Академический Документы
Профессиональный Документы
Культура Документы
F. Khedher1
Evaluation of Sewed Thread Consumption
of Jean Trousers Using Neural Network
and Regression Methods
1Textile Engineering Laboratory, Abstract
2National Engineering School of Monastir,
This paper deals with the prediction of the sewing thread consumption of jean trousers us-
ing the neural network technique. The neural network results and analysis are discussed
University of Monastir, and investigated. Indeed the findings show that neural network consumption values give
Ibn ElJazzar Street, Monastir 5019- Tunisia better fitting of experimental results than the ones obtained using regression technique.
E-mail: boubaker.jaouachi@enim.rnu.tn/
boubaker.jaouachi@gmail.com However, compared to the experimental consumption results, theoretical ones of the sewn
jean pants seem widely predictable in the desired field of interest. Among the all parameters
studied, statistical analysis results also indicate that five inputs can be considered as influ-
ential ones. When classifying these five influential inputs, only three parameters are con-
sidered most significant. In fact the thread consumed to sew jean trouser samples remains
influenced especially by the thread properties and needle fineness as well. Compared with
the regression model, the neural network model gives a more accurate prediction and to a
great extent provides the amount of sewing thread.
mathematical methods (fuzzy, math- regulation points were varied: the thread
n Introduction ematical, geometrical, statistical, subjec- components expressed by the number of
tive, etc.) in defining, prediction and in twisted thread ends (Tye), the needle size
With actual increases in sewing speeds
modelling both complex and non-linear (Ns), fabric thickness (Fth), the mass or
and with the advent of both synthetic fab-
problems, the neural network method of- weight of fabric (Wf), the stitch length
rics and threads, the problem of exces-
fers large levels of flexibility and remains (SL), and sewing machine type (Smt).
sive thread consumption in sewing op-
an excellent method for predictors [2, Each level of the input parameters (1 and
erations has become more critical and of
3, 13 - 15, 26 - 28, 32]. The purpose of 2) for adjustment represents a regulation
greater interest. A reasonable estimation
this study is to determine accurately the point, with 1 corresponding to the mini-
of potential thread sewn for the garment
amount of sewing thread as a function mum, and 2 referring to the maximum, as
industry should be determined objective-
of some influential inputs. Moreover in shown in Table 1.
ly not only to predict suitable thread con-
this work, the amount needed by the type
sumed and to reduce the stocks unused,
of garment was deduced by selecting the These input parameters are considered
but it also allows us to use a better-quali-
ty of sewing thread for the same cost and most significant parameters using statisti- because of their probable contribution to
to avoid stock rupture. However, in the cal analysis, therefore it is focused on the the thread consumption value of the jean
general literature survey, the thread con- evaluation and prediction of the thread trousers. They are investigated objective-
sumption problem has not been studied consumption of jean trousers using neu- ly according to the factorial experimental
sufficiently for two important reasons: ral network modelling. The efficiency of design. Factorial designs allow for simul-
first because of the complexity of sewn our model was then analysed and inves- taneous study of the effects that several
thread evaluation, and second the high tigated. factors may have on a process. When
input parameters. Some works relating to performing an experiment, varying the
thread consumption evaluation are tack- levels of the factors simultaneously rath-
n Materials and methods
led and conducted in the literature survey. er than one at a time is efficient in terms
Thread consumption evaluation using Data collection and analysis of time and cost, and also allows for the
such techniques was measured as a func- Six different input parameters were cho- study of interactions between the factors.
tion of some input parameters such as the sen and used in the experiment to sew Interactions are the driving force in many
stitch length, thread tension and its com- jean trouser specimens in order to evalu- processes. Without the use of factorial
pressive modulus [1, 4 - 12, 15, 17 - 22, ate their experimental thread consump- experiments, important interactions may
25, 34 and 35]. However, the sewing tion values. Table 1 shows the input pa- remain undetected. Hence, in a full facto-
thread consumption of a garment needs rameters and their corresponding levels. rial experiment, responses are measured
to be provided and predicted accurately. To regulate and adjust the sewing condi- at all combinations of the experimental
Several factors determine the extent of tions, overall input parameters with two factor levels. The combinations of the
thread consumption in any sewn gar-
ment, such as the stitch length, stitch Table 1. Input parameters and their corresponding levels.
