A Markov Chain Monte Carlo Based Method for System Identification

Conference · Tue Oct 22 04:00:00 EDT 2002

OSTI ID:15002274

Glaser, R E; Lee, C L; Nitao, J J; Hanley, W G

This paper describes a novel methodology for the identification of mechanical systems and structures from vibration response measurements. It combines prior information, observational data and predictive finite element models to produce configurations and system parameter values that are most consistent with the available data and model. Bayesian inference and a Metropolis simulation algorithm form the basis for this approach. The resulting process enables the estimation of distributions of both individual parameters and system-wide states. Attractive features of this approach include its ability to: (1) provide quantitative measures of the uncertainty of a generated estimate; (2) function effectively when exposed to degraded conditions including: noisy data, incomplete data sets and model misspecification; (3) allow alternative estimates to be produced and compared, and (4) incrementally update initial estimates and analysis as more data becomes available. A series of test cases based on a simple fixed-free cantilever beam is presented. These results demonstrate that the algorithm is able to identify the system, based on the stiffness matrix, given applied force and resultant nodal displacements. Moreover, it effectively identifies locations on the beam where damage (represented by a change in elastic modulus) was specified.

• Research Organization: Lawrence Livermore National Lab., CA (US)
• Sponsoring Organization: US Department of Energy (US)
• DOE Contract Number: W-7405-ENG-48
• OSTI ID: 15002274
• Report Number(s): UCRL-JC-150494
• Country of Publication: United States
• Language: English

Related Subjects

42 ENGINEERING
ALGORITHMS
FLEXIBILITY
MONTE CARLO METHOD
SIMULATION