TY - GEN
T1 - Development of a Surrogate Model for Structural Health Monitoring of a UAV Wing Spar
AU - Razzini, Adrielly H.
AU - Kressel, Iddo
AU - Ofir, Yoav
AU - Tur, Moshe
AU - Yehoshua, Tal
AU - Todd, Michael D.
N1 - Publisher Copyright:
© 2022, The Society for Experimental Mechanics, Inc.
PY - 2022
Y1 - 2022
N2 - A critical part to implementing a structural health monitoring system is being able to understand the structural response under different operational and environmental conditions. In this work, a detailed finite element model of an unmanned aerial vehicle’s wings’ spar was developed to serve as a synthetic data generator. A probabilistic understanding of the aerodynamic loads and debonding damages at different locations and with different sizes were implemented to simulate observations of the spar’s performance in service. The target measurements are uniaxial strain, measured in several paths throughout the spar. Given measured strain, the damage assessment problem is probabilistically formulated by defining local buckling from debonding as the observable damage, which is fundamentally characterized by load-dependent buckling eigenvalues. This FE physical model is highly computationally intensive, so a Gaussian process regressor and a multilayer artificial neural network (MANN) were designed to serve as a “run time” surrogate model to learn the relationships between inputs (loads and damage conditions) and outputs (strain measurements and buckling eigenvalues). The results illustrate that the surrogate models presented are a reliable replacement to the computationally expensive inverse finite element model in damage identification.
AB - A critical part to implementing a structural health monitoring system is being able to understand the structural response under different operational and environmental conditions. In this work, a detailed finite element model of an unmanned aerial vehicle’s wings’ spar was developed to serve as a synthetic data generator. A probabilistic understanding of the aerodynamic loads and debonding damages at different locations and with different sizes were implemented to simulate observations of the spar’s performance in service. The target measurements are uniaxial strain, measured in several paths throughout the spar. Given measured strain, the damage assessment problem is probabilistically formulated by defining local buckling from debonding as the observable damage, which is fundamentally characterized by load-dependent buckling eigenvalues. This FE physical model is highly computationally intensive, so a Gaussian process regressor and a multilayer artificial neural network (MANN) were designed to serve as a “run time” surrogate model to learn the relationships between inputs (loads and damage conditions) and outputs (strain measurements and buckling eigenvalues). The results illustrate that the surrogate models presented are a reliable replacement to the computationally expensive inverse finite element model in damage identification.
KW - Finite element
KW - Gaussian process regressors
KW - Neural networks
KW - Structural health monitoring
KW - Surrogate model
UR - http://www.scopus.com/inward/record.url?scp=85135064970&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-04122-8_12
DO - 10.1007/978-3-031-04122-8_12
M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???
AN - SCOPUS:85135064970
SN - 9783031041211
T3 - Conference Proceedings of the Society for Experimental Mechanics Series
SP - 99
EP - 102
BT - Data Science in Engineering, Volume 9 - Proceedings of the 40th IMAC, A Conference and Exposition on Structural Dynamics, 2022
A2 - Madarshahian, Ramin
A2 - Hemez, Francois
PB - Springer
T2 - 40th IMAC, A Conference and Exposition on Structural Dynamics, 2022
Y2 - 7 February 2022 through 10 February 2022
ER -