Escolar Documentos
Profissional Documentos
Cultura Documentos
http://trj.sagepub.com Predicting the Seam Strength of Notched Webbings for Parachute Assemblies Using the Taguchi's Design of Experiment and Artificial Neural Networks
Levent Onal, Mithat Zeydan, Mahmut Korkmaz and Sheik Meeran Textile Research Journal 2009; 79; 468 DOI: 10.1177/0040517508099921 The online version of this article can be found at: http://trj.sagepub.com/cgi/content/abstract/79/5/468
Published by:
http://www.sagepublications.com
Additional services and information for Textile Research Journal can be found at: Email Alerts: http://trj.sagepub.com/cgi/alerts Subscriptions: http://trj.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.co.uk/journalsPermissions.nav Citations http://trj.sagepub.com/cgi/content/refs/79/5/468
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
Article
Predicting the Seam Strength of Notched Webbings for Parachute Assemblies Using the Taguchi's Design of Experiment and Artificial Neural Networks
Abstract
Levent Onal1
Erciyes University, Department of Textile Engineering, 38039 Kayseri, Turkey
Webbings are used in parachute assemblies as reinforcing units for the strength they provide. The strength of these seams is an important characteristic which has a substantial influence on the mechanical property of the parachute assemblies. It is well established that factors such as fabric width, folding length of joint, seam design and seam type will all have an impact on seam strength. In this work, the effect of these factors on seam strength was studied using both Taguchis design of experiment (TDOE) as well as an artificial neural network (ANN). In TDOE, two levels were chosen for the factors mentioned above. An L8 design was adopted and an orthogonal array was generated. The contribution of each factor to seam strength was analyzed using analysis of variance (ANOVA) and signal to noise ratio methods. From the analysis it was found that the fabric width, folding length of joint and interaction between the folding length of joint and the seam design affected seam strength significantly. Further, using TDOE, an optimal configuration of levels of factors was found. In order to contrast and compare the results from TDOE, an ANN was also used to predict seam strength using the above mentioned factors as inputs. The prediction from TDOE and ANN methodologies were compared with physical seam strength. It was established from these comparisons, in which the root mean square error was used as an accuracy measure, that the predictions by ANN were better in accuracy than those predicted by TDOE.
Mithat Zeydan
Erciyes University, Department of Industrial Engineering, 38039, Kayseri, Turkey
Mahmut Korkmaz
Erciyes University, Mustafa Cikrikcioglu Vocational School, 38039 Kayseri, Turkey
Sheik Meeran
School of Management, The University of Bath, Bath, U.K.
Key words webbing, parachute, seam strength, artificial neural network, Taguchis design of experiment
Corresponding author: tel: +90 (352) 4374901 ext.32877; fax: +90 (352) 4375784; e-mail: lonal@erciyes.edu.tr
Textile Research Journal Vol 79(5): 468478 DOI: 10.1177/0040517508099921 Figures 68 appear in color online: http://trj.sagepub.com
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
Predicting the Seam Strength of Notched Webbings for Parachute Assemblies Using L. Onal et al. Webbings have been used in different forms as reinforcement in different types of parachute assemblies, such as paratrooper parachute, plane decelerating parachute, cargo decelerating parachute and decelerating parachute for cluster bombs. Webbings support the canopy fabric of parachutes. Depending on the type and end use of parachutes, the number and location of the webbings on the canopy of the parachutes vary. To our knowledge and understanding, there are no publications which use both Taguchis design of experiment (TDOE) and artificial neural network (ANN) to predict the strength of webbings and then compare them with the strength physically obtained from mechanical tests on notched webbing specimens. In any case, there are only a handful of studies that attempt to predict webbing strength irrespective of the method employed. However, there have been various studies looking at different aspects which either relate to or lead to predicting webbing strength. Many studies can be found on predicting fabric strength. These studies in general have concentrated on matters relating to the canopy of parachutes [19]. Edberg [1] studied the stress-strain curves of ultraviolet-aged nylon parachute fabrics where as Ericksen [4] studied the effects of folding on the strength of parachute materials. Li et al. [8] and Lee and Li [9] made a three-dimensional analysis of various parachutes, which included study of parachute fabrics. Of the very few studies on webbings in parachute assemblies [10, 11], Ericksen et al. [10] evaluated the effect of warp and weft variables on the strength of joints in Kevlar webbings. Onal [11] studied the mechanical properties of notched webbings in cluster sub-munition parachutes. He also studied the effect