0
Journal of Vibration and Acoustics | Volume 135 | Issue 5 | research-article
Research Papers

Characterization of Dynamic Response of Structures With Uncertainty by Using Gaussian Processes

[+] Author and Article Information
J. Tang

e-mail: jtang@engr.uconn.edu
Department of Mechanical Engineering,
University of Connecticut,
Storrs, CT 06269

1Corresponding author.

Contributed by the Design Engineering Division of ASME for publication in the Journal of Vibration and Acoustics. Manuscript received August 3, 2011; final manuscript received March 8, 2013; published online June 18, 2013. Assoc. Editor: Michael Brennan.

J. Vib. Acoust 135(5), 051006 (Jun 18, 2013) (13 pages) Paper No: VIB-11-1168; doi: 10.1115/1.4023998 History: Received August 03, 2011; Revised March 08, 2013
Article
References
Figures
Tables
Citing Articles

Characterizing dynamic characteristics of structures with uncertainty is an important task that provides critical predictive information for structural design, assessment, and control. In practical applications, sampling is the fundamental approach to uncertainty analysis but has to be conducted under various constraints. To address the frequently encountered data scarcity issue, in the present paper Gaussian processes are employed to predict and quantify structural dynamic responses, especially responses under uncertainty. A self-contained description of Gaussian processes is presented within the Bayesian framework with implementation details, and then a series of case studies are carried out using a cyclically symmetric structure that is highly sensitive to uncertainties. Structural frequency responses are predicted with data sparsely sampled within the full frequency range. Based on the inferred credible intervals, a measure is defined to quantify the potential risk of response maxima. Gaussian process emulation is proposed for Monte Carlo uncertainty analysis to reduce data acquisition costs. It is shown that Gaussian processes can be an efficient data-based tool for analyzing structural dynamic responses in the presence of uncertainty. Meanwhile, some technical challenges in the implementation of Gaussian processes are discussed.
FIGURES IN THIS ARTICLE
<>
Copyright © 2013 by ASME
Your Session has timed out. Please sign back in to continue.

References

1
Ghanem, R. G., and Spanos, P. D., 1991, Stochastic Finite Elements: A Spectral Approach, Springer-Verlag, New York.
2
Panayirci, H. M., Pradlwarter, H. J., and Schueller, G. I., 2011, “Efficient Stochastic Finite Element Analysis Using Guyan Reduction,” Adv. Eng. Software, 42(4), pp. 187–196. [CrossRef]
3
Friswell, M. I., and Adhikari, S., 2000, “Derivatives of Complex Eigenvectors Using Nelson's Method,” AIAA J., 38(12), pp. 2355–2357. [CrossRef]
4
Soize, C., 2000, “A Nonparametric Model of Random Uncertainties for Reduced Matrix Models in Structural Dynamics,” Probab. Eng. Mech., 15, pp. 277–294. [CrossRef]
5
Soize, C., Capiez-Lernount, E., Durand, J.-F., Fernandez., C., and Gagliardini., L., 2008, “Probabilistic Model Identification of Uncertainties in Computational Models for Dynamical Systems and Experimental Validation,” Comput. Methods Appl. Mech. Eng., 198, pp. 150–163. [CrossRef]
6
Adhikari, S., and Sarkar, A., 2009, “Uncertainty in Structural Dynamics: Experimental Validation of a Wishart Random Matrix Model,” J. Sound Vib., 323, pp. 802–825. [CrossRef]
7
Hinke, L., Dohnal, F., Mace, B. R., Waters, T. P., and Ferguson, N. S., 2009, “Component Mode Synthesis as a Framework for Uncertainty Analysis,” J. Sound Vib., 324, pp. 161–178. [CrossRef]
8
Bladh, R., Pierre, C., and Castanier, M. P., 2002, “Dynamic Response Predictions for a Mistuned Industrial Turbomachinery Rotor Using Reduced-Order Modeling,” ASME J. Eng. Gas Turbines Power, 124, pp. 311–324. [CrossRef]
9
Lee, S.-Y., and Castanier, M. P., 2006, “Component-Mode-Based Monte Carlo Simulation for Efficient Probabilistic Vibration Analysis,” Proceedings of the 47th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, Newport, RI, May 1–4, AIAA Paper No. 2006-1990. [CrossRef]
10
Dohnal, F., Mace, B. R., and Ferguson, N. S., 2009, “Joint Uncertainty Propagation in Linear Structural Dynamics Using Stochastic Reduced Basis Methods,” AIAA J., 47, pp. 961–969. [CrossRef]
11
Tournour, M. A., Atalla, N., Chiello, O., and Sgard, F., 2001, “Validation, Performance, Convergence and Application of Free Interface Component Mode Synthesis,” Comput. Struct., 79, pp. 1861–1876. [CrossRef]
12
Craig, R. R., Jr., 2000, “Coupling of Substructure for Dynamic Analyses: An Overview,” Proceedings of the 41st AIAA/ASME/ASCE/AHSIASC Structures, Structural Dynamics, and Materials Conference and Exhibit, Atlanta, GA.
13
Feiner, D. M., and Griffin, J. H., 2002, “A Fundamental Model of Mistuning for a Single Family of Modes,” ASME J. Turbomach., 124, pp. 597–605. [CrossRef]
14
Moyroud, F., Fransson, T., and Jacquet-Richardet, G., 2002, “A Comparison of Two Finite Element Reduction Techniques for Mistuned Bladed Disks,” ASME J. Eng. Gas Turbines Power, 124, pp. 942–952. [CrossRef]
15
Mbaye, M., Soize, C., and Ousty, J.-P., 2010, “A Reduced-Order Model of Detuned Cyclic Dynamical Systems With Geometric Modifications Using a Basis of Cyclic Modes,” ASME J. Eng. Gas Turbines Power, 132, p. 112502. [CrossRef]
16
Bladh, R., Castanier, M. P., and Pierre, C., 2001, “Component-Mode-Based Reduced Order Modeling Techniques for Mistuned Bladed Disks—Part I: Theoretical Models,” ASME J. Eng. Gas Turbines Power, 123, pp. 89–99. [CrossRef]
17
Bah, M. T., Nair, P. B., Bhaskar, A., and Keane, A. J., 2003, “Stochastic Component Mode Synthesis,” Proceedings of the 44th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, Norfolk, VA, April 7–10, AIAA Paper No. 2003-1750. [CrossRef]
18
