We present an optimal sensor placement methodology for structural health monitoring (SHM) purposes, relying on a Bayesian experimental design approach. The unknown structural properties, e.g. the residual strength and stiffness, are inferred from data collected through a network of sensors, whose architecture, i.e., type and position may largely affect the accuracy of the monitoring system. In tackling this issue, an optimal network configuration is herein sought by maximizing the expected information gain between prior and posterior probability distributions of the parameters to be estimated. Since the objective function linked to the network topology cannot be analytically computed, a numerical approximation is provided by means of a Monte Carlo analysis, wherein each realization is obtained via finite element modeling. Since the computational burden linked to this procedure often grows infeasible, a Polynomial Chaos Expansion (PCE) approach is adopted for accelerating the computation of the forward problem. The analysis expands over joint samples covering both structural state and design variables, i.e., sensor locations. Via increase of the number of deployed sensors in the network, the optimization procedure soon turns computationally costly due to the curse of dimensionality. To this end, a stochastic optimization method is adopted for accelerating the convergence of the optimization process and thereby the damage detection capability of the SHM system. The proposed method is applied to thin flexible structures, and the resulting optimal sensor configuration is shown. The effects of the number of training samples, the polynomial degree of the approximation expansion and the optimization settings are also discussed.

An optimal sensor placement method for SHM based on Bayesian experimental design and polynomial chaos expansion

CAPELLARI, GIOVANNI;MARIANI, STEFANO
2016-01-01

Abstract

We present an optimal sensor placement methodology for structural health monitoring (SHM) purposes, relying on a Bayesian experimental design approach. The unknown structural properties, e.g. the residual strength and stiffness, are inferred from data collected through a network of sensors, whose architecture, i.e., type and position may largely affect the accuracy of the monitoring system. In tackling this issue, an optimal network configuration is herein sought by maximizing the expected information gain between prior and posterior probability distributions of the parameters to be estimated. Since the objective function linked to the network topology cannot be analytically computed, a numerical approximation is provided by means of a Monte Carlo analysis, wherein each realization is obtained via finite element modeling. Since the computational burden linked to this procedure often grows infeasible, a Polynomial Chaos Expansion (PCE) approach is adopted for accelerating the computation of the forward problem. The analysis expands over joint samples covering both structural state and design variables, i.e., sensor locations. Via increase of the number of deployed sensors in the network, the optimization procedure soon turns computationally costly due to the curse of dimensionality. To this end, a stochastic optimization method is adopted for accelerating the convergence of the optimization process and thereby the damage detection capability of the SHM system. The proposed method is applied to thin flexible structures, and the resulting optimal sensor configuration is shown. The effects of the number of training samples, the polynomial degree of the approximation expansion and the optimization settings are also discussed.
2016
ECCOMAS Congress 2016 PROCEEDINGS
978-618-82844-0-1
Structural Health Monitoring, Damage Detection, Optimal Sensor Placement, Bayesian Inference, Uncertainty Quantification.
File in questo prodotto:
File Dimensione Formato  
P6762_eccomas2016.pdf

accesso aperto

: Post-Print (DRAFT o Author’s Accepted Manuscript-AAM)
Dimensione 414.81 kB
Formato Adobe PDF
414.81 kB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11311/1006369
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 17
  • ???jsp.display-item.citation.isi??? ND
social impact