Weaving & Knitting

Abstract

In this work, bursting strength and pilling rating of polyester-cotton fibre blended three thread fleece fabric was predicted using Artificial Neural Network (ANN) Back Propagation (BP) model. Bursting strength and pilling has been predicted from 50 three thread fleece fabrics with different Stitch length, Grams per square meter (GSM), yarn counts and twist per inch. All three thread fleece fabrics has their own Bursting strength and pilling rate which was used for the prediction. To validate the two models in the training steps, training precision, and simulation precision, 40 fabrics were used for training and 10 fabrics were used for testing, and the predicted pilling property was obtained. The results shows that the predicted value of bursting strength and pilling rating are closer to the experimental values which are determined by ANN Back propagation (BP) model. This shows that optimized model with BP can predict the bursting strength and pilling rating of polyester–cotton three thread fleece fabrics with acceptable accuracy.

Key words: Three thread Fleece Fabric, ANN, Back Propagation, Prediction, Bursting strength, Pilling

Introduction

Three-thread fleece fabric is widely used as winter clothing where both strength and pilling. Pilling is a fabric surface property with small tightly entangles balls of fibres that occurs as a result of friction of protruding fibres on the surface [1]. Pilling creates discomfort for the user and also deteriorates the outer look of garment. The tendency of pilling formation is associated with the yarn properties like stitch length, twist per inch, yarn count, and yarn type [2]. Fabrics with less twist per inch in yarns shows more pilling. Uncombed yarns also create excessive pilling due to their hairy surface on the yarn body [3]. Bursting strength is a key measure of knit fabric strength that indicates durability of garment to be produced by the fabric. Bursting strength indicates how much pressure a fabric can withstand before it breaks by air pressure. The bursting strength property of fabric can also be associated with stitch length twist per inch, yarn count, and fabric grams per square meter (GSM). Higher GSM fabrics exhibit higher bursting strengths [4]. The impact of GSM on knit fabric pilling is negligible. Higher GSM of fabric would produce small percentage increase in pilling. Applying heat setting and singeing can effectively improve knit fabric pilling. In recent, biopolishing is an enzymatic process used to pill resistant fabric [5]. Moreover, it is challenging to produce pill resistant three thread polyester cotton blended fleece fabric because of cottons and polyester blends. The resistance of knitted fabrics to pilling is dependent on the density of the fabric; that is, the stitch length. Resistance to pilling increases as the surface density of the knitted fabric increases and the stitch length decreases [6]. In the analysis of the relationship between knit structure and pilling propensities, it was found that knitted fabrics with rib structure were the most resistant to pilling, knitted fabrics with interlock structure were less resistant, and woven fabrics with plain weave were more likely to pill. Any fabric, knit or woven, is more susceptible to pilling if the number of yarns per unit length of the fabric reduces [7]. Spinning process has technological advancement such as compact and vortex spinning for the reduction of pilling andbursting strength of the fabric. Vortex yarn structure is very compact and the outer fibres are lodged in the center of the yarns, so that the outer surface fibre cannot form pill easily [8]. Efficient use of textile fibres, process machinery and technical parameters, it is necessary for the manufacturer to produce fabrics or garment as per customers specification and for that prediction of fabric or garment parameters from process variables is essential now a days. Statistical regression can provide predicted value as a fundamental method [9]. Predicting using advanced computational techniques like artificial neural networks (ANN) offers a promising solution for manufacturers for process optimization 10].The main objective of artificial intelligence like ANN in textile industries is to create a system that is capable of thinking and acting like humans. Backpropagation with a variable learning rate and several linear regressions are being used in the ANN methodology. ANN is recently used in in textile for the prediction of fabric defects [11].

Developing a predictive model using ANNs can streamline the process, allowing manufacturers to anticipate how different fabric construction parameters affect both bursting and pilling performance. ANNs were also used for predicting different fabric properties for several fabric types [12]. But still now no research has been done on predicting pilling rate and bursting strength for three thread fleece fabric. With a view to fulfil the gap of this field of study on three thread fleece fabric, this work has been conducted in order to develop an artificial neural network-based model capable of predicting the bursting strength and pilling rating of three thread fleece fabrics and to design and train an ANN model using experimental data to predict these fabric properties with high accuracy, and also analyze the effect of hyperparameters like number of hidden layer nodes, number of epochs and loss functions on model accuracy. 

2. Methodology

2.1 Raw materials

The polyester-cotton blended three thread Fleece Fabric where the first thread fibre type was polyester and cotton and their blend ratio 40% polyester and 60% cotton. The second thread fibre type was 100% polyester, and third thread fibre type was 80% cotton and 20% polyester. 50 different three thread Fleece fabrics were collected from a local garment factory of Bangladesh. The fabrics were different in terms of their parameters. Yarns and Fabric parameters of those fabrics are summarized as their lowest and highest values in Table 1.

2.2 Analytical methods

Backpropagation (BP) Model model mainly consists of a multilayer forward neural network. There are two parts in the backpropagation model training and testing parts. The 40 sets of data are used for training the model, and 10 sets of data are allocated for testing the model. Four physical factors, yarn count, twist, GSM, Twist are taken as the training input vector. Therefore, the number of input nodes (m) of training is 4. The number of pills and the busting strength are used as the learning target data and therefore the number of nodes in the output node (n) is 2, and what will be the number of nodes in the hidden layer (N) is calculated using the equation no. 1 considering the constant (a) [1,10] To minimize the errors the same samples with different node numbers should be trained and the mean absolute percent error (MAPE) the error, root mean square error (RMSE), and mean absolute error (MAE) are calculated as per following equations.

Where abs denote the absolute value, the output value of the model is denoted by A(i), the experimental value is denoted by B(i), the number of samples are denoted by k.

2.4 Process Flowchart

Process flowchart of ANN model developed to predict pilling rating and bursting strength. Python coding to develop ANN model can be found as supplementary information.

3. Results and Discussion

3.1 Effect of the number of hidden layer nodes

Pytorch 3.0 and Pylance is used to develop and analysis of Neural Network model. As per equation (1), the value range of the hidden layer node is [5, 14]. Errors of different hidden layer nodes for Bursting Strength and.Errors of differenthidden layer nodes for Pilling Rating is tabulated in Table 3 and Table 4 respectively.

It is evident that error functions of bursting strength show its minimum for MAPE at node 5, RMSE at node 8 and MAE at node 13. Similarly, error functions of pilling rating show its minimum for MAPE at node 5, RMSE at node 11 and MAE at node 11. The reason of this is the too small number of hidden layer nodes. Following the same, for nodes 11 and 12 of Bursting strength and nodes 7 and 8 for Pilling, the error values rose.

3.2 Effect of epoch number and loss function

The performance function that comes with the neural network toolbox is the MAE and RMSE. The Y-axis of the error curve of the training is mean absolute error. From the Figure, the training step of BP neural network for bursting strength is 820. On the other side, it is seen from Figure that the training step of BP neural network for pilling rating is990. This result indicates that the network reaches the target value using mean absolute error function for bursting strength and using root mean square error function for pilling rating.

Predicting of Bursting strength of three thread fleece fabric from the input data using ANN Back Propagation (BP) model to determine the mean absolute error and root mean square error, it is seen that that close predicted value has been found MAE function as well as RMSE function have predicted very well. Pilling of three thread fleece fabric is also determined by ANN Back Propagation (BP) model by calculating some functional measure such as MAE and RMSE error. From the graphical presentation, RMSE function predicted more close pilling value than MAE function.

4. Conclusion

In his study, bursting strength and pilling rating of polyester and cotton fibre blended three thread fleece fabrics was investigated using Artificial Neural Network (ANN). Back Propagation (BP) model has predicted the bursting strength and pilling from 50 sets of three thread fleece fabrics with respect to their four input (TPI, Count, Stitch Length and GSM) and 2 output parameters (Bursting strength and Pilling rate). The results shows that the predicted value of bursting strength and pilling rating are closer to the experimental values. This study shows that optimized model with BP can predict the bursting strength and pilling rating of polyester cotton fibre blended three thread fleece fabrics with acceptable accuracy.

References

[1] Okubayashi, S., Campos, R., Rohrer, C., & Bechtold, T. (2005). A pilling mechanism for cellulosic knit fabrics – effects of wet processing. The Journal of The Textile Institute, 96(1), 37–41.

[2] Daiva MIKUČIONIENĖ, The Influence of Structure Parameters of Weft Knitted Fabrics on Propensity to Pilling, TEXTILE MATERIALS, vol. 15, no. 4 (2009).

[3] Ukponmwan, J. O., Mukhopadhyay, A., & Chatterjee, K. N. (1998). PILLING. Textile Progress, 28(3), 1–57.

[4] Islam, A., Billal Hossain, M., Haq, E., Saber Shravan, A., & Rahman, A. (2022). Factors Influencing Bursting Strength of Single Jersey Knitted Fabrics. European Scientific Journal, ESJ, 18(36), 68.

[5] Mccloskey SG, Jump JM. Bio-Polishing of Polyester and Polyester/Cotton Fabric. Textile Research Journal. 2005;75(6):480-484.

[6] Uyanik, S., & Topalbekiroglu, M. (2017). The effect of knit structures with tuck stitches on fabric properties and pilling resistance. The Journal of The Textile Institute, 108(9), 1584–1589.

[7] Azita Asayesh and Fatemeh Kolahi Mahmoodi, “The effect of fabric structure on the pilling and abrasion resistance of interlock weft-knitted fabrics” , International Journal of Clothing Science and Technology Vol. 36 No. 2, 2024 pp. 287-303.

[8] Kim, H. A. (2017). Physical properties of ring, compact, and air vortex yarns made of PTT/wool/modal and wearing comfort of their knitted fabrics for high emotional garments. The Journal of The Textile Institute, 108(9), 1647–1656.

[9] Hisham E Eltayib , Akram H M Ali , Isam A Ishag, “The Prediction of Tear Strength of plain weave fabric Using Linear Regression Models”, International Journal of Advanced Engineering Research and Science Vol-3, Issue-11, 2016.

[10] Rolich, T., Šajatović, A.H. & Pavlinić, D.Z. Application of artificial neural network (ANN) for prediction of fabrics’ extensibility. Fibers Polym 11, 917–923 (2010).

[11] Kuo C-FJ, Lee C-J. A Back-Propagation Neural Network for Recognizing Fabric Defects. Textile Research Journal. 2003;73(2):147-151.

[12] Pamuk G. Multi Response Optimization for Bursting Strength and Pill Density of Lacoste Fabrics. Journal of Engineered Fibers and Fabrics. 2015;10(1).