TY - JOUR
T1 - Data-driven reduced order model with temporal convolutional neural network
AU - Wu, Pin
AU - Sun, Junwu
AU - Chang, Xuting
AU - Zhang, Wenjie
AU - Arcucci, Rossella
AU - Guo, Yi-Ke
AU - Pain, Christopher C.
N1 - Funding Information:
The authors would like to thank the reviewers for their valuable comments and effort to improve the manuscript. This work is supported by State Key Laboratory of Aerodynamics, China Aerodynamics Research and Development Center under Grant SKLA20180303 , Natural Science Foundation of Shanghai under Grant 19ZR1417700 .
PY - 2020/3/1
Y1 - 2020/3/1
N2 - This paper presents a novel model reduction method based on proper orthogonal decomposition and temporal convolutional neural network. The method generates basis functions of the flow field by proper orthogonal decomposition, and the coefficients are taken as the low-dimensional features. Temporal convolutional neural network is used to construct the model for predicting low-dimensional features. In this work, the training data are obtained from high fidelity numerical simulation. Compared with recurrent networks, temporal convolutional neural network is more effective with fewer parameters. The model reduction method developed here depends only on the solution of flow field. The performance of the new reduced order model is evaluated using numerical case: flow past a cylinder. Experimental results illustrate that time cost is reduced by three orders of magnitude, and convolutional architecture is beneficial to construct reduced order model. The speed-up ratio is linear with the computational scale of the numerical simulation.
AB - This paper presents a novel model reduction method based on proper orthogonal decomposition and temporal convolutional neural network. The method generates basis functions of the flow field by proper orthogonal decomposition, and the coefficients are taken as the low-dimensional features. Temporal convolutional neural network is used to construct the model for predicting low-dimensional features. In this work, the training data are obtained from high fidelity numerical simulation. Compared with recurrent networks, temporal convolutional neural network is more effective with fewer parameters. The model reduction method developed here depends only on the solution of flow field. The performance of the new reduced order model is evaluated using numerical case: flow past a cylinder. Experimental results illustrate that time cost is reduced by three orders of magnitude, and convolutional architecture is beneficial to construct reduced order model. The speed-up ratio is linear with the computational scale of the numerical simulation.
KW - Deep learning
KW - Proper orthogonal decomposition
KW - Reduced order model
KW - Temporal convolutional network
UR - http://www.scopus.com/inward/record.url?scp=85076769876&partnerID=8YFLogxK
U2 - 10.1016/j.cma.2019.112766
DO - 10.1016/j.cma.2019.112766
M3 - Journal article
AN - SCOPUS:85076769876
SN - 0045-7825
VL - 360
JO - Computer Methods in Applied Mechanics and Engineering
JF - Computer Methods in Applied Mechanics and Engineering
M1 - 112766
ER -