrics emerged as a rope in the washing process were
squeezed by water in a tube-cutting machine and turned
2.2.3. Areal Density Measurement
into a dekatur (folded fabric) and dried in a drying machine.
2
The weight of the 100 cm fabric specimens prepared with Sanforizing process has been applied to the fabrics for
the sample cutter was weighed using a sensitive scale. This dimensional stability. At the end of the sanforizing process,
measurement was repeated for five different areas of fabrics the dimensional change and areal density values were
and their average was taken into consideration (TS 251). measured and recorded as shown in Table 3.
2.2.4. Fabric Width Measurement 2.3.1. Areal Density Measurement
2
The width of all manufactured fabric rolls were measured by The weight of the 100 cm fabric specimens prepared with
tapeline and recorded. This measurement was repeated the sample cutter was weighed using a sensitive scale. This
three times at different areas of fabrics; and their average measurement was repeated for five different areas of fabrics
was considered. and their average was taken into consideration (TS 251).
2.3. Fabric Finishing Processes 2.3.2. Fabric Width Measurement
Following measurements of the raw fabrics used in the The width of all manufactured fabric rolls were measured by
empirical study, they were incurred in finishing process to tapeline and recorded. This measurement was repeated
obtain finished fabric specimens. Knitted fabrics were three times at different areas of fabrics; and their average
processed according to their fiber types in the same way so was considered.

TEKSTİL ve KONFEKSİYON 28(1), 2018 45

 Table 3. Fabric Finishing Processes

Raw Expander Thermo fixing Wet Treatment Drying Sanforizing

Machine Fabric Expander Machines Stenter HT Drying Machine Sanfor
Machine Speed (m/min) 60 m/min 15-30 m/min 2 rev./min. 25-30 m/min 15 m/min
Manufacturing Capacity 21 000 kg/day 200kg/hr. 20 000 kg/day 12 000 kg/day 4000 kg/day
0 0 0
Process Temperature (ºC) - 185-198 C Washing80 C 120 C 120 0C
Fabric Processing Speed 60 m/min. 24 m/min. 20-25 m/min 20-25 m/min 10 m/min.

Figure 1.Shrinkage scale (for measuring dimensional change) and signing fabric.

2.3.3. Dimensional Change In the present study, MATLAB ® Neural Network Toolbox
was used for creating Multilayer Feed-Forward ANN that
In order to determine longitudinal and latitudinal shrinkage has three layers as input layer, hidden layer and output
or elongation afterwards of the finishing process, layer.
dimensional change test was applied to specimens. After
each finishing process, the dimensional change was In the ANN model, initial weight and bias values were
measured (TS EN ISO 3759- 2009). A special pattern was assigned randomly; and task of changing weights was
used to determine longitudinal and latitudinal shrinkage or performed online basis. In other words, the established ANN
elongation. In the beginning of finishing treatments, all alters weights one unit at each iteration online basis.
fabrics were drawn on the dimensional pattern (50cm x Whereas sigmoid activation function was used as the
50cm) using a permanent marker. After each finishing activation function, training of the ANN model, commenced
process, dimensional change was measured by means of a with determined network parameter, is continued until
shrinkage scale that is shown in Figure1. Measurements certain number of iteration (epoch) (It was set as 1.000 for
were taken inner borders of the first drawn line on the ruler this study). The activation function employed in the study is
(35x35 cm). a sigmoid function that yields continuous responds in certain
pattern with respect to the data entered into the function.
3. Artificial neural networks and structure These responds are certainly not discrete. Thus, sigmoid
ANN, which is the computer system, is developed by function is commonly applied because it refers the most
modelling of human brain. Main properties of ANN are appropriate function for the problems requiring sensitive
automatically creating, deriving and exploring new evaluations. The important point is that function needs to
information using learning like human brain. Technically, the allow derivation operation.
major task of artificial neural networks is to predict an output 3.1. Input Parameters
data set with respect to the entry data set given to them. In
accomplishing this task, the network is required to be Dimensional change in finished fabrics prepared for garment
trained with examples of the relevant incident so that it could manufacturing is determined by numbers of factors such as
make a generalization which enables the system to predict raw material, knitting conditions and finishing processes.
output set corresponds to the similar events [19-20]. Before delivered to garment manufacturing facility, fabrics
incur sanforizing process at the finishing stage as the last