density, seam type and material thickness
[16, 33]. Until now, there has been no an- Input Twisted thread Needle size Sewing machine
Fabric Stitch Weight of
thickness length fabric
alytical model using the neural network level ends, Tye Ns, Nm type, Smt
Fth, mm SL, mm Wf, g/m2
technique to evaluate the amount of sewn Without automatic
1 2 90 0.60 2.25 268
thread required to make up jean trouser thread trimmer
garments and to predict accurately the 2 3 120
Within automatic
0.92 2.85 417
thread trimmer
consumed thread value. Compared with
Jaouachi B, Khedher F. Evaluation of Sewed Thread Consumption of Jean Trousers Using Neural Network and Regression Methods. 91
FIBRES & TEXTILES in Eastern Europe 2015; 23, 3(111): 91-96.
which are chosen to analyse the overall
specimens. The output predicted using
the neural network (STCNN) and regres-
sion models (STCact) is the thread length
sewn, which represents the thread con-
sumed to sew classic Jean trousers with
5 pockets. These values refer to the sewn
thread lengths predicted and the actual
Lockstitch
measures of unstitch thread length val-
ues founded experimentally. As shown in
Table 1, two types of jean fabrics (heavy
and light) were used in the study within
two different compositions: 100% Cot-
ton and 98% Cotton + 2% Elastane. The
Safety-stitch majority of assemblies made on the jean
pants were sewn with different types of
stitches, for example: flat felled seam
(2 × 401), safety stitch machine (ISO-
516), overlockstitch machine (401)
and lockstitch machine (301). In spite
of its cross section, which is different,
the appearance of this stitch, 401-401,
as shown in Figure 1, can, however,
be made using a JUKI DDL-8700 ma-
chine, to obtain a similar stitch. That is
why we used this kind of sewing ma-
Lockstitch Felled seam
chine in our case. We will study their
effect on thread consumption. All adjust-
ment conditions are regulated to obtain
Figure 1. Different stitch types used to sew the jean trousers. good quality of assembly. Jean fabrics
were seamed on JUKI DDL-8700 and
Table 2. Sewing thread consumption of trousers predicted and evaluated by the ANN model MO-3316 sewing machines with two
developed. ♦STCact - Represents the actual length of thread consumed during sewing of jean different sewing needle sizes: 111 and
trouser samples, measured experimentally by unstitching. ♣ STCNN - Represents the theo- 83 dtex. All machine settings were the
retical consumption of sewing thread predicted using the ANN model developed.
same for all trouser samples. Denim fab-
Inputs tested within their levels Sewing thread consumption, m ric layers were seamed with stitch den-
Test Experimental ♦ Theoretical ♣ sities of 4.5 and 3.5 stitches/cm, which
Smt SL Tye Ns Fth Wf
STCact values STCNN values were not identical to those selected by
1
1
2.25 120 0.60 268 300 290.0 Webster et al. [34, 35].
2 0.92 417 420 416.3
2.85 3
3 0.60 268 420 413.1 Therefore this study is essentially carried
2 90
4 2.25 0.92 417 375 373.6
out according to these stitches. Overall
5 1 2.85 268 360 372.1
0.60 seam operations are realised within the
6 2 120 417 190 165.8
2.25 2 same type of mean sewing thread count,
7 1 345 364.2
0.92 268 16.6 tex, to obtain jean trousers with
8 2 2.85 90 348 355.1
5 pockets, as shown in Figure 1. In the
9 1 2.25 0.60 380 390.6
3 same Figure, all stitch kinds in some spe-
10 0.92 417 250 258.3
11
120
200 182.6
cific and basic zones are shown. In the
12 2
2.85
90
0.60
268 350 346.3
area of the belt, belt loops and pockets,
13 2 120 417 210 191.4
the stitches used to sew them are men-
14 2.25
0.92
268 330 338.3 tioned by lockstitch type 301 and over-
15 1 2.85 0.60 417 345 340.6 lockstitch type 401. Indeed this use of
16 3 90 0.92 405 405.2 many more different types of stitch in
268 Figure 1 than was explained previously
17 2.25 310 329.5
2 2 0.60
18 417 300 297.9 is required to prevent redundant informa-
19 2.85 3 320 289.8 tion, which makes Figure 1 and the ex-
120 0.92 268
20 1 2.25 2 230 232.0 planation in the text complementary and
not doubtful.