of warp density and fiber type on the load extension property of notched webbings. On the wider aspect of using TDOE, there have been only a few publications on the application of TDOE in textile technology. Park and Ha [12] applied TDOE for optimizing sewing conditions in woven garments. Liasi et al. [13] used the TDOE method for finding the optimum needle temperature in sewing. On the other hand, ANN has been used in various areas of textile technology for nearly 10 years. Park et al. [14] applied fuzzy logic and ANNs for the evaluation of knitted fabrics. Chen et al. [15] focused on predicting the fiber diameter of melt-blown non-wovens by supplying processing parameters as input variables to their ANN models. An ANN was also used by Majumdar et al. [16] for predicting spinning quality index and fiber micronaire from the given fiber properties. Majumdar and Majumdar [17] employed an ANN model for predicting the breaking elongation of ring spun cotton yarns. They used yarn counts and cotton fiber properties as inputs to their models. Yao et al. [18] investigated the predictability of warp breakage rate from a sizing yarn quality index using a feed-forward back-propagation ANN. Tilocca et al. [19] presented a new approach to automatic fabric inspection based on an optical acquisi-
469
TRJ
tion system and an ANN to analyze the acquired data. ANN methodology has been also successfully used for predicting the strength of fabrics. Ertugrul and Ucar [20] used an ANN to predict bursting strength of cotton plain-knitted fabrics. Ogulata et al. [21] predicted elongation and recovery test results of woven stretch fabric for warp and weft direction using different test points with regression and ANN models. Chattopadhyay and Guha [22] reviewed extensively the application of ANN to various research areas of textile science. Zeydan [23] compared consistency and efficiency of both ANN and TDOE methodologies to predict woven fabric strength. Although webbings have been extensively used in parachute assemblies, no specific study has been reported on the seam strength of notched webbing. Hence in this work, the seam strength of notched webbings for the parachute assemblies is addressed. The effect of factors such as fabric width, folding length of joint, seam design and seam type on seam strength was studied using both TDOE as well as an ANN. In TDOE, two levels were chosen for the factors mentioned above: an L8 design was adopted and an orthogonal array was generated using MINITAB (Minitab Inc. (2003) statistical software, Release 14 for Windows, State College, Pennsylvania, U.S.A.). The contribution of each factor to seam strength was analyzed using analysis of variance (ANOVA) and signal to noise ratio (S/N) methods. Further, an optimal configuration of levels of factors was found using TDOE as well as an ANN using the above mentioned factors as inputs. The prediction from TDOE and ANN methodologies were compared using root mean square error (RMSE) as an accuracy measure against the physical seam strength, which was found through conducting mechanical tests on specimens.
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
TRJ
470
Figure 1 Schematic view of sample, where the hole is at the centre of a folded section of combined narrow fabric.
in the Federal Specification V-T-295D. The summary of which is, sewing nylon yarn, Type-1 Size-3, is a three-plied yarn having 2 twist/cm, a breaking force of 120 N and a yarn count of 205 Tex. Binding thread is Type-1 Size-E yarn, which is a three-plied stitch yarn having 2.4 twist/cm, a breaking force of 40 N and a yarn count of 68 Tex. Properties of webbings are given in Table 1. As mentioned earlier, this work is about predicting seam strength using TDOE and ANN methodologies and verifying and comparing the results obtained by these methodologies against the physical strength, which is obtained by experimental measuring. The experimental measuring of seam strength is carried out as follows. Samples were notched with 4.5 mm holes at the center of the seam zone (as shown in Figure 1) by a sharp pin acting swiftly in order not to cause any deformation of the fabric. Seam strength of the samples were tested with Zwick Universal tester, which used a loading rate of 305 mm/minute. Breaking force (FT) and elongation were recorded for further analysis. Sewn seam strength was calculated using equation 1 given below. Whenever there was a slippage of fabric from the tester jaws that was also recorded. All tests were performed under standard atmospheric conditions (65+/2% relative humidity and 202oC temperature) and samples were conditioned for 24 hours before testing. Five samples were tested in each group and the average was recorded; the group being the one identified by the L8
experiment. None of the samples failed at or close to the grip region. FT F S = 1000 ----w
(1)
where FS is sewn seam strength, FT is the seam breaking force of each sample, and w is specimen width.
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
Predicting the Seam Strength of Notched Webbings for Parachute Assemblies Using L. Onal et al.
471
TRJ
be seen that there are four major categories that influence the seam strength. Performing a TDOE on all of the parameters given in Figure 3 will be unwieldy. Hence, all individual parameters within these categories are kept constant apart from the sewing parameters. This led to selection of the key factors and levels of the sewing parameters used in setting TDOE in this work. Further steps after choosing the key factors and levels in establishing TDOE are: selecting an appropriate orthogonal array, collecting the data related to the groupings according to the scheme
of orthogonal array, performing ANOVA for identifying significant key factors, finding the optimal combination of factor levels to maximize (optimize) the objective metric, in this case, seam strength of notched webbings. Key factors (sometimes called control factors) are the factors which have an influence on the outcome/design parameter/product/process. The factors chosen here are fabric width, fabric length of joint, seam design and seam type, as they have an impact on seam strength, the design parameter of interest and are specified in Table 2. In the
Figure 3 Fishbone (Ishikawa cause-effect) diagram for factors affecting seam strength of webbing.
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
TRJ
472
Textile Research Journal 79(5) putation, is a mathematical/computational model based on biological neural network. Most widely used ANN algorithm is feed forward back-propagation algorithm proposed by Rumelhart and McClelland [25], while the most popular architecture of the ANN is a multilayer network consisting of input, hidden and output layers. The capability to learn from examples and the ability to generalize are the most important features of back-propagation ANN (BP ANN), which are achieved by the learnt information stored as weights across the network. BP ANN has a great potential in many industrial and scientific areas where prediction has to be made based on existing patterns and hence used in this work as one of the methods to predict seam strength. A three-layer, feed-forward BP ANN model having four input neurons and five hidden neurons was used in this study. During training, the output is compared with an expected value and the mean square error is calculated. If the mean square error is more than a prescribed limiting value, it is used as a trigger to back-propagate a correction of weights from output to input. The weights are further modified until the error or number of iterations is within a prescribed limit. Mean square error, EP, for pattern p is defined as
n
Taguchis design of experiment, two levels, namely level 1 (minimum value) and level 2 (maximum value) are considered for each of the above factors. An L8 experiment with two levels for each factor and seven degrees of freedom was found be appropriate. Every experiment (corresponding to a row in the orthogonal array) was repeated five times and performance variance is captured in terms of S/N ratio. In general, S/N ratio is used as a measure of performance variability of products/processes (that is, robustness in the presence of noise factors). In statistical term S stands for mean and is called signal and N stands for standard deviation and is called noise. Mathematically, the S/N ratio is derived from quadratic loss function. The S/N data for all the experiments given by orthogonal array was analyzed using mains effect diagram and ANOVA to identify the significant key factors which have maximum impact on the variance of the characteristic of interest, that is, seam strength. The overall aim of any TDOE is to find factor levels that maximize S/N ratio, which suggests a lower variability. This leads to the premise: the better the S/N ratio, the better the quality and hence the effort to increase the S/N ratio in any design. In general, the S/N ratio could be formulated on three modes: smaller is better, nominal is better, larger is better. As suggested above, the larger is better mode is found to be appropriate in this design of experiment. On this basis the following formula is used to calculate S/N ratio:
n 1- --2 y i=1 i S N = 10 log ------------ n
Ep =
i=1
-(D -2
pi
O pi )
(3)
where Dpi is the target value, and Opi is the computed output for the ith pattern. Also we have Ep ( t ) - + W ( t 1 ) W ( t ) = --------------W
(4)
(2)
where n is the number of repetitions for an experimental combination, i is a numerator, and yi is a performance value of the ith experiment. S/N ratio is abbreviated as SNR.
where is the learning rate, which is in the interval of 0 < < 1; is the momentum coefficient, which is in the interval of 0 < < 1; EP(t) is the error that varies with time t; and W is the weight of the connections between neurons in different layers. The nature of the training depends partially on the choice of activation function. In order to generate the optimum activation function for the study, tanh, linear, Gaussian, and sigmoid functions were experimented in the QwickNet v2.23 software. Sigmoid activation function used in hidden and output layers resulted in faster training. The sigmoid activation function is given in equation 5. Figure 4 was drawn varying constant, c. 1 . y = ------------------- ( cx ) 1+e (5)
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
Predicting the Seam Strength of Notched Webbings for Parachute Assemblies Using L. Onal et al.