46 TEKSTİL ve KONFEKSİYON 28(1), 2018

process. To the end of predicting dimensional change at the b. Dimensional change (cm –width)
end of sanforizing process, input data were selected c. Dimensional change (cm –length)
respectively. Raw fabric parameters (at the moment
3.2 Output Parameters
operation on fabrics is completed in the machine in a way
that they were processed in the dry relaxation) that could be According to the established Network, parameters of
effective on the ultimate properties of the fabric were finished fabric parameters at the end of the sanforizing
selected as mean areal density, fabric width, course and process were given below:
wale density. Additionally, at the finishing process, areal
1. Dimensional change (cm –width)
density, longitudinal and latitudinal dimensional change
2. Dimensional change (cm –length)
values in the main drying process, the most effective
process on the dimensional change, were selected as input In the training period of the Artificial Neural Networks model,
parameters. Selected finished fabric input parameters were the assessment was conducted based on 72 different
coded as below for the purpose of digitalization of entry values from 14 independent input variables and 2 output
data. variables (Figure 1). Whereas 70% of these values were
used for training, 15% were for confirmation and 15% were
1. Fabric Code: 4 types of fabric in S, LS, INT and LINT
for testing purposes. In other words, randomly selected 50
were coded as fabric 1, 2, 3 and 4, respectively.
fabric rolls were used for training, 11 fabric rolls were for
2. Yarn Type: 100% cotton, 50% cotton – 50% viscose, confirmation and 11 fabric rolls were for testing.
100% viscose, 50% cotton – 50% polyester, 100%
polyester, and 50% polyester – 50% viscose. Since it In the light of information given above, this study examined
these verbal values could not be employed in an network performance with respect to different learning rule,
artificial neural networks model, they were transformed training function, momentum coefficient, number of hidden
into digital codes a 1, 2, 3, 4, 5 and 6, respectively. layer and number of nodes in hidden layer in order to
determine the optimal ANN performance value. For training
3. Stitch length (100 needle/cm) desired: values purpose of the ANN, Levenberg-Marquardt model was
determined as 27, 29, 32, 34 and 36 were coded for the utilized; error was determined by means of the Mean
ANN model as 1, 2, 3, 4 and 5, respectively. Square Error method. Accordingly, number of hidden layer
4. At the end of the machine operation (momentarily) and cell numbers in hidden layer in the study were
2
a. Areal Density (gr/m ) determined based on the learning rate and momentum
b. Fabric Roll Width (cm.) coefficient found frequently in the literature. When learning
c. Course Density (course per cm.) rate was 0.001; momentum coefficient was 0.8; then, hidden
d. Wale Density (wale per cm.) layer was selected 1. In the proposed model, number of
5. After dry relaxation neurons of the hidden layer was tried starting from 3 up to
2
a. Areal Density (gr/m ) 50. Table 6 exhibits structure parameters of the established
b. Fabric Roll Width (cm) ANN model.
c. Course Density (course per cm.) Given input parameters, dimensional and width change that
d. Wale Density (wale per cm.) could arise after sanforizing process were predicted
6. After drying according to the yarn types in “cm”. Table 7 exhibits input
2
a. Areal Density (gr/m ) values of specimen taken randomly.

