Abstract
In the context of manufacturing, support vector machines (SVM) are commonly used to predict quality, i.e., predict the characteristics of a product according to the manufacturing parameters. The prediction accuracy of a SVM model is affected by a number of factors: training set size, data set quality, etc. Manufacturing datasets are usually prone to measurement uncertainties. Such uncertainties affect the observed values of the manufacturing parameters, thereby affecting the predictive performance of the SVM. To address this issue, several works in the literature have been proposed to improve the robustness of SVM to measurement uncertainties. These works, however, do not evaluate the contribution of the uncertainties of each parameter to the overall impact. For this reason, this paper focuses on quantifying the impact of the uncertainties of each parameter on the accuracy of the SVM prediction. Three approaches are proposed to do so. The first two approaches are based on Monte-Carlo simulation and allow providing quantitative measures that represent the impact of the uncertainties of each manufacturing parameter on the accuracy of the SVM. On the other hand, the third approach relies on simple statistical tools in order to estimate the impact of the uncertainties of each parameter. The proposed approaches would eventually make it possible to identify the uncertainties of the parameters that mostly affect the SVM. Such parameters are referred to as key measurement uncertainties. Identifying the key measurement uncertainties would provide a better understanding of how the SVM is affected by uncertainties, as it would provide a strong basis for improving the robustness of SVM in future works. The proposed approaches are applied to four datasets, and their performances are discussed and compared.
Similar content being viewed by others
References
Kwon, K., Kang, D., Yoon, Y., Sohn, J.-S., Chung, I.-J.: A real time process management system using RFID data mining. Comput. Ind. 65, 721–732 (2014). https://doi.org/10.1016/j.compind.2014.02.007
Hofmann, E., Rüsch, M.: Industry 4.0 and the current status as well as future prospects on logistics. Comput. Indus. 89, 23–34 (2017). https://doi.org/10.1016/j.compind.2017.04.002
Kusiak, A.: Data mining: manufacturing and service applications. Int. J. Prod. Res. 44, 4175–4191 (2006). https://doi.org/10.1080/00207540600632216
Padhy, N.: The survey of data mining applications and feature scope. IJCSEIT 2, 43–58 (2012). https://doi.org/10.5121/ijcseit.2012.2303
Silwattananusarn, T.: Data mining and its applications for knowledge management: a literature review from 2007 to 2012. IJDKP 2, 13–24 (2012). https://doi.org/10.5121/ijdkp.2012.2502
Rostami, H., Dantan, J.-Y., Homri, L.: Review of data mining applications for quality assessment in manufacturing industry: support vector machines. Int. J. Metrol. Qual. Eng. 6, 401 (2015). https://doi.org/10.1051/ijmqe/2015023
Wang, J., Ma, Y., Zhang, L., Gao, R.X., Wu, D.: Deep learning for smart manufacturing: Methods and applications. J. Manuf. Syst. 48, 144–156 (2018). https://doi.org/10.1016/j.jmsy.2018.01.003
Harding, J.A., Shahbaz, M., Srinivas, K.A.: Data mining in manufacturing: a review. J. Manufact. Sci. Eng. 128, 969–976 (2006). https://doi.org/10.1115/1.2194554
Köksal, G., Batmaz, İ, Testik, M.C.: A review of data mining applications for quality improvement in manufacturing industry. Exp. Syst. Appl. 38, 13448–13467 (2011). https://doi.org/10.1016/j.eswa.2011.04.063
Wuest, T., Weimer, D., Irgens, C., Thoben, K.-D.: Machine learning in manufacturing: advantages, challenges, and applications. Prod. Manufact. Res. 4, 23–45 (2016). https://doi.org/10.1080/21693277.2016.1192517
Cortes, C.: WSupport-vector network. Mach. Learn. 20, 1–25 (1995). https://doi.org/10.1007/BF00994018
Cholette, M.E., Borghesani, P., Di Gialleonardo, E., Braghin, F.: Using support vector machines for the computationally efficient identification of acceptable design parameters in computer-aided engineering applications (2017)
Zanaty, E.A.: Support vector machines (SVMs) versus multilayer perception (MLP) in data classification. Egyp. Inform. J. 13, 177–183 (2012). https://doi.org/10.1016/j.eij.2012.08.002
Diao, G., Zhao, L., Yao, Y.: A dynamic quality control approach by improving dominant factors based on improved principal component analysis. Int. J. Prod. Res. 53, 4287–4303 (2015). https://doi.org/10.1080/00207543.2014.997400
González, J.P.N.: Vehicle fault detection and diagnosis combining an AANN and multiclass SVM. Int. J. Inter. Des. Manufact. (IJIDeM) 12(1), 273–279 (2018). https://doi.org/10.1007/s12008-017-0378-z
Zhang, Z., Ming, W., Zhang, G., Huang, Y., Wen, X., Huang, H.: A new method for online monitoring discharge pulse in WEDM-MS process. Int. J. Adv. Manuf. Technol. 81, 1403–1418 (2015). https://doi.org/10.1007/s00170-015-7261-5
Ay, M., Stemmler, S., Schwenzer, M., Abel, D., Bergs, T.: Model predictive control in milling based on support vector machines. IFAC-PapersOnLine 52, 1797–1802 (2019). https://doi.org/10.1016/j.ifacol.2019.11.462
Hickey, R.J.: Noise modelling and evaluating learning from examples. Artif. Intell. 82, 157–179 (1996). https://doi.org/10.1016/0004-3702(94)00094-8
Leung, C.K.-S.: Mining uncertain data: mining uncertain data. WIREs Data Min. Knowl. Discov. 1, 316–329 (2011). https://doi.org/10.1002/widm.31
Zhu, X., Wu, X., Chen, Q.: Bridging local and global data cleansing: identifying class noise in large, distributed data datasets. Data Min. Knowl. Disc. 12, 275–308 (2006). https://doi.org/10.1007/s10618-005-0012-8
Yin, H., Dong, H.: The problem of noise in classification: past, current and future work. In: 2011 IEEE 3rd international conference on communication software and networks. In: Presented at the 2011 IEEE 3rd international conference on communication software and networks (ICCSN), pp. 412–416. IEEE, Xi’an, China (2011). https://doi.org/10.1109/ICCSN.2011.6014597
Wang, X., Pardalos, P.M.: A survey of support vector machines with uncertainties. Ann. Data Sci. 1(3–4), 293–309 (2014). https://doi.org/10.1007/s40745-014-0022-8
Bi, J., Zhang, T.: Support vector classification with input data uncertainty. Adv. Neural. Inf. Process. Syst. 17, 8 (2005)
Niaf, E., Flamary, R., Lartizien, C., Canu, S.: Handling uncertainties in SVM classification, in: 2011 IEEE Statistical Signal Processing Workshop (SSP). Presented at the 2011 IEEE Statistical Signal Processing Workshop (SSP), IEEE, Nice, France, pp. 757–760 (2011). https://doi.org/10.1109/SSP.2011.5967814
Jeyakumar, V., Li, G., Suthaharan, S.: Support vector machine classifiers with uncertain knowledge sets via robust optimization. Optimization 63, 1099–1116 (2014). https://doi.org/10.1080/02331934.2012.703667
Pant, R., Trafalis, T. B., Barker, K.: Support vector machine classification of uncertain and imbalanced data using robust optimization. In: Proceedings of the 15th WSEAS international conference on computers, pp. 369–374. World Scientific and Engineering Academy and Society (WSEAS) Stevens Point, Wisconsin, USA (2011)
Mohammadi, M., Siadat, A., Dantan, J.-Y., Tavakkoli-Moghaddam, R.: Mathematical modelling of a robust inspection process plan: Taguchi and Monte Carlo methods. Int. J. Prod. Res. 53, 2202–2224 (2015). https://doi.org/10.1080/00207543.2014.980460
Mantovani, R.G., Rossi, A.L.D., Alcobaça, E., Vanschoren, J., de Carvalho, A.C.P.L.F.: A meta-learning recommender system for hyperparameter tuning: predicting when tuning improves SVM classifiers. Inform. Sci. 501, 193–221 (2019). https://doi.org/10.1016/j.ins.2019.06.005
Bäck, T., Schwefel, H.-P.: An overview of evolutionary algorithms for parameter optimization. Evol. Comput. 1, 1–23 (1993). https://doi.org/10.1162/evco.1993.1.1.1
Chou, J.-S., Cheng, M.-Y., Wu, Y.-W., Pham, A.-D.: Optimizing parameters of support vector machine using fast messy genetic algorithm for dispute classification. Exp. Syst. Appl. 41, 3955–3964 (2014). https://doi.org/10.1016/j.eswa.2013.12.035
Lin, S.-W., Ying, K.-C., Chen, S.-C., Lee, Z.-J.: Particle swarm optimization for parameter determination and feature selection of support vector machines. Exp. Syst. Appl. 35, 1817–1824 (2008). https://doi.org/10.1016/j.eswa.2007.08.088
Yang, D., Liu, Y., Li, S., Li, X., Ma, L.: Gear fault diagnosis based on support vector machine optimized by artificial bee colony algorithm. Mech. Mach. Theory 90, 219–229 (2015). https://doi.org/10.1016/j.mechmachtheory.2015.03.013
Olatomiwa, L., Mekhilef, S., Shamshirband, S., Mohammadi, K., Petković, D., Sudheer, C.: A support vector machine–firefly algorithm-based model for global solar radiation prediction. Sol. Energy 115, 632–644 (2015). https://doi.org/10.1016/j.solener.2015.03.015
Özdemir, Ö., Çavuş, M.: Performance of the inverse transformation method for extreme value distributions. In: Xth international statistics days conference (ISDC’2016), Giresun, Turkey 8 (2016)
Idriss, D., Beaurepaire, P., Homri, L., Gayton, N.: Key Characteristics identification by global sensitivity analysis. Int. J. Inter. Des. Manuf. 14(2), 423–434 (2020)
Jacques, J., Lavergne, C., Devictor, N.: Sensitivity analysis in presence of model uncertainty and correlated inputs. Reliab. Eng. Syst. Saf. 91, 1126–1134 (2006). https://doi.org/10.1016/j.ress.2005.11.04723
Abdiansah, A., Wardoyo, R.: Time complexity analysis of support vector machines (SVM) in LibSVM. IJCA 128, 28–34 (2015). https://doi.org/10.5120/ijca2015906480
Zouhri, W., Homri, L., Dantan, J.Y.: Handling the impact of feature uncertainties on SVM: a robust approach based on Sobol sensitivity analysis. Exp. Syst. Appl. 115691 (2021)
Zou, X., Zhao, X., Li, G., Li, Z., Sun, T.: Sensitivity analysis using a variance-based method for a three-axis diamond turning machine. Int J Adv Manuf Technol 92, 4429–4443 (2017). https://doi.org/10.1007/s00170-017-0394-y
Quality level of a chemical product, Chem_4/8. Available from: https://cloud.ensam.eu/index.php/s/FE5yhJp4Xequ5Zt
Quality level of a chemical product, Chem_5/7. Available from: https://cloud.ensam.eu/index.php/s/FE5yhJp4Xequ5Zt
Quality level of floatation process, Mine_1/2. Available from: https://cloud.ensam.eu/index.php/s/FE5yhJp4Xequ5Zt
Quality assurance in manufacturing system -Confidential manufacturing process-, Roll_0/1. Available from: https://cloud.ensam.eu/index.php/s/FE5yhJp4Xequ5Zt
Acknowledgements
The authors gratefully acknowledge the support of the research chair on reconfigurable, safe, and efficient production systems, co-founded by Thyssenkrupp, the European Regional Development Fund “Programme opérationnel FEDER-FSE Lorraine et Massif de Vosges 2014-2020”, UIMM F2I and UIMM Lorraine.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Support vector machine
SVM was first introduced for the classification of linearly separable binary problems. Given a set of N observations \(D = \left\{ {\left( {x_{1} ,y_{1} ), \ldots ,(x_{N} ,y_{N} } \right)} \right\}\), where \(x_{i}\) represents the ith training data and \(y_{i}\) is the class label of \(x_{i}\), the SVM tries to find a hyperplane that separates the set space into two areas, each of which includes individuals from one of the two classes. The separating hyperplane is given as \(w^{T} x + b = 0\), where \(b\) is the bias of the hyperplane, \(x\) are the data located within the hyperplane, and \(w{ }\) are the weights that determine the hyperplane’s orientation. SVM seeks to maximize the margin by establishing formulation (Eq. 4) that allows defining the parameters of the optimal hyperplane [39]
Additionally, the linear SVM formulation might be useful to deal with the classification of some non-linearly separable problems. This can be done by allowing the constraint (of formulation 3) to be violated by some datapoints. To this end, the formulation of the SVM is modified so that it can accept that the margin is crossed, and that certain datapoints are on the wrong side of the hyperplane. A new variable “\({\upxi }_{i}\)” associated to each datapoint is therefore introduced to inform how far this point violates the constraint. This new variable, called the slack variable, can be either:
-
ξ = 0: if the data point complies with the constraint.
