Sunteți pe pagina 1din 4

The Development of Cotton-yarn-quality Predicting System

Ying Xiao1 Shulin Zhao2


Tianjin Polytechnic University, School of Textiles Tianjin Polytechnic University, School of Textiles
Tianjin, China Tianjin, China
yingxiao_sun@tom.com

Abstract: A new cotton-yarn-quality predicting system was


developed by using the merging programming technique of
VB and Matlab in the paper. Using this system, the tenacity
and evenness CV% of the cotton-yarn processing by
conventional spinning at standard temperature and
humidity can be predicted through inputting some fiber
properties into the system, such as the percentage of
impurities, the principal length, the percentage of short
fiber, the degree of maturity, fiber strength and the value of
Micronaire. And also it was verified that the system did the
good job for predicting yarn tenacity and evenness CV%
exactly with the relative error of less than 4% after the
model being trained. The accuracy can meet the demand of
spinning factories and so the predicting results would be
useful for guiding the spinning practice.
Figure1.The Overall System structure
Key words: BP network; network model; yarn quality
prediction; merging programming of VB and Matlab
The system mainly consisted of the neural network
I. INTRODUCTION building module, network training module, yarn-quality
predicating module, the model database and other
The quality of yarns is mainly determined by fiber
modules. Each module could operate coordinately under
properties, spinning processing, spinning equipments and
the domination and regulation of main control program of
the temperature and humidity of workshops, and so on. In
VB and realize many functions, such as data document
these factors, fiber properties are the most important one
read and writing, network modeling, network model
for the yarn quality. In order to predict the yarn quality
training, retention, call, and yarn quality prediction.
through digital simulation, it is essential to analyze the
relationship between fiber properties and yarn quality III. BP NEURAL NETWORK MODEL DEVELOPMENT
under any other conditions unchanged, such as processing
temperature and humidity. In this paper, a BP neural 1. The Introduction of BP Neural Network
network model was set up and based on the model, a new The BP Neural Network ( short for BP network), also
yarn-quality predicting system was developed by which known as back propagation network, is a multiple layer
the tenacity and evenness CV %of cotton yarns by ring feed-forward network, which generally includes an input
spinning process at the standard temperature and humidity layer, an output layer and one or several hidden layers,
could be predicted through inputting six fiber properties with its operation mainly consisting of two processes: the
into the system. Furthermore, it was validated that the feed-forward transmission of the information and the
system model did a good job for predicting both the yarn reverse transmission of the error[1]. In the former process,
tenacity and the evenness CV% exactly with the relative the information was calculated from the input layer to
error of less than 4%. output layer, layer by layer via hidden layer ,with the
neural unit state of each layer only affecting the next
II. THE DESIGN OF THE OVERALL STRUCTURE one's ;and in output layer, the system would turn to back
A system was developed by using the merging propagation if the desired results were not achieved when
programming technology of VB and Matlab and the compared with the target value. Then the error was
system integrated the user-friendly interface of VB and transferred from the output layer to input layer with the
the powerful function for data processing of Matlab. The network algorithmically adjudging its own weights and
overall system structure was shown in Figure 1. thresholds and the adjusting process was repeated till it
reached the desired results.
2. Neural Network Model Development
Aiming at the above issue, the strength and evenness
CV % of the yarns was predicted when the 3 layers and 4
layers BP networks were developed with neural network