factor levels represent the conditions at vation. The entire set of runs is the “de- Sewing thread consumption values
which responses will be measured. Each sign.” The number of tested samples giv- (STC) are determined by unstitching the
experimental condition is a called a “run” en by the full experimental design, using seam and measured according to French
and the response measurement an obser- Minitab 14 software, is 64 combinations Standard NF G07 101. Thus after every
operation STCact. Besides this, to com- and artificial neural-network models for
pare theoretical and experimental results, predicting thread consumption, we in- c
twenty jean trouser specimens were sewn vestigated the experimental consumption
and unstitched in order to measure thread values under applied factorial design. In Figure 2. Example of Neural Network
consumption at the levels prescribed by fact, by classifying the contribution of structure.
the factorial design. each input parameter to the mean thread
consumption, we analysed the factorial [15], for the highest needle size the sew-
Artificial Neural Network parameters design using Minitab software analysis. ing thread can cause a surplus of length-
Figure 2 shows an example of a neural Figure 3 shows the overall means at each ening and a large loop of thread, which
network structure which is composed of level of input, represented using main ef- we can avoid. Thus, according to Seyam
three different inputs (a, b, and c) and fect plots. The main effect plots can be and El Sheikh’s study [31], an appro-
one predictable output. To build a neural drawn for either the raw response data, priate thread for the correct needle size
network model, some parameters should the means of the output parameter for should be selected carefully. However,
be considered. Hence the size of the hid- each input level, fitted values after ana- to reduce the mean value of the thread
den layers, the number of neurons in each lysing the design and predicted values for consumed, it is suitable to use sewing
hidden layer and the neurons in the out- each input level. These plots are used to thread composed of two twisted threads.
put layer play a vital role during model- compare the magnitudes of all different Furthermore increasing the number of
main effects studied. A main effect oc- twisted threads to sew jean trousers en-
ling steps.
curs when the mean response changes courages a high consumption value. An
For example, the model shown in Fig- across the input levels. average increase value of 21.82% was
ure 2 is built using one hidden layer of saved when the thread was composed
Moreover, Figure 3 also shows the vari- of three twisted threads instead of two
three neurons and to test the training
ation of main input parameter effects on only. Thus in order to minimise sewing
of network inputs. Moreover a suitable
the mean sewing thread consumption thread consumption, a low twisted thread
transfer function, adjusted weights and
values. According to the same figure, it number seems more recommended. By
thresholds should be selected to start
is clearly noted that the needle size and classification of the contributions of in-
prediction. In fact, these neural network
type of thread used to sew jean trousers puts, we can also remark that the weight
parameters are automatically modified
are the most influential parameters. Due of fabric is significant. Indeed the weight
during training as a function of the best
to the high variation presented and saved fabric parameter participates in the vari-
accuracy of the model. The performance when the input value changes from one ation in consumption because it presents
results of the network express the accu- level to another, the mean thread con- a non negligible effect when the level is
racy of the model, measured after train- sumed changes enormously too. Hence modified from the lowest to the highest.
ing and evaluated by the error values we can remark also that the decrease in Regarding the increase in the weight of
between the actual and predicted output. the size of needle (83 dtex instead of fabric, the consumption value decreases
Some works were conducted using the 111 dtex) caused an average decrease by an average of 10.18%. Compared to
neural networks method [3, 13, 16]. To equal to 33.64% of the mean consump- Ns, Tye and Wf, the other input parameters
test our network model, we divided each tion value. According to earlier work can be considered as less significant due
under-design of experiments formed
of 64 tests into two samples: A training
sample that contains 44 random tests Smt SL Tve
and a validation sample with the rest as 360
remaining tests. In fact, our neural net-
330
work is characterised by 300 iterations,
for 6 inputs and one output. To build our 300
Mean STC actual value, m
the stitch length, (SL), and the sewing 100 Linear regression
Since 2004 Laboratory has had the accreditation of the Polish Centre for Accreditation No. AB 551, confirming that
the Laboratory meets the requirements of Standard PN-EN ISO/IEC 17025:2005.
Contact: