TY - GEN
T1 - Physics based multi-fidelity data fusion for efficient characterization of mode shape variation under uncertainties
AU - Zhou, K.
AU - Tang, J.
N1 - Funding Information:
This research is supported in part by National Science Foundation under grant CMMI – 1825324.
Funding Information:
This research is supported in part by National Science Foundation under grant CMMI ? 1825324.
Publisher Copyright:
Copyright © 2020 ASME.
PY - 2020
Y1 - 2020
N2 - Efficient prediction of mode shape variation under uncertainties is important for design and control. While Monte Carlo simulation (MCS) is straightforward, it is computationally expensive and not feasible for complex structures with high dimensionalities. To address this issue, in this study we develop a multi-fidelity data fusion approach with an enhanced Gaussian process (GP) architecture to evaluate mode shape variation. Since the process to acquire high-fidelity data from full-scale physical model usually is costly, we involve an order-reduced model to rapidly generate a relatively large amount of low-fidelity data. Combining these with a small amount of high-fidelity data altogether, we can establish a Gaussian process meta-model and use it for efficient model shape prediction. This enhanced meta-model allows one to capture the intrinsic correlation of model shape amplitudes at different locations by incorporating a multi-response strategy. Comprehensive case studies are performed for methodology validation.
AB - Efficient prediction of mode shape variation under uncertainties is important for design and control. While Monte Carlo simulation (MCS) is straightforward, it is computationally expensive and not feasible for complex structures with high dimensionalities. To address this issue, in this study we develop a multi-fidelity data fusion approach with an enhanced Gaussian process (GP) architecture to evaluate mode shape variation. Since the process to acquire high-fidelity data from full-scale physical model usually is costly, we involve an order-reduced model to rapidly generate a relatively large amount of low-fidelity data. Combining these with a small amount of high-fidelity data altogether, we can establish a Gaussian process meta-model and use it for efficient model shape prediction. This enhanced meta-model allows one to capture the intrinsic correlation of model shape amplitudes at different locations by incorporating a multi-response strategy. Comprehensive case studies are performed for methodology validation.
KW - Mode shape
KW - Multi-level Gaussian process
KW - Multi-response Gaussian process
KW - Order-reduction
KW - Uncertainty quantification
UR - http://www.scopus.com/inward/record.url?scp=85100923270&partnerID=8YFLogxK
U2 - 10.1115/DSCC2020-3199
DO - 10.1115/DSCC2020-3199
M3 - Conference article published in proceeding or book
AN - SCOPUS:85100923270
T3 - ASME 2020 Dynamic Systems and Control Conference, DSCC 2020
BT - Intelligent Transportation/Vehicles; Manufacturing; Mechatronics; Engine/After-Treatment Systems; Soft Actuators/Manipulators; Modeling/Validation; Motion/Vibration Control Applications; Multi-Agent/Networked Systems; Path Planning/Motion Control; Renewable/Smart Energy Systems; Security/Privacy of Cyber-Physical Systems; Sensors/Actuators; Tracking Control Systems; Unmanned Ground/Aerial Vehicles; Vehicle Dynamics, Estimation, Control; Vibration/Control Systems; Vibrations
PB - American Society of Mechanical Engineers
T2 - ASME 2020 Dynamic Systems and Control Conference, DSCC 2020
Y2 - 5 October 2020 through 7 October 2020
ER -