0
Research Papers

A New Design Approach for Rapid Evaluation of Structural Modifications Using Neural Networks

[+] Author and Article Information
O. Demirkan

Mechanical Engineering Department,
Middle East Technical University,
Ankara 06531, Turkey
e-mail: ozlemdemirkan@gmail.com

E. Olceroglu

e-mail: eolceroglu@gmail.com

I. Basdogan

e-mail: ibasdogan@ku.edu.tr
Mechanical Engineering Department,
Koc University,
Istanbul 34450, Turkey

H. N. Özgüven

Mechanical Engineering Department,
Middle East Technical University,
Ankara 06531, Turkey
e-mail: ozguven@metu.edu.tr

1Corresponding author.

Contributed by the Design Automation Committee of ASME for publication in the JOURNAL OF MECHANICAL DESIGN. Manuscript received January 17, 2012; final manuscript received October 1, 2012; published online January 7, 2013. Assoc. Editor: Shinji Nishiwaki.

J. Mech. Des 135(2), 021004 (Jan 07, 2013) (10 pages) Paper No: MD-12-1056; doi: 10.1115/1.4023156 History: Received January 17, 2012; Revised October 01, 2012

Design optimization of structural systems is often iterative, time consuming and is limited by the knowledge of the designer. For that reason, a rapid design optimization scheme is desirable to avoid such problems. This paper presents and integrates two design methodologies for efficient conceptual design of structural systems involving computationally intensive analysis. The first design methodology used in this paper is structural modification technique (SMT). The SMT utilizes the frequency response functions (FRFs) of the original model for the reanalysis of the structure that is subjected to structural modification. The receptances of the original structure are coupled with the dynamic stiffness of the components that are added to or removed from the original structure to perform the structural modification. Then, the coupled matrices are used to calculate the mobility matrices of the modified structure in an efficient way. The second design methodology used in this paper is neural networks (NN). Once sufficient input and output relationships are obtained through SMT, a NN model is constructed to predict the FRF curves of the system for further analysis of the system performance while experimenting different design parameters. The input–output sets used for training the network are increased by applying an interpolation scheme to improve the accuracy of the NN model. The performance of the proposed method integrated through SMT and NN technique is demonstrated on a rectangular plate to observe the effect of the location and thickness of stiffeners on the frequency response of the structure. It is observed that both methods combined with the proposed interpolation scheme work very efficiently to predict the dynamic response of the structure when modifications are required to improve the system performance.

Copyright © 2013 by ASME
Your Session has timed out. Please sign back in to continue.

References

Figures

Grahic Jump Location
Fig. 3

Point FRF of point P for two different mesh sizes

Grahic Jump Location
Fig. 2

FRF results of the mid node at the flexible plate

Grahic Jump Location
Fig. 1

Flexible plate and the modifying strip

Grahic Jump Location
Fig. 4

Rectangular plate used in the structural modification studies

Grahic Jump Location
Fig. 5

Cross FRF of selected node after adding the stiffener at x = 0 (FRFs are in mm/N)

Grahic Jump Location
Fig. 6

Cross FRF of the selected node for different positions of the added strip (FRFs are in mm/N)

Grahic Jump Location
Fig. 7

Cross FRF of the selected node for different width of the strip added in the middle (FRFs are in mm/N)

Grahic Jump Location
Fig. 13

FRF amplitudes at 33 Hz obtained by training NN with 150 data sets (120 of them are artificial data) (FRFs are in mm/N)

Grahic Jump Location
Fig. 14

FRF amplitudes at 10 Hz obtained by training NN with 150 data sets (120 of them are artificial data) (FRFs are in mm/N)

Grahic Jump Location
Fig. 8

The NN model used in this study

Grahic Jump Location
Fig. 9

FRF amplitudes at 33 Hz obtained by training NN with 30 data sets (FRFs are in mm/N)

Grahic Jump Location
Fig. 10

FRF amplitudes at 10 Hz obtained by training NN with 30 data sets (FRFs are in mm/N)

Grahic Jump Location
Fig. 11

FRF amplitudes at 33 Hz obtained by training NN with 60 data sets (30 of them are artificial data) (FRFs are in mm/N)

Grahic Jump Location
Fig. 12

FRF amplitudes at 10 Hz obtained by training NN with 60 data sets (30 of them are artificial data) (FRFs are in mm/N)

Grahic Jump Location
Fig. 15

Cross FRF of selected node using neural network trained with 30 data sets (strip at x = −280 mm) (FRFs are in mm/N)

Grahic Jump Location
Fig. 16

Cross FRF of the selected node using neural network trained with 60 data sets (strip at x = −280 mm) (30 of them are artificial data) (FRFs are in mm/N)

Grahic Jump Location
Fig. 17

Cross FRF of selected node using neural network trained with 150 data sets (strip at x = −280 mm) (120 of them are artificial data) (FRFs are in mm/N)

Grahic Jump Location
Fig. 18

Cross FRF of the selected node using neural network node trained with 10 data sets (30 mm wide strip) (FRFs are in mm/N)

Grahic Jump Location
Fig. 19

Cross FRF of the selected node using neural network node trained with 19 data sets (30 mm wide strip) (9 of them are artificial data) (FRFs are in mm/N)

Grahic Jump Location
Fig. 20

Cross FRF of the selected node using neural network trained with 37 data sets (30 mm wide strip) (27 of them are artificial data) (FRFs are in mm/N)

Tables

Errata

Discussions

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