Research Papers

Developing Metamodels for Fast and Accurate Prediction of the Draping of Physical Surfaces

[+] Author and Article Information
Esben Toke Christensen

Department of Mechanical and
Manufacturing Engineering,
Aalborg University,
Fibigerstræde 16,
Aalborg East 9220, Denmark
e-mail: esben@m-tech.aau.dk

Alexander I. J. Forrester

Centre of Excellence,
University of Southampton Engineering,
Burgess Road,
Southampton SO16 7QF, UK
e-mail: alexander.forrester@soton.ac.uk

Erik Lund

Department of Mechanical and
Manufacturing Engineering,
Aalborg University
Fibigerstræde 16,
Aalborg East 9220, Denmark
e-mail: el@m-tech.aau.dk

Esben Lindgaard

Department of Mechanical and
Manufacturing Engineering,
Aalborg University,
Fibigerstræde 16,
Aalborg East 9220, Denmark
e-mail: elo@m-tech.aau.dk

Manuscript received April 18, 2017; final manuscript received October 28, 2017; published online March 15, 2018. Assoc. Editor: Ying Liu.

J. Comput. Inf. Sci. Eng 18(2), 021003 (Mar 15, 2018) (12 pages) Paper No: JCISE-17-1080; doi: 10.1115/1.4039334 History: Received April 18, 2017; Revised October 28, 2017

In this paper, the use of methods from the meta- or surrogate modeling literature, for building models predicting the draping of physical surfaces, is examined. An example application concerning modeling of the behavior of a variable shape mold is treated. Four different methods are considered for this problem. The proposed methods are difference methods assembled from the methods kriging and proper orthogonal decomposition (POD) together with a spline-based underlying model (UM) and a novel patchwise modeling scheme. The four models, namely kriging and POD with kriging of the coefficients in global and local variants, are compared in terms of accuracy and numerical efficiency on data sets of different sizes for the treated application. It is shown that the POD-based methods are vastly superior to models based on kriging alone, and that the use of a difference model structure is advantageous. It is demonstrated that patchwise modeling schemes, where the complete surface behavior is modeled by a collection of locally defined smaller models, can provide a good compromise between achieving good model accuracy and scalability of the models to large systems.

Copyright © 2018 by ASME
Your Session has timed out. Please sign back in to continue.


Eckers, V. , Janetzko, S. , and Gries, T. , 2012, “ Drape Study on Textiles for Concrete Applications,” Autex Res. J., 12(2), pp. 50–54. http://www.autexrj.com/cms/zalaczone_pliki/0010_12.pdf
Munro, C. , and Walczyk, D. F. , 2007, “ Reconfigurable Pin-Type Tooling: A Survey of Prior Art and Reduction to Practice,” ASME J. Manuf. Sci. Eng., 129(3), pp. 551–565. [CrossRef]
Montgomery, D. C. , 2012, Design and Analysis of Experiments, Wiley, Hoboken, NJ.
MacKay, D. J. C. , 2004, Information Theory, Inference, and Learning Algorithms, Cambridge University Press, Cambridge, UK.
Forrester, A. I. J. , and Keane, A. J. , 2009, “ Recent Advances in Surrogate-Based Optimization,” Prog. Aerosp. Sci., 45(1–3), pp. 50–79. [CrossRef]
Smola, A. , and Schölkopf, B. , 2004, “ A Tutorial on Support Vector Regression,” Stat. Comput., 14(3), pp. 199–222. [CrossRef]
Forrester, A. I. J. , Sóbester, A. , and Keane, A. J. , 2008, Engineering Design Via Surrogate Modelling: A Practical Guide, Wiley, Hoboken, NJ.
Kalogirou, S. A. , 2001, “ Artificial Neural Networks in Renewable Energy Systems Applications: A Review,” Renewable Sustainable Energy Rev., 5(4), pp. 373–401. [CrossRef]
Carr, J. C. , Fright, W. R. , and Beatson, R. K. , 1997, “ Surface Interpolation With Radial Basis Functions for Medical Imaging,” IEEE Trans. Med. Imaging, 16(1), pp. 96–107. [CrossRef] [PubMed]
Klass, V. , Behm, M. , and Lindbergh, G. , 2015, “ Capturing Lithium-Ion Battery Dynamics With Support Vector Machine-Based Battery Model,” J. Power Sources, 298, pp. 92–101. [CrossRef]
Shi, X. , Palos Teixeira, Â. , Zhang, J. , and Guedes Soares, C. , 2014, “ Kriging Response Surface Reliability Analysis of a Ship-Stiffened Plate With Initial Imperfections,” Struct. Infrastruct. Eng., 11(11), pp. 1450–1465. [CrossRef]
Hamdaoui, M. , Oujebbour, F.-Z. , Habbal, A. , Breitkopf, P. , and Villon, P. , 2014, “ Kriging Surrogates for Evolutionary Multi-Objective Optimization of CPU Intensive Sheet Metal Forming Applications,” Int. J. Mater. Form., 8(3), pp. 469–480. [CrossRef]
Rayas-Sanchez, J. E. , 2004, “ EM-Based Optimization of Microwave Circuits Using Artificial Neural Networks: The State-of-the-Art. Microw,” Theory Tech. IEEE Trans., 52(1), pp. 420–435. [CrossRef]
Shlens, J. , 2014, “A Tutorial on Principal Component Analysis,” e-print arXiv: 1404.1100. https://arxiv.org/abs/1404.1100
Xiao, M. , Breitkopf, P. , Filomeno Coelho, R. , Knopf-Lenoir, C. , Sidorkiewicz, M. , and Villon, P. , 2010, “ Model Reduction by CPOD and Kriging,” Struct. Multidiscip. Optim., 41(4), pp. 555–574. [CrossRef]
Bandler, J. W. , Biernacki, R. , Chen, S. H. , Grobelny, P. A. , and Hemmers, R. H. , 1994, “ Space Mapping Technique for Electromagnetic Optimization,” IEEE Trans. Microwave Theory Tech., 42(12), pp. 2536–2544. [CrossRef]
Friedman, J. H. , 1991, “ Multivariate Adaptive Regression Splines,” Ann. Stat., 19(1), pp. 1–67. [CrossRef]
Mullur, A. A. , and Messac, A. , 2005, “ Extended Radial Basis Functions: More Flexible and Effective Metamodeling,” AIAA J., 43(6), pp. 1306–1315. [CrossRef]
Rasmussen, C. E. , 2006, Gaussian Processes for Machine Learning, MIT Press, Cambridge, MA.
Krige, D. G. , 1951, “ A Statistical Approach to Some Basic Mine Valuation Problems on the Witwatersrand,” J. Chem. Metall. Min. Soc. South Africa, 52(6), pp. 119–139. http://journals.co.za/content/saimm/52/9/AJA0038223X_4858
Sacks, J. , Welch, W. J. , Mitchell, T. J. , and Wynn, H. P. , 1989, “ Design and Analysis of Computer Experiments,” Stat. Sci., 4(4), pp. 409–423. [CrossRef]
Forrester, A. I. J. , Sóbester, A. , and Keane, A. J. , 2015, “Optimization Codes,” Optimization Codes, accessed Feb. 28, 2018, https://optimizationcodes.wordpress.com
Joseph, V. R. , and Ying, H. , 2008, “ Orthogonal-Maximin Latin Hypercube Designs,” Stat. Sin., 18(1), pp. 171–186. http://www3.stat.sinica.edu.tw/statistica/oldpdf/A18n17.pdf
Fuentes, M. , 2001, “ A High Frequency Kriging Approach for Non-Stationary Environmental Processes,” Environmetrics, 12(5), pp. 469–483. [CrossRef]
Xiong, Y. , Chen, W. , Apley, D. , and Ding, X. , 2007, “ A Non-Stationary Covariance-Based Kriging Method for Metamodelling in Engineering Design,” Int. J. Numer. Methods Eng., 71(6), pp. 733–756. [CrossRef]


