Monitoring and Control of Industrial Sewing Machines

Posted on : by : Writer

Monitoring and Control of Industrial Sewing

Processing textile materials is generally very
difficult due to the flexible nature of the material. In industries
using sewing as assembly process, most processes rely on human
labor, being difficult or even impossible to automate. The relations
between machine configuration and adjustment, material
properties, and the resulting product quality are also complex.
This paper describes current work using an instrumented
lockstitch sewing machine to study the dynamics and variations of
one of the important process parameters during high-speed sewing
of shirts: thread tensions. The objective is study the principles that
may allow for an automatic setting of the machines, quality control
and for real-time process control. It has been found that
differences in material properties result in measurable features of
the thread tension signals acquired VS enterprises

A software application has been developed in Labview
allowing the acquisition and processing of the resulting signals.
The signal processing functions of this software have been
reported elsewhere [3]. The most important one is splitting the
thread tension signals into stitch cycles (each cycle
corresponding to one rotation of the machine’s main shaft) and
in turn dividing each stitch cycle into phases, which are
associated to specific events of stitch formation. For each one of
these phases, that will be described later, features such as peak
values, power, energy or average of the signal is computed.
In the current experimental work, thread force waveforms
throughout the stitch cycle are being analysed when varying
parameters such as static thread tension adjustment, number of
fabric layers, mass per unit area and thickness of fabric, needle
size and sewing speed. Both the effect of the machine settings
and process variables on the thread tensions, as well as the effect
of the material properties are investigated. In this paper, the
effect of static thread tension and the influence of the fabric on
the dynamic tension signals are analysed.
The first step was to observe the resulting thread tension
signals and interpret their relation to the stitch formation
process. Some trials with the adjustment of the needle thread
pre-tensions were made.
Afterwards, a more comprehensive experiment was set up to
investigate on the influence of the material being sewn.
Three similar shirt fabrics with different mass per unit area were
used, namely
• Fabric 1 : 1×1 plain weave ; 100% cotton; 102 g/m2;
thickness 0,22 mm
• Fabric 2 : 2×1 twill fabric; 100% cotton; 127 g/m2;
thickness 0,23 mm
• Fabric 3 : Mixed structure; 100% cotton; 118 g/m2;
thickness 0,23 mm vs sewing machine

Monitoring and Control of Industrial Sewing

For each fabric, strips of fabric of 10 cm width and 30 cm
length were cut. Specimens with two and four layers of these
strips were prepared. On each one of them, 10 seams with 20
stitches each were performed.
Peak values for each of the three defined stitch cycle phases
(see next section) were extracted by the developed software.
Results were compared between specimens of two and four
layers. For this purpose, the Statistical Package for the Social
Sciences (SPSS 20.0) was used.
For each experiment a MANOVA (one-way between-groups
multivariate analysis of variance) was performed. Three
dependent values were used: Peak values of thread force in
phase 1, 2 and 3. The independent variable was the number of
layers: 2 or 4. The analysis was carried out following the
recommendations of Pallant [16].
Preliminary assumption testing was conducted to check for
normality, linearity, univariate and multivariate outliers,
homogeneity of variance-covariance matrices, and
multicollinearity, with no serious violations note

The results of the MANOVA analysis include the F-statistic
value, average (M), standard deviation (SD), Wilks’ Lambda,
significance level p and partial eta squared. Wilks’ Lambda is
one of the most reported statistics. If the associated significance
level p is less than 0.05, then it can be concluded that there is a
significant difference between groups. Partial eta squared, also
known as effect size, shows the proportion of the variance in the
dependent variable than can be explained by the independent
variable. The guidelines proposed by Cohen [17] have been
used in this work: 0.01=small effect, 0.06=moderate effect,
0.14=large effect.
When the results for the dependent variables were
considered separately, a Bonferroni adjusted alpha level of 0.017
was used. In this case, a significance level p smaller than 0.017
represents a significant difference.