473
TRJ
into the TDOE module of MINITAB to start the process of experiment. The physical manifestation of the orthogonal arrays can be seen in the columns headed A, B, C, D in Table 3. The data collection was replicated for five cycles and results are given in Table 3. S/N, mean, standard deviation (given in parentheses in Table 3) were calculated for these data using MINITAB. It can be also seen from Table 3 by varying factors A, B, C, and D, the variance of the process (which is given in terms of S/N) could be controlled and in turn could be reduced. From further TDOE analysis using the same software, it was found that factor A (fabric width) and factor B (folding length of joint) were significant based on ANOVA and SNR calculations with their P values less than 0.05 as shown in Table 4. However, factor C (seam design) and factor D (seam type) were found to be not significant even at 90%, 95% and 99% significance level as F values for factors C and D are less than critical value for F as given in the F-table (F0.1,1,3 =5.54). This is confirmed by the main effect plots for SNR (with larger is better mode) shown in Figure 5, derived from Table 5, in which the effect of each factor for seam strength at levels 1 and 2 as well as the corresponding delta values are given. The delta value is the difference between level 1 effect (response) and level 2 effect, which in turn represents the net effect. A higher delta value
Table 3 Experimental results for five cycles including SNR and mean value.
Trial 1 2 3 4 5 6 7 8 A 10 10 10 10 13 13 13 13 B 15 15 20 20 15 15 20 20 C X X II II II II X X D Lsb Lsc Lsb Lsc Lsb Lsc Lsb Lsc 1. Cycle 39799.2 59801.8 71897.5 80540.1 62670.8 73160.0 75303.1 84034.0 2. Cycle 35353.7 50649.8 61606.8 70684.8 60889.9 62708.5 77159.4 90402.9 3. Cycle 35587.7 52309.9 58452.5 57803.5 63508.4 74231.5 75559.6 71092.3 4. Cycle 32848.4 48408.6 62255.8 64708.3 61463.4 75687.9 77732.9 76563.3 5. Cycle 30561.9 51087.5 54528.5 65742.1 60497.5 63666.9 75008.8 87007.2 Mean 34830 (3453) 52451.5 (4344) 61748.2 (6447) 67895.8 (8431) 61806 (1256) 69891(6834) 76152.8 (1348) 81819.9 (7877) SNR 90.73 94.33 95.70 96.48 95.81 96.80 97.63 98.15
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
TRJ
474
means that the difference at the selected level for a given factor is highly pronounced and in turn, the level change of this factor has an impact on seam strength. As can be seen, the delta values for factors A and B (occupying ranks 1 and 2) are higher than the values for factors C and D. The TDOE analysis further suggested that the optimum combination of factors (which will give optimum seam strength) is A2B2C2D2, which translates into fabric width of 13 mm, folding length of joint of 20 mm, seam design of II, and seam type of LSc.
In order to predict the usefulness of the TDOE model, which has just been established, additional trials with different combinations of levels of factors were carried out on additional samples as shown in Table 6. Within this set of additional tests, the optimal configuration given by TDOE, namely A2B2C2D2, is also given. In Table 6, the mean value for each of the trials is given along with standard deviation for each configuration. These values are compared against the values predicted by MINITAB for these configurations to validate the predictions by the TDOE. As the optimum configuration is A2B2C2D2, it may inadvertently imply that each factor at level 2 will have a better impact on seam strength. From inspection of Tables 3 and 6, it can be seen that this is not the case, that is, level 2 of each factor does not always give the higher performance (seam strength). This indicates that it is important to look at the interaction between factors and their change in levels. In order to find which interactions between factors are significant, the main interaction plot of seam strength was
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
Predicting the Seam Strength of Notched Webbings for Parachute Assemblies Using L. Onal et al.
475
TRJ
drawn using MINITAB. TDOE suggests studying the interactions between two single factors only. The interactions between each two factors (all possible combinations) are given in Figure 6. An interaction is said to exist if the effect of change in level of one factor is opposite to the effect of change in level in another factor. In Figure 6, if the line connecting levels 1 and levels 2 of both factors are parallel or almost parallel, this indicates that there is no meaningful interaction taking place between these factors. No interaction means the effect of the first factor remains the same regardless of the level of the second factor and vice versa. From this standpoint, there are no meaningful interactions between factors A and B (A*B), A and D (A*D), B and D (B*D), and C and D (C*D). An interaction means the effect of the first factor (factor indicated in the row in Figure 6) depends on the level at which the second factor (factor indicated in the column in Figure 6) is used and vice versa. The greater the skew between the lines, then the greater the strength of the interactions between factors. Hence from Figure 6, it can be seen that there are interactions between factors A and C (A*C) and B and C (B*C). In order to demonstrate the strength of interactions seen in Figure 6, an ANOVA test was performed using 95% significance level. Tables 7 and 8 show B*C interaction is significant, while A*C interaction is not significant, because the P value in the first case is less than 0.05, and in
the second case it is more than 0.05. This result indicates that interactions of factors B and C should be considered when the joining of notched webbing is planned.