Xia, Z., and Tang, J., 2011, “Uncertainty Analysis of Structural Dynamics by Using Two-Level Gaussian Processes,” Proceedings of the ASME International Mechanical Engineering Congress and Exposition, Denver, CO, November 11–17, ASME Paper No. IMECE2011-63708. [CrossRef]
19
Neal, R. M., 1998, “Regression and Classification Using Gaussian Process Priors,” Bayesian Statistics 6, J. M.Bernardo, ed., Oxford University, New York.
20
MacKay, D. J. C., 2003, Information Theory, Inference and Learning Algorithms, Cambridge University Press, Cambridge, UK.
21
Rasmussen, C. E., and Williams, C. K. I., 2006, Gaussian Processes for Machine Learning, MIT, Cambridge, MA.
22
DiazDelaO, F. A., and Adhikari, S., 2010, “Structural Dynamic Analysis Using Gaussian Process Emulators,” Eng. Comput., 27, pp. 580–605. [CrossRef]
23
DiazDelaO, F. A., and Adhikari, S., 2011, “Gaussian Process Emulators for the Stochastic Finite Element Method,” Int. J. Numer. Methods Eng., 87, pp. 521–540. [CrossRef]
24
Hasen, J., Murray-Smith, R., Johansen, T. A., 2005, “Nonparametric Identification of Linearizations and Uncertainty Using Gaussian Process Models—Application to Robust Wheel Slip Control,” Proceedings of the Joint 44th IEEE Conference on Decision and Control, and European Control Conference(CDC-ECC’05), Seville, Spain, December 12–15. [CrossRef]
25
Gregorcic, G., and Lightbody, G., 2009, “Gaussian Process Approach for Modeling of Nonlinear Systems,” Eng. Applic. Artif. Intell., 22, pp. 522–533. [CrossRef]
26
Azman, K., and Kocijan, J., 2009, “Fixed-Structure Gaussian Process Model,” Int. J. Syst. Sci., 40, pp. 1253–1262. [CrossRef]
27
Mohanty, S., Das, S., Chattopadhyay, A., and Peralta, P., 2009, “Gaussian Process Time Series Model for Life Prognosis of Metallic Structures,” J. Intell. Mater. Syst. Struct., 20, pp. 887–896. [CrossRef]
28
Kennedy, M. C., and O'Hagan, A., 2000, “Predicting the Output From a Complex Computer Code When Fast Approximations are Available,” Biometrika, 87, pp. 1–13. [CrossRef]
29
Sivia, D. S., and Skilling, J., 2006, Data Analysis—A Bayesian Tutorial, 2nd ed., Oxford University, New York.
30
Loredo, T. J., 1990, From Laplace to Supernova SN 1987A: Bayesian Inference in Astrophysics, in Maximum Entropy and Bayesian Methods, Academic Publishers, Dordrecht, The Netherlands.
31
Mockus, J., Eddy, W., and Reklaitis, G., 1997, Bayesian Heuristic Approach to Discrete and Global Optimization, Kluwer Academic Publishers, Dordrecht, The Netherlands.
32
Alvarez, M. A., and Lawrence, N. D., 2011, “Computationally Efficient Convolved Multiple Output Gaussian Processes,” J. Mach. Learn. Res., 12, pp. 1459–1500.
33
Kennedy, M. C., and O'Hagan, A., 2000, Supplementary Details on Bayesian Calibration of Computer Codes, University of Sheffield, Sheffield, UK.
34
Swiler, L. P., 2006, “Bayesian Methods in Engineering Design Problems,” Sandia National Laboratories, Paper No. SAND2005–3249.
35
Kennedy, M. C., and O'Hagan, A., 2001, “Bayesian Calibration of Computer Models,” J. R. Stat. Soc., Ser. B, 63, p. 425–464. [CrossRef]
36
Whitehead, D. S., 1998, “Maximum Factor by Which Forced Vibration of Blades Can Increase Due to Mistuning,” ASME J. Eng. Gas Turbines Power, 120, pp. 115–119. [CrossRef]
37
Kenyon, J. A., and Griffin, J. H., 2003, “Forced Response of Turbine Engine Bladed Disks and Sensitivity to Harmonic Mistuning,” ASME J. Eng. Gas Turbines Power, 125, pp. 113–120. [CrossRef]
38
Brooks, S. P., 1998, “Markov Chain Monte Carlo Method and Its Application,” J. R. Stat. Soc., Ser. D, 47(1), pp. 69–100. [CrossRef]
39
Bastos, L. S., and O'Hagan, A., 2009, “Diagnostics for Gaussian Process Emulators,” Technometrics, 51, pp. 425–438. [CrossRef]