-
ξ > 1: if the data point is misclassified.
-
0 < ξ < 1: if the data point has crossed the margin.
Consequently, the new formulation (Eq. 5) aims to minimize the sum of errors \(\left( {\sum\nolimits_{i = 1}^{n} {{\upxi }_{i} } } \right)\) in addition to \(w^{2}\), while it introduces a new regularization parameter “C” that manages the compromise between the margin width and the relaxation of the constraint.
However, to address most of the non-linearly separable problems, the kernel trick method is adopted to create non-linear SVM models. The kernel trick consists on mapping the datapoints to a larger dimensional space called the feature space \(\phi :R^{d} \to R^{p}\) which is equivalent to replacing each datapoint “\(x_{i}\)” by its image “\(z_{i} = \phi (x_{i} )\)” in the new space. The motivation of applying the kernel trick comes from the fact that data which is not separable in the input space can always be separated in a space of high enough dimensionality. This data transformation refers us to the new formulation (Eq. 6).
To compute the SVM formulations, the dual formulation is used. The formulations seen above are constrained quadratic optimization problems that are written as follows (Eq. 7), and can be reformulated as the equation (Eq. 8) using Lagrange multipliers “αi”:
Similarly, the formulation of the SVM can be written:
At the minimum, we can take the derivatives with respect to the primal variables “w” and “b” and set these to zero:
By replacing the two previous formulas in the formulation (8):
This formulation has a major advantage when using the kernel trick because the functional form of the mapping \(\phi ({\text{x}}_{i} )\) does not need to be known since it is implicitly defined in the feature space by the choice of the kernel function \(K\left( {x_{i} \times x_{j} } \right) = \phi (x_{i} ) \times \phi (x_{j} ).\)
Appendix B: Sobol sensitivity analysis
Sobol’s analysis is one the different methods to analyse the global sensitivity of a model. It is based on variance decomposition techniques to provide a quantitative measure of the contributions of the input to the output variance. The decomposition of the output variance in a Sobol sensitivity analysis employs the same principal as the classical analysis of variance in a factorial design, which allows representing the output’s variance as follow:
where: V: Total variance of the model output. \({\text{V}}_{{\text{i}}}\): The first order contribution of the ith model parameter. \({\text{V}}_{{{\text{ij}}}}\): The contribution of the interaction of the ith and jth parameters.
Based on this decomposition, Sobol defines first order sensitivity indices, Eq. 14, as well as higher-order sensitivity indices Eq. 15, such as:
Also, to measure the total sensitivity of the variance Y due to a variable Xi, Homma and Saltelli [40] introduced Total-effect indices Eq. 16 defined as the sum of the contribution caused by Xi and by its interactions -of any order- with any other input variables.
where #i represents all sets containing the index i.
One of the ways to estimate Sobol indices is by using the Monte-Carlo simulation. The Monte-Carlo estimation consists on estimating: the output expected value (E[Y]), the output variance (V[Y]), and both quantities (\(\^U_{i}\)) and (\(\^U_{\sim i}\)). The estimation of these four quantities requires two -samples \(\tilde{X}_{\left( N \right)}^{\left( 1 \right)} = (x_{k1}^{\left( 1 \right)} , \ldots ,x_{kp}^{\left( 1 \right)} )_{k = 1..N}\) and \({ }\tilde{X}_{\left( N \right)}^{\left( 2 \right)} = (x_{k1}^{\left( 2 \right)} , \ldots ,x_{kp}^{\left( 2 \right)} )_{k = 1..N}\), using the following formulas:
where “f” is the model that links the inputs to the output.
Based on the previous quantities, the Sobol indices can be estimated in the following manner:
-
First order sensitivity indices: \(\hat{S}_{i} = \frac{{\^U_{i} - \^E^{2} }}{{\hat{V}}}\)
-
Total-effect indices: \(\hat{S}_{Ti} = 1 - \frac{{\^U_{\sim i} - \^E^{2} }}{{\hat{V}}}\)
Rights and permissions
About this article
Cite this article
Zouhri, W., Homri, L. & Dantan, JY. Identification of the key manufacturing parameters impacting the prediction accuracy of support vector machine (SVM) model for quality assessment. Int J Interact Des Manuf 16, 177–196 (2022). https://doi.org/10.1007/s12008-021-00807-8
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s12008-021-00807-8