978-1-4577-0860-2/11/$26.00 ©2011 IEEE


toolbox in Matlab environment and trained [2-3]; viewed reduce the time for network training. Meanwhile, the
from the training effect, not only the training time of network model structure was not too complicated and the
3-layer-network was shorter than the 4-layer-network, the problems were relatively simple in the quality predicting
prediction precision was also higher. Therefore, 3-layer system, the actual total memory consumption was not too
network model network structure was chosen in this much when the network was trained by this method.
yarn-quality predicting system, which consisted of an The network model has been built completely after the
input layer, a hidden layer and an output layer. above design, according to which the topology map could
be drawn in Figure 2. The BP network could be divided
A. Neurons Number in Input Layer
into three layers: the input layer, the hidden layer, and the
This system was mainly used to predict the yarn output layer. There were 6 nodes in the input layer, an
tenacity and its evenness CV% under any other conditions adjustable number in the intermediate hidden layer, and 2
unchanged, such as processing, temperature and humidity. in the output layer.
To reach the above objective, the yarns quality was
predicted through inputting six indicators of fiber quality
such as impurity rate, the principal length, the percentage
of short fiber, the degree of maturity, fiber strength and
Micronaire [4].Firstly, the 11 fiber quality indicators the
spinning plants often measured were statistically analyzed
and then the experts' opinions were gotten in this field,
finally six neurons were taken in this layer to meet the
requirement that the network neurons number should
correspond to the fiber index number.
B. Neurons Number in Hidden Layer
There was no effective principle to follow to
determine the neurons numbers in this layer, so the Figure2. The BP Net Structure
optional one was mainly determined by training and
contrast on different neurons numbers. It was revealed that IV. SYSTEM DEVELOPMENTS
the network predicting accuracy was highest when the
neurons number in this layer was 16 through training and 1. The Neural Network Construction Module
contrast in the course of modeling the network. However, Developments
this was only for the training data used by the author, if According to the requirements of network mode
the training sample changed, the optional number of mentioned above, this part was developed by Matlab
neurons would have a corresponding change, therefore the neural network toolbox, in which the Newcf. Function was
neurons number in hidden layer could be set to a variables called to design the BP network; finally the designed
with a recommended value 16 in the quality predicting network was packaged into the same function file named
system, and the best suited hidden nodes number for net.m together with the data pre-designed program and the
samples could be determined by the system operator network training program. During the construction and
through training and contrast. training process, the network was activated by transferring
the external data information and the design of the
C. Neurons Number in Output Layer network nodes of hidden layer into the internal of function
The yarn tenacity and evenness CV % were chosen for file through a call to this function file, after that the
the predicting output items in this system. As a result, the network training was completed.
corresponding output layer neurons number in the network
2. The Modules Developments on Network
was taken as 2.
Training and Yarn Quality Prediction
D. Activation Function of Each Layer These two parts were developed with the technique of
The common hyperbolic tangent function VB and Matlab merging programming, the running of
−2 n −2 n
f (n) = (1 − e ) (1 + e ) in BP network was adopted Matlab application program was controlled by the master
control program of VB and the data transfer was carried
as activation function in the hidden layer. This function
out between the environments of VB and Matlab.
could compress each input to a value between -1 and 1;
while the linear function f (n) = w×n +bcould be adopted
A. The Realization Base of VB and Matlab
Merging Programming
as the activation function in the output layer [5].
ActiveX Automation protocol is a protocol that allows
E. Network Training Algorithm one application program (the control side) to control
Levenberg-Marquardt algorithm was adopted to train another (the server side). As VB could supports the
the network model in the system to overcome the ActiveX automation control side protocol and Matlab was
disadvantages of criteria BP algorithm, such as long able to support automated server-side protocol, therefore
training time and vulnerability to local optimum. It has when the ActiveX connection established between them.
been proved that the improved algorithm could markedly The Matlab command could be called directly in VB
environment, and then the realization of data transfer object named MT being created in the VB environment,
between the two programs followed. the trained network model the user specified was loaded to
Matlab environment through a call for Matlab to
B. The Network Training Module Developments
implement the "LOAD" command by MT.Execute
The function this module require to be achieved was method;Then fiber index value was transferred into
completing the network training and the retention of the Matlab environment from the VB environment by
trained model on the basis of transferring the input MT.Putfullmatrix method and passed to the loaded
messages from the network training interface in VB network mode after being pre-processed, then the yarn
environment to the network models in Matlab quality was predicted through a call for the neural network
environment as well as making them activated. The emulation function "SIM". Finally the predicting results
realization processes were as follows: First, a Matlab. obtained were processed, transmitted in the VB
application object named MT was created in the VB environment by MT.Getfullmatrix method and displayed
environment so that the ActiveX connection could be on the quality-predicting interface [7].
achieved between VB and Matlab ; then the activation and
training of the network could be realized on the basement V. MODEL LIBRARY CONSTRUCTING
of transferring the input messages from the network Cotton yarns with different count were collected from
training interface in VB environment to Matlab several cotton mills within a year, along with the
environment by MT.Putfullmatrix method, the call for corresponding raw cotton information indicators. After
Matlab by MT.Execute method to implement the net.m being analyzed and processed, the data could be used to
function files and the transfer of data information and the train the network models constructed by this system, and
setting to the network hidden layer nodes to the network then all the trained models were saved in model database
through function parameters latter ; After training, the for the yarn quality prediction.
trained network model was saved into model base, which
was available to be called by yarn quality prediction VI. THE INSTANCE AND ANALYSIS
module through a call for Matlab program to implement The model"JC18.2T" was called from the model
the "SAVE" order by MT.Execute method [6]. library to predict the quality of the JC18.2T yarn produced
C. The Yarn Quality Predicting Module by a factory. The properties of the raw cotton the yarn
used were summarized in Table 1, while both the
The function this module to be realized was
predictive values and the measured values were shown in
transferring the fiber quality index value from the
Table 2; As it could be seen in Table 2, the yarn tenacity
quality-predicting interface in VB environment to the
trained network model that loaded in Matlab environment and its evenness CV%were predicted with the relative
to realize the prediction and not only returning the error of less than 4% by the trained model. Since the
prediction results to VB environment but also displaying predicting accuracy was in an acceptable range, the
them on quality prediction interface, the above functions predicting results would be useful for guiding yarn
was realized as follows: after the Matlab application -spinning practice.