Figures

Grahic Jump Location
Fig. 1

Geometry and meshing of the example bladed disk

+Geometry and meshing of the example bladed disk
View Large  |  Save Figure  |  Download Slide (.ppt)
Geometry and meshing of the example bladed disk
Grahic Jump Location
Fig. 2

The seconds engine order frequency responses on 18 blade tips. Solid lines—tuned tip responses (identical for all blades); dotted lines—mistuned tip responses. (a) 2% mistuning SD; (b) 5% mistuning SD; (c) 8% mistuning SD; and (d) 12% mistuning SD.

+The seconds engine order frequency responses on 18 blade tips. Solid lines—tuned tip responses (identical for all blades); dotted lines—mistuned tip responses. (a) 2% mistuning SD; (b) 5% mistuning SD; (c) 8% mistuning SD; and (d) 12% mistuning SD.
View Large  |  Save Figure  |  Download Slide (.ppt)
The seconds engine order frequency responses on 18 blade tips. Solid lines—tuned tip responses (identical for all blades); dotted lines—mistuned tip responses. (a) 2% mistuning SD; (b) 5% mistuning SD; (c) 8% mistuning SD; and (d) 12% mistuning SD.
Grahic Jump Location
Fig. 3

The second engine order frequency responses on blade 2 tip by Monte Carlo sampling (5% mistuning SD). Gray lines—1000 sampled responses; black line—nominal responses; crosses—response peaks of individual runs.

+The second engine order frequency responses on blade 2 tip by Monte Carlo sampling (5% mistuning SD). Gray lines—1000 sampled responses; black line—nominal responses; crosses—response peaks of individual runs.
View Large  |  Save Figure  |  Download Slide (.ppt)
The second engine order frequency responses on blade 2 tip by Monte Carlo sampling (5% mistuning SD). Gray lines—1000 sampled responses; black line—nominal responses; crosses—response peaks of individual runs.
Grahic Jump Location
Fig. 4

Predictions of the second engine order frequency responses at the tip of blade 2. Plus—training point; cross—validation point; solid line—inferred responses; dashed line—true responses; shaded area—95% credible interval. Plots (a), (d), and (g) are obtained with same nine training data but different initial hyperparameters; plots (b), (e), and (h) are obtained with same seven training data but different initial hyperparameters; and plots (c), (f), and (i) are obtained with different posterior hyperparameters φ∧={1.0,0.2106392,0}, {0.3,0.1051911,0}, and {0.07,0.09349025,0}, respectively.

+Predictions of the second engine order frequency responses at the tip of blade 2. Plus—training point; cross—validation point; solid line—inferred responses; dashed line—true responses; shaded area—95% credible interval. Plots (a), (d), and (g) are obtained with same nine training data but different initial hyperparameters; plots (b), (e), and (h) are obtained with same seven training data but different initial hyperparameters; and plots (c), (f), and (i) are obtained with different posterior hyperparameters φ∧={1.0,0.2106392,0}, {0.3,0.1051911,0}, and {0.07,0.09349025,0}, respectively.
View Large  |  Save Figure  |  Download Slide (.ppt)
Predictions of the second engine order frequency responses at the tip of blade 2. Plus—training point; cross—validation point; solid line—inferred responses; dashed line—true responses; shaded area—95% credible interval. Plots (a), (d), and (g) are obtained with same nine training data but different initial hyperparameters; plots (b), (e), and (h) are obtained with same seven training data but different initial hyperparameters; and plots (c), (f), and (i) are obtained with different posterior hyperparameters φ∧={1.0,0.2106392,0}, {0.3,0.1051911,0}, and {0.07,0.09349025,0}, respectively.
Grahic Jump Location
Fig. 5

Statistical characterization of frequency responses on blade 2 tip: (a) Monte Carlo sampling with 1000 runs; (b) Monte Carlo sampling with 50 runs; and (c) inferred envelope compared with true envelope. Gray line in (a) and (b)—individual sampled response curve; dotted line in (a) and (c)—upper envelope of 1000 sampled response curves; dotted line in (b)—upper envelope of 50 sampled response curves; solid line in (a)—nominal response curve; solid line in (c)—inferred upper envelope; plus sign - feature point for inference; shaded area in (c)—95% credible interval.

+Statistical characterization of frequency responses on blade 2 tip: (a) Monte Carlo sampling with 1000 runs; (b) Monte Carlo sampling with 50 runs; and (c) inferred envelope compared with true envelope. Gray line in (a) and (b)—individual sampled response curve; dotted line in (a) and (c)—upper envelope of 1000 sampled response curves; dotted line in (b)—upper envelope of 50 sampled response curves; solid line in (a)—nominal response curve; solid line in (c)—inferred upper envelope; plus sign - feature point for inference; shaded area in (c)—95% credible interval.
View Large  |  Save Figure  |  Download Slide (.ppt)
Statistical characterization of frequency responses on blade 2 tip: (a) Monte Carlo sampling with 1000 runs; (b) Monte Carlo sampling with 50 runs; and (c) inferred envelope compared with true envelope. Gray line in (a) and (b)—individual sampled response curve; dotted line in (a) and (c)—upper envelope of 1000 sampled response curves; dotted line in (b)—upper envelope of 50 sampled response curves; solid line in (a)—nominal response curve; solid line in (c)—inferred upper envelope; plus sign - feature point for inference; shaded area in (c)—95% credible interval.
Grahic Jump Location
Fig. 6

Response peaks on blade 2 tip inferred with 21 training data (a) at all 1000 configuration points and (b) in a zoomed-in region. Cross—true peak value; plus—inferred peak value; circle—data point; shaded area—95% credible interval.

+Response peaks on blade 2 tip inferred with 21 training data (a) at all 1000 configuration points and (b) in a zoomed-in region. Cross—true peak value; plus—inferred peak value; circle—data point; shaded area—95% credible interval.
View Large  |  Save Figure  |  Download Slide (.ppt)
Response peaks on blade 2 tip inferred with 21 training data (a) at all 1000 configuration points and (b) in a zoomed-in region. Cross—true peak value; plus—inferred peak value; circle—data point; shaded area—95% credible interval.
Grahic Jump Location
Fig. 7

Response peaks on blade 2 tip inferred with 51 training data (a) at all 1000 configuration points and (b) in a zoomed-in region. Cross—true peak value; plus—inferred peak value; circle—data point; shaded area—95% credible interval.

+Response peaks on blade 2 tip inferred with 51 training data (a) at all 1000 configuration points and (b) in a zoomed-in region. Cross—true peak value; plus—inferred peak value; circle—data point; shaded area—95% credible interval.
View Large  |  Save Figure  |  Download Slide (.ppt)
Response peaks on blade 2 tip inferred with 51 training data (a) at all 1000 configuration points and (b) in a zoomed-in region. Cross—true peak value; plus—inferred peak value; circle—data point; shaded area—95% credible interval.
Grahic Jump Location
Fig. 8

Comparison of true and inferred response peaks on blade 2 tip (a) with 21 training data and (b) with 51 training data

+Comparison of true and inferred response peaks on blade 2 tip (a) with 21 training data and (b) with 51 training data
View Large  |  Save Figure  |  Download Slide (.ppt)
Comparison of true and inferred response peaks on blade 2 tip (a) with 21 training data and (b) with 51 training data

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
Introduction
Turbine Aerodynamics: Axial-Flow and Radial-Flow Turbine Design and Analysis> Chapter 1
PSA Level 2 — NPP Ringhals 2 (PSAM-0156)
Proceedings of the Eighth International Conference on Probabilistic Safety Assessment & Management (PSAM)
Evaluation of Terrorist Risk Using Belief and Plausibility (PSAM-0384)
Proceedings of the Eighth International Conference on Probabilistic Safety Assessment & Management (PSAM)
Performance-Based Expert Judgement Weighting Using Moment Methods (PSAM-0264)
Proceedings of the Eighth International Conference on Probabilistic Safety Assessment & Management (PSAM)
How to Build an Adequate Set of Minimal Cut Sets for PSA Importance Measures Calculation (PSAM-0071)
Proceedings of the Eighth International Conference on Probabilistic Safety Assessment & Management (PSAM)
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
This site uses cookies. By continuing to use our website, you are agreeing to our privacy policy. | Accept
Sign In