Figure 1. Finished fabric network diagram

TEKSTİL ve KONFEKSİYON 28(1), 2018 47

 Table 6. Artificial neural network parameters
Network Parameters
Number of İnput Parameters 14
Number of Hidden Layer 1
Number of Neuron in the Hidden Layer Between 3 and 50
Learning Rate 0.001
Momentum Coefficient 0.8
Number of Iteration 1,000
Number of Output Parameters 2
The Most Successful Network Number of Hidden Layer Neuron 14
The Most Successful Network Number of Iteration 11
2
The Most Successful Network R Value 0.96132

Table 7. Specimen finished fabric input data

At the end of machine operation At the end of dry relaxation At the end of drying
process

Course Density (course per cm.)

Course Density (course per cm.

Dimensional change cm -length

Dimensional change cm -width
Stitch length (100 needle /cm)

Wale Density (wale per cm.)

Wale Density (wale per cm.)

Areal density(g/m2)

Areal density(g/m2)

Areal density(g/m2)
Fabric Roll Width

Fabric Roll Width

Fabric Code

Yarn Type

desired

1 1 1 128 113 22 12 135 112 23 12 140 39,95 52,75

1 2 1 133 114 21 12 143 112 22 12 162 38,85 47,6
1 3 1 135 116 20 12 142 114 21 12 179 37 49,95
1 4 1 130 115 21 12 136 113 22 12 172 41,5 42,5
1 5 1 137 114 20 12 150 112 21 12 199 44,75 40

Given input and output parameters, longitudinal and hidden layer of the ANN model was increased from 3 to 50
latitudinal dimensional changes that could arise as a result so that the weights of the model yielding the most accurate
of sanforizing process were predicted in centimeter with result could be recorded.
respect to yarn types.
In this article, it was determined that the ANN model gave
the most accurate result with hidden layer having 14 nodes
4. Results and Discussion
(Figure 2). Mean Absolute Error (MAE) value of this ANN
In order to test consistency of the obtained results, model is dimensional width 0.7623 and dimensional change
performance value (accomplishment or accurate estimation height 1.1619. MAE values of the first 10 sample is shown in
value) was measured. Thus, the number of node(s) in the Table 8, and all values is plotted in Figure 3.

Figure 2. Artificial neural network diagram illustrated for dimensional change in finished fabric

48 TEKSTİL ve KONFEKSİYON 28(1), 2018

 Table 8. Mean Absolute Error Values of First 10 Sample

Sample No 1 2 3 4 5 6 7 8 9 10
Dimensional Change
0.1760 1.9848 1.0700 2.4036 0.2165 5.7511 1.6119 1.4067 0.4178 3.6911
(cm - Width)
Dimensional Change
2.4464 6.7724 1.6965 7.5441 1.0408 3.4202 0.7061 3.7837 0.8337 7.7827
(cm - Length)

Figure 3. Mean Absolute Error Value of Output

For evaluation of the dimensional change in finished fabrics, process displayed difference with respect to the one
output values were digitalized according to measurement predicted by the ANN.
results. For instance at the end of the drying process, if
Given the obtained results through the ANN model, the
shrinkage was measured as -13.2% cm with the 50x50 cm
performance plotting in Figure 4 ceased the iteration after
pattern with respect to the fabric input value, the real value
11 steps according to the MSE (Mean Square Error) method
was digitalized as 43.50 cm similar to the value at the first
after the best error value (where minimum error occurs) was
raw in Table 8. In other words, width of a fabric with 50 cm
obtained regarding the Training, Validation and Test values.
original dimension was measured as 43.40 cm at the end of
In Figure 4, after the data was determined for training of the
the sanforizing process. On the other hand, the ANN model
ANN model, test and validation data were run in the ANN
predicted the output dimension as 43.24 cm. In terms of the
model following the training, the smallest of these three
fabric width, the difference between the measured and
values was displayed on the MSE value plotting. In other
predicted values was 0.16 cm that was considered
words, whereas the data isolated for the training purpose
prediction error of the ANN model. After the raw fabric was
were used during training of the ANN model, the most
incurred all operations following its entry, width of the raw
appropriate MSE value was given as validation plotting for
fabric in 50 cm dimension was measured at the end of the
the values predicted based on the training, validation and
sanforizing process as 43.40 cm. On the other hand, the
test data during performance testing of the model.
proposed ANN model predicted this dimension as 43.24 cm.
Accordingly, it was observed that the model exhibited the
Similarly, whereas the length of the fabric in 50 cm original th
best performance at the 11 iteration. This iteration was
size was measured as 46.70 cm at the end of the
determined as the most successful prediction model since
sanforizing process, it was predicted by the ANN model as 2
the model displayed correlation coefficient (R ) was 0.96132
46.297 cm.
(Figure 5). It could be observed that the error displayed by
Table 9 exhibits data regarding randomly selected specimen the ANN model reduced as number of iteration increased.
taken at end of processes. According to the table, fabric After the training of the model, correlation coefficient was
width and lengths measured at the end of the sanforizing determined as 0.96132 that suggests strong correlation
between the real measurements and predicted values.