TABLE I. THE PROPERTIES OF COTTON FIBER

Group Impurity Principle Length Short Fiber Content Strength


Maturity Micronaire
Number (%) (mm) (%) (cN)
1 1.34 30.03 9.75 1.44 4.5940 3.84
2 1.06 30.18 9.46 1.40 4.4523 3.84
3 1.22 30.18 11.00 1.39 4.4113 3.84
4 1.17 30.21 10.53 1.40 4.4021 3.79
5 0.98 30.10 9.50 1.38 4.5115 3.73

TABLE II. THE TABLE OF COMPARISON BETWEEN PREDICTING VALUE AND MEASURING VALUE

Predictive Value Measured Value Relative Error


Group
Number Strength CV % Strength CV % Strength CV %
(cN·tex-1) (%) (cN·tex-1) (%) (%) (%)

1 15.7 12.6 15.5 12.5 +1.29 +0.80


2 15.3 12.6 14.9 13.0 +2.61 -3.08
3 14.3 12.8 14.8 12.9 -3.38 +0.78
4 14.5 12.7 14.8 13.0 -2.03 -2.31
5 14.6 12.9 14.1 12.8 +3.55 -0.78
VII. CALCULATIONS
In order to predict the yarn quality through digital
simulation, it is essential to predict the yarn quality with
higher accuracy in the spinning process. In this paper, a
further exploration was made to this part and artificial
neural network model was built to predict yarn strength
and evenness. And also it was verified that the system
did the good job for predicting yarn tenacity and
evenness CV% exactly with the relative error of less
than 4% after the model being trained. The accuracy can
meet the demand of spinning factories and so the
predicting results would be useful for guiding the
spinning practice.
REFERENCES
[1] Xiangsheng Huang,Hanbin Xiao.Application of Wavelet
Analysis and ANN in Ultrasonic Quantitative Testing[J],Port
Engineering Technology,2008(6):13-15.
[2] Bao Liu, Gang Wu. Surface Prediction Model Based on BP
Neural Network [J].Heat Treatment Technology and Equipment,
2006(6):37-40.
[3] Xiancai Gui. Realization of BP Networks and Their Application
on MATLAB [J].Journal of Zhanjiang Normal College, 2004, 25
(3):79-84.
[4] Xiaomei Zou, Guan Liang. The Relationships of Performance of
Rraw Cotton,Spinning Process and Yarn Quality[J]. Inspection
and Quarantine Science, 2005, (12):143-144.
[5] Cong Shuang.Matlab Neural Network Toolbox and its
Application [M]. University of Science and Technology of China
Press.2003-5 (2).
[6] Zhisong Zhu, Xiaoli Guo. Research on Merging Programming of
VB and MATLAB [J]. Computer Applications, 2003(9):18-19.
[7] Li Feng, Ju Li. Method of Hybird Programming with VB and
Matlab[J], Software Technology, 2005 (5):110-112.

S-ar putea să vă placă și