Grahic Jump Location
Fig. 1

Illustration of the variable shape mold concept

Grahic Jump Location
Fig. 2

Layout of the substitute model. Each dot represents an actuator. Dimensions as well as the selected actuator numbering are shown.

Grahic Jump Location
Fig. 3

Schematic illustration of the actuator to membrane implementation in the numerical substitute model

Grahic Jump Location
Fig. 4

Illustration of the interpolation done during post-processing. Due to in-plane displacements of the FE nodes, a piecewise linear interpolation is used to evaluate the displacements in a fixed grid to achieve a consistent data representation.

Grahic Jump Location
Fig. 5

The principle in the difference method in two cases: (a) at training and (b) at prediction. The punctuated box encloses what is collectively referred to as the surrogate model.

Grahic Jump Location
Fig. 6

Illustration of the construction of the spline-based coarse model. Splines are fitted and sampled along the in-plane directions in the succession shown in the figure legend. The total number of sampled points in the plane is N2 (see Table 1).

Grahic Jump Location
Fig. 7

Illustration of the arrangement of patch models on the mold surface. Translucent squares mark the prediction regions of three models. The red square in full line marks the input region of the red patch model. Squares in broken black line exemplify quarter, half, and full size patch models.

Grahic Jump Location
Fig. 8

Comparison of the UM and the four surrogate models. Mean, standard deviation, and min/max are shown for the RMAE (Eq. (35)) and RRMSE (Eq. (36)) measures. Numbers attached to each point on the mean graphs indicate the number of mold configurations in the data set for that particular model.

Grahic Jump Location
Fig. 9

Plots of the hyper values for krigs trained on 9 × 9 spatial domains. The hyper parameters shown correspond to (a) actuator 1 with center (0,0), (b) actuator 41 with center (0.5 L,0.5L) (c) actuator 72 with center (L,0), and (d) the x coordinate.

Grahic Jump Location
Fig. 10

Comparison of the UM and the four approximation methods without the use of an UM. Mean, standard deviation, and min/max are shown for the RMAE (Eq. (35)) and RRMSE (Eq. (36)) measures. Numbers attached to each point on the mean graphs indicate the number of mold configurations in the data set for that particular model. It is seen that without including the knowledge of an approximate UM, the models struggle to even reach the base performance of the UM.




Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In