Table 9. Data of samples of finished fabrics at the end of processes (in cm)

Mean Absolute Percentage

Measured Value Predicted Value by ANN Measured – Predicted (Error)
Error (%)

Dimensional Dimensional Dimensional Dimensional Dimensional Dimensional

Dimensional Dimensional Change Change Change Change Change Change
Change Width Change Length Width Length Width Length Width Length
43.4 50.0 43.240 51.582 0.160 -1.582 0.37 3.16
42.5 46.7 41.990 46.297 0.510 0.403 1.20 0.86
38.5 54.5 37.585 56.433 0.915 -1.933 2.38 3.55
44.1 39.35 43.318 38.316 0.782 1.034 1.77 2.63
45.0 39.0 44.661 37.92 0.339 1.080 0.75 2.77

TEKSTİL ve KONFEKSİYON 28(1), 2018 49

 learning rule and sigmoid transfer function were
implemented. As a result of the study, it was observed that
the ANN model displayed successful predictions regarding
dimensional change in finished fabric.
Owing to the tensions subject to finishing process and effect
of the wet treatments, dimensional changes would occur in
fabrics and they would try to return their very original form
as soon they are relieved from these tensions. This situation
is considered as a significant nuisance since it would lead
shrinkage of clothing after washed. Afterwards of the
finishing process, it is required to deliver fabrics with desired
shrinkage values to the garment manufacturing stage. In
order to avoid garment manufacturers to experience
problems with fabrics sent from the finishing facilities, it is
required to implement non-shrinkage process on fabrics
Figure 4. Training of the network yielding the best result for the from the very beginning accurately and to keep all
finished fabric dimensional change. processes under control since the raw fabric stage. It is
considered that in case shrinkage prediction methods such
as artificial neural networks are utilized for the fabrics with
5. Conclusions
high dimensional change potential, it would be possible to
In the ANN model structured for prediction of dimensional reduce additional precautions necessary to eliminate
change, feed-forward, back propagation, momentum shrinkage and to increase productivity.

Figure 5. Correlation coefficient of the network

REFERENCES
1. Marmaralı A. Atkı Örmeciliğine Giriş, İzmir, Ege Üniversitesi Yayınları, 2004.
2. Çoban, S., (1999), “Genel Tekstil Terbiyesi ve Bitim İşlemleri, İzmir”, Ege Üniversitesi Yayınları, pp:248-265
3. Matusiak M. (2015), “Application of Artificial Neural Networks to Predict the Air Permeability of Woven Fabrics”, FIBRES & TEXTILES in Eastern Europe,
Vol: 23, 1(109), pp: 41-48.
4. Bhattacharjee D, Kothari VK. (2007), “A Neural Network System for Prediction of Thermal Resistance of Textile Fabrics”, Textile Research Journal, Vol: 77,
pp: 4-12,