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
TRJ
476
Textile Research Journal 79(5) confidence level (CL99.9%). The test results for five samples were 94032.6, 93081.8, 81973.9, 91014.2, and 92515.8, respectively. All these values lie in the confidence interval of TDOE predicted value (87698.12 7503.8). Therefore, the experimental design could be accepted as satisfactory and valid.
(6)
where Ve represents error variances of mean standard error, F, 1, n is the F value from the F-table for significance level , 1 degree of freedom, and n sample size; Ne is the effective number of replications. For the present study, mean standard error = 34919131, Ne = 8, table value of F0.001,1,35 = 12.90, and hence CL99.9% = 7503.8. This suggests the actual experimental seam strength could vary within 7503.8 of the predicted value of TDOE at 99.9%
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
Predicting the Seam Strength of Notched Webbings for Parachute Assemblies Using L. Onal et al.
477
TRJ
Figure 8 The comparison of experimental values with predicted values derived from TDOE and ANN.
the ANN, respectively. Both the data sets are given in Tables 9 and 10. The test data set was used to evaluate the prediction capability of ANN as well as TDOE. This evaluation was made against the experimental data obtained for specimens prepared for the parameters specified by the ten patterns in the test data set. It is worthwhile noting that the same experimental data set was used to assess the TDOE result.
Conclusion
The TDOE and ANN approaches were successfully applied to predict the seam strength of webbings used in many parachute assemblies. The effects of fabric width, folding length of joint, seam type and seam design on the seam strength were analyzed. An L8 orthogonal array was used in planning the Taguchi experiments in this work. The TDOE revealed (based on SNR performance criteria) that fabric width,
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009
TRJ
478
folding length of joint and the interaction of folding length of joint and seam design significantly affect seam strength. The TDOE was also used to find the optimal configuration, which was A2B2C2D2: this corresponds to a fabric width of 13, folding length of joint of 20, seam design of II, seam type of LSc. The TDOE and ANN predictions were found to be satisfactory for several of the patterns tested. In addition, the prediction by the ANN model was compared with that of TDOE. It was found that ANN had better predictions of the experimental seam strength of jointed notched webbing than the TDOE predictions. In these comparisons, RMSE values were used as comparative metrics. As a result, it can be said that ANN appears to be a reliable and useful tool in characterizing the effect of some critical manufacturing parameters on the seam strength of webbing, if a sufficient number of replicated experimental data are available to train the ANN.
11.
12.
13.
14.
15.
Acknowledgements
The authors are grateful to Dr Kerim Guney, Dean of Engineering, Dr Kadir Bilisik, Head of the Textile Engineering Department and Dr Nafiz Kahraman, Head of the Mustafa Cikrikcioglu Vocational School for their approval to carry out the research.
16.
17.
Literature Cited
1. 2. Edberg, B., An Investigation of Stress-strain Curve of Nylon Parachute Cloth, Text. Res. J., 40, 665676 (1970). Ericksen, R. H., Davis, A. C., and Warren, W. E., DeflectionForce Measurements and Observations on Kevlar 49 Parachute Fabrics, Text. Res. J., 62, 628637 (1992). Peterson, C. W., Pepper, W. B. D., Johnson, W., and Holt, I. T., Application of Kevlar to Parachute System Design, J. Aircr., 23, 87 (1986). Ericksen, R. H., Effect of Folding on the Strength of Kevlar 29 and Nylon Parachute Materials, Fibre Sci. Technol., 15, 161172 (1981). Ericksen, R. H., and Orear L., Wet Strength of Kevlar 29 Ribbon Parachute Fabrics, J. Aircr., 17, 267271 (1980). Pepper, W.B., Evaluation of Kevlar 29 vs Nylon for 3.81m Diameter Ribbon Parachutes, J. Aircr., 17, 198205 (1980). Horenstein, M. N., ,Surface Charging Limit For A Woven Fabric On A Ground Plane, J. Electrostat., 35(1), 3140 (1995). Li, P., Lee, C., and Corner, B., Computing Surface Area and Volume of Parachutes from 3D Scans, Engineering Computations, 22 (24), 393408 (2005).
18.
19.
20.
3.
21.
4.
5. 6. 7.
22. 23.
24. 25.
8.
Downloaded from http://trj.sagepub.com by Ngoc Nguyen Thi Thuy on October 28, 2009