50 TEKSTİL ve KONFEKSİYON 28(1), 2018

5. Hui CL, Lau TW, Ng SF, Chan KCC. (2004), “Neural Network Prediction of Human Psychological Perceptions of Fabric Hand”, Textile Research Journal,
Vol: 74 pp:375-383.
6. Park SW, Hwang YG, Kang BC, Yeo SW. (2000), “Applying Fuzzy Logic and Neural Networks to Total Hand Evaluation of Knitted Fabrics”, Textile
Research Journal, Vol:70, pp: 675-681,
7. Majumdar A., (2011), “Modeling of Thermal Conductivity of Knitted Fabrics Made of Cotton-Bamboo Yarns Using Artificial Neural Network”. The Journal of
The Textile Institute, Vol: 102(9), pp: 752-762.
8. Wong ASW, Li Y, Yeung PKW, Lee PWH., (2003),“Neural Network Predictions of Human Psychological Perceptions of Clothing Sensory Comfort”, Textile
Research Journal, Vol:73, pp:31-37.
9. Kumar V, Sampath VR., (2013), “Investigation on the Physical and Dimensional Properties of Single Jersey Fabrics made from Cotton Sheath–Elastomeric
Core Spun”, FIBRES & TEXTILES in Eastern Europe; Vol: 21, 3(99) pp: 73-75.
10. Farooq A., (2014), “Predicting the Dynamic Cohesion in Drafted Slivers at Draw Frame Using Artificial Neural Networks”, Textile and Apparel, Vol: 24(3).
11. Murrels, C.M., Tao, X.M., Xu, B.G., and Cheng, K.P.S., (2009), “An Artificial Neural Network Model for the Prediction of Spirality Fully Relaxed Single Jersey
Fabrics”, Textile Research Journal, Vol:79(3), pp: 227-234.
12. Jıanda, C., Xıaojun, G., Lıanfu, Y., (2004), “Research on BP Neural Network Applied to Predict Cotton Fabric Handle”, Proceedings of The Textile Institute
83rd World Conference, pp:1265-1268.
13. Huı, C-L, NG, S-F, (2005), “A New Approach for Prediction of Sewing Performance of Fabrics in Apparel Manufacturing using Artificial Neural Networks”,
The Journal of Textile Institute, Vol: 96,.6, pp: 401-405.
14. Huı, C-L, NG, S-F,( 2005), “A new Approach for Prediction of Sewing Performance of Fabrics in Apparel Manufacturing Using Artificial Neural Networks”,
The Journal of Textile Institute, Vol: 96,6, pp: 401-405.
15. Warren, J. Jasper, Kovacs, E. and Berkstresser, G. A., ( 1993 ). “Using Neural Networks to Predict Dye Concentrations in Multiple-Dye Mixtures”, Textile
Research Journal, Vol. 63, pp:545 - 551.
16. Arıkan Kargı, V. Sinem, (2014)., “A Comparison of Artifıcial Neural Networks and Multiple Linear Regression Models as in Predictors of Fabric Weft
Defects”, Textile and Apparel , Vol: 24(3), pp: 309-316.
17. Saravana, K.T., Sampath, V., (2011 ), “An Artificial Neural Network System for Prediction of Dimensional Properties of Weft Knitted Rib Fabric”, Journal of
the Textile Association, Vol: 71(5) pp: 247–250.
18. Saravana K.T, Sampath, VR.,(2012), “Prediction of Dimensional Properties of Weft Knitted Cardigan Fabric by Artificial Neural Network System”, Journal of
Industrial Textiles, Vol: 42(4), pp: 446–458.
19. Öztemel E., (2006), Yapay Sinir Ağları. Papatya Yayıncılık Eğitim, İstanbul, Türkiye.
20. Özdemir H., (2013), “Yapay Sinir Ağları ve Dokuma Teknolojisinde Kullanımı”, Tekstil Teknolojileri Elektronik Dergisi, Vol:7(1), pp: 51-68.

TEKSTİL ve KONFEKSİYON 28(1), 2018 51